EPA-600/4-76-014
                                                   March 1976
            COLLABORATIVE  STUDY OF
PARTICULATE  EMISSIONS  MEASUREMENTS BY
       EPA METHODS 2,  3, AND  5 USING
    PAIRED PARTICULATE SAMPLING TRAINS
             (Municipal  Incinerators)
                           by
                       Henry F. Hamil
                      Richard E. Thomas

                   Southwest Research Institute
                    San Antonio, Texas 78284
                   EPA Contract No. 68-02-0626
                       Project Officer
                      M. Rodney Midgett
                    Quality Assurance Branch
             Environmental Monitoring and Support Laboratory
              Research Triangle Park, North Carolina 27711
             U.S. ENVIRONMENTAL PROTECTION AGENCY
             OFFICE OF RESEARCH AND DEVELOPMENT
        ENVIRONMENTAL MONITORING AND SUPPORT LABORATORY
              RE SEARCH TRIANGLE PARK, N.C. 27711
                               '".--'"-" kVrtorii Street, Eoom 1670
                               Cfliokao",'"!!.   60604

-------
                            DISCLAIMER

This report  has been  reviewed  by the  Environmental Research Center,
Research Triangle Park, Office of Research and Development, U.S. Environ-
mental Protection Agency, and approved for publication. Approval does not
signify that  the contents necessarily reflect  the views and policies of the
U.S. Environmental Protection Agency, nor  does mention  of trade names
or commercial products constitute endorsement or recommendation for use.

-------
                               SUMMARY AND CONCLUSIONS
      This report presents the results of statistical analyses of data from a collaborative test using paired
paniculate sampling trams.  The purposes of the test were to estimate the minimum variability that can be
expected with the use of BPA Method S and to determine any noticeable effect in the particulate concen-
trations due to spatial/temporal changes in the gas flow at this test site.  The paired train consists of two
mirror-image Method 5 trains in a single box, and allows two independent laboratories to obtain simultane-
ous particulate concentration data with probe nozzles only 5.8 cm apart. This report deals with Method 5,
and also Method 2 (Velocity) and Method 3 (Stack Gas Analysis) which are called for in the use of Me-
thod 5. In addition, the particulate concentrations are converted to the applicable compliance test standard
for the source tested, and these are also analyzed. The latest in-house revisions of the EPA methods were
used in  this test, and the results contained here  are applicable to these revisions.

      The collaborative test was conducted at a municipal incinerator in Dade County, Florida. Four paired
particulate sampling trains were used in the test, and eight concentration determinations were made in each
run.  Of these, 6 were made by laboratories operating on one side of a paired  train, while the other 2 were
made by a single laboratory operating the paired train by themselves.  There were 13 sampling runs made
over a three-week period, with 3 runs the first week and 5 each of the remaining two weeks.  The paired-
train laboratory was changed each week so that a total of 9 laboratories participated and a total of 104 con-
centration determinations were made. One of these determinations had to be deleted from the data set due
to a broken probe liner and contaminated filter but  was replaced with a substituted value for the statistical
analysis of the results.

      For each factor of interest in the report, precision components are estimated in terms of within-
laboratory, laboratory bias and between-laboratory  standard deviations and coefficients of variation.  These
precision estimates pertain to the determination of a single run result and not to the average of three results
that is specified in the performance test for compliance section of the Federal Register. The results are sum-
marized below.

     Method 5—The determined particulate concentration ranged from 81.7 to  254.5 mg/scm. The within-
laboratory term is estimated using the differences between determinations by the paired-train laboratory on
a given run.  The standard deviation estimated using all three laboratories' data is 13.81 mg/scm, or 10.4%
of the average determined concentration with 13 degrees of freedom.  The laboratory bias standard devia-
tion is estimated from an Analysis of Variance (ANOVA) of the data from the six single-tram laboratories.
The estimated value is 8.15  mg/scm, or 6.1% of the average determination with 3 degrees of freedom.  Com-
bining these two estimates gives a between-laboratory standard deviation of 16.04 mg/scm, or 12.1% of the
mean value.

     Method 2—The determined velocities ranged from 13.6 to 16.5 m/sec.  Only data from the six single-
train laboratories are used in this analysis. The  precision estimates are obtained using a coefficient of vari-
ation approach, where the standard deviations are expressed as a percentage of an unknown mean. The
within-laboratory standard deviation is estimated to be 2.2%  of the  mean value, with 66 degrees of freedom.
The between-laboratory standard deviation is estimated as 4.3% of the mean with 5 df. This gives an esti-
mated laboratory bias standard deviation of 3.7% of the mean.

     Method J—There were 7 gas analyses performed on all but the first 4 runs. The paired-train  labora-
tory took one sample for both trains, while one laboratory did not take  Orsats during the first four runs.
There were  87 determinations of percent carbon dioxide, percent oxygen and dry gas molecular weight, and
one substituted value was used to complete the  data set. Precision estimates were obtained using an ANOVA
approach and are summarized in terms of standard deviations.

     •     %COi —The determined concentrations ranged from 1.8 to 3.1 percent C02 . The within-labora-
           tory standard deviation is estimated to be 0.20 percent C02  with 67 degrees of freedom.  The
                                                  m

-------
           estimated laboratory bias standard deviation is 0.35 percent fO2, with 6 degrees of freedom.
           This results in an estimated between-laboratory standard deviation of 0.40 percent CO2.  In
           terms of relative variation, the between-laboratory  coefficient of variation is 16.7% of the mean
           determined value.

     •    %O-i  The determined oxygen concentration ranged from 16.5 to  19.5 percent O2. The within-
           laboratory standard deviation is estimated as 0.32 percent O2  with 67 degrees of freedom. The
           estimated laboratory bias standard deviation is 0.52 percent O2 with 6 degrees of freedom.
           These give a between-laboratory standard deviation estimate of 0.61  percent C)2.  No coefficient
           of variation is expressed for these data since the variation in the O2 is considered to be more re-
           lated to the variation in the CO2 determination than to the  actual  oxygen level in the stack gas.

     •    Dry Gas Molecular Weight -The dry gas molecular weights determined ranged from 28.06 to
           29.21 gm/gm-mole. The estimated within-laboratory standard deviation is 0.035 gm/gm-mole
           with 67 degrees of freedom.  The laboratory bias standard deviation  estimate is 0.033 gm/gm-
           mole with 6 degrees of freedom.  This gives an estimated between-laboratory standard deviation
           of 0.048 gm/gm-mole. The variation in this determination is related to the variation in the CO2
           and O2 determinations rather than the true stack gas molecular weight.

     Moisture Fraction-The determined moisture fraction ranged from 0.117 to 0.175. Using the data from
all 8 trains, there are 104 separate  determinations of the proportion of water vapor in the gas stream. These
are used in an ANOVA approach to obtain  the following estimates. The within-laboratory standard deviation
is estimated to be 0.009 with 70 degrees of freedom.  The laboratory bias standard deviation estimate is
0.008 with 7 degrees of freedom.  This gives a between-laboratory standard deviation estimate of 0.012.

     Paniculate Concentration Corrected to 12% C02 -In order to show compliance with Federal regulations
concerning incinerators, particulate concentrations are to be converted to a common base of 12 percent C02
by multiplying the concentration obtained  by the factor 12/% CO2, where the percent CO2 is obtained from
the stack gas analysis. The concentrations determined in the collaborative test were corrected to 12% C02 us-
ing the Orsat data and reanalyzed.  The corrected particulate concentrations ranged from 377.1 to 1513.7
mg/scm.  From  the differences between the paired-train laboratories' concentrations, a within-laboratory
standard deviation of 55.43 mg/scm is calculated. This represents an 8.2% variation relative to the mean level.
This does not include any added effect due to determination of percent CO2 in the gas stream, however,
since these laboratories made a single stack gas analysis for both trains.  A separate error term gives an esti-
mated within-laboratory standard  deviation of 96.52 mg/scm, or  14.3% of the mean, with variability due to
CO2  determination included.  There are 13 and 69 degrees of freedom,  respectively, for the two estimates.
The laboratory bias standard deviation is estimated from differences between the independent laboratories
operating on a single-paired train.  The estimated value is  1 14.96 mg/scm with 3  degrees of freedom, or 17.0^
of the mean value. The between-laboratory standard deviation is estimated separately using each of the within-
laboratory estimates above. From the paired-train laboratory estimate,  the between-laboratory standard devi-
ation is 127.62 mg/scm, or an 18.8% variation.  Using  the ordinary ANOVA error term, the between-laboratory
standard deviation is 150.10 mg/scm, giving an estimated coefficient of variation of 22.2%.

     The apparent conclusion from the above is that qualified teams using the revised Method 5 and care-
fully following the procedural details, specified to a much greater degree than  in the  published method, can
obtain particulate concentration measurements with reasonable precision. The precision for the associated
methods appears adequate  for the  purposes for which  they were intended, with the possible exception of the
CO2 determination.  It can be seen from the corrected particulate loadings that significant variation was in-
duced in the loadings due to variation in the CO2 data.

     Comparison is made in this report between the results of this study  and those of previous studies deal-
ing with the above variables.
                                                  IV

-------
                                    TABLE OF CONTENTS

                                                                                              Page

LIST OF ILLUSTRATIONS	"'

LIST OF TABLES	w

I.    INTRODUCTION	    1

II.   COLLABORATIVE TESTING OF METHODS AND ASSOCIATED METHODS	    2

     A.   Collaborative Test Site	    2
     B.   Collaborators	    2
     C.   Pretest Calibration  Requirements	    7
     D.   Conduct of the Test	    7
     E.   Philosophy of Collaborative Testing	15
     F.   Acknowledgements	15

III.  STATISTICAL ANALYSIS	16

     A.   Definitions and Terms	16
     B.   Particulate Test Data	17
     C.   Precision of Method 5	18
     D.   Rank Test for Equality of  Laboratories	21
     E.   Velocity	22
     F.   Stack Gas Analysis	24
     G.   Moisture Fraction	29
     H.   Precision of Particulate  Loadings Corrected to 12% CO2	31

IV.  COMPARISON WITH PREVIOUS STUDIES	35

APPENDIX A-REVISED EPA METHODS FOR PARTICULATE EMISSION MEASUREMENT	37

     A.1  Method 2—Determination  of Stack Gas Velocity and Volumetric Flow Rate (Type S Pitot Tube)   .   39
     A.2  Method 3—Gas  Analysis for Carbon Dioxide, Oxygen, Excess Air, and Dry Molecular Weight   .   .   48
     A.3  Method 5—Determination  of Particulate Emissions from Stationary Sources	54

APPENDIX B-STATISTICAL  METHODS	71

     B.1   Statistical Model for Particulate  Concentration Data	73
     B.2  Precision Estimation and Tests of Hypotheses for Method 5	75
     B.3  Rank Test for Equality of  Laboratories	77
     B.4  Weighted Coefficient of Variation Estimates	77
     B.5  Precision Estimation for Velocity Determination	80
     B.6  Precision Estimation for Method 3	82
     B.7   Precision Estimation for Moisture Fraction Determination	87
     B.8   Precision Estimation and Tests of Hypotheses for Particulate Concentration Corrected to 12% C02-   88

LIST OF REFERENCES	91

-------
                                   LIST OF ILLUSTRATIONS




Figure                                                                                           Page




   I       Sampling Site Configuration	    3




   2       Sampling Platform Configuration	    4




   3       Sampling Train Operation	    4




   4       Average Velocity Profiles	    5




   5       Metro Dade County Northeast Incinerator	    6




   6       Meter Box Calibration Form	    8




   7       Pilot Tube Calibration Form	    9




   8       Nomograph Calibration Form	10




   9       Paired Particulate Sampling Train	11




  10      Dual Sample Box—Dimensions and Glassware Layout    	12




  11       Dual Sample Box    	13







                                        LIST OF TABLES




Table                                                                                             Page




   1       Particulate Concentration, mg/scm	17




   2       Percentage of Collected Mass in Acetone Wash	18




   3       Estimated Precision Between Paired-Train Laboratories	18




   4       Estimated Within-Laboratory Components from Interaction Term	19




   5       Ranks of Concentration Determinations—Individual Weeks	22




   6       Ranks of Concentration Determinations—Three Weeks Data	22




   7       Determined Velocity of the Gas Stream, Arranged by Block (m/sec)	22




   8       Blocking Criterion for Velocity	23




   9       Percent CO2 Determinations	25




   10      Percent 02 Determinations	27




   11      Dry Gas Molecular Weight Determinations (gm/gm-mole)	28
                                                 VI

-------
                                  LIST OF TABLES (Cont'dj





Table




  12      Determined Moisture Fractions	30




  13      Particulate Concentration Corrected to 12% CO2 , mg/scm	32




  14      Precision Estimation Between Paired-Train Laboratories	32




  15      Comparison of Coefficients of Variation for Method 5	35




  Bl      Contrasts Among Laboratory  Means	74




  B2      ANOVA for Method 5 Concentration Determinations, Individual Weeks	75




  B3      ANOVA for Method 5 Concentration Determinations, 3-Week Analysis	76




  B4      Summary of Friedman Test	77




  B5      Run Data Summary	81




  B6      Collaborator-Block Data Summary	82




  B7      ANOVA for CO2 Determination	83




  B8      ANOVA for O2 Determination	85




  B9      ANOVA for Md Determination	86




 BIO     ANOVA For Bws Determination	88




 Bll      ANOVA for Paniculate Concentration Corrected to 12% CO2,  Individual Weeks	89




 B12     ANOVA For Paniculate Concentration Corrected to 12% CO2 , Three-Week Analysis	90

-------
                                     I. INTRODUCTION
     This report describes the work performed and results obtained on Southwest Research Institute Project
01-3462-014, Contract No. 68-02-0626, which includes collaborative testing of the most recent revisions of
Method 5 and related methods for particulate emissions in accordance with the test methods and procedures
specified for municipal incinerators given in "Standards of Performance for New Stationary Sources."^ '

     This report describes the collaborative testing of the most recent revisions of Method 5 and associated
methods in a municipal incinerator using paired particulate sampling trains. The statistical analysis of the data
from the collaborative test and the conclusions and recommendations based upon this analysis of data are
presented.

-------
                      II.  COLLABORATIVE TESTING OF METHOD 5
                               AND ASSOCIATED METHODS
A.   Collaborative Test Site

     Arrangements were made for collaborative testing of Method 5 and associated methods at the North-
east Municipal Incinerator of the Metro Dade County Public Works Department, Miami, Florida. Arrange-
ments for use of this site were made by Dr. William Mitchell, EPA.

     The incinerator consists of a single furnace train with a capacity of 272 metric tons of refuse per
24-hour day.  The furnace is a dual-grate design and is fed refuse by a hydraulic ram on an automatic time
cycle. Refuse is transferred from the storage bins to the ram-charging hopper by a traveling bridge crane.

     The unit is operated to maintain a furnace outlet temperature of 980°C.  Gases leaving the furnace are
cooled by water injection through an array of spray heads and by addition of dilution air.  Water feed rates
and dilution air volume are controlled by regulators which use furnace outlet temperature  as a control
signal. The gases are cooled  to about 260°C and are then fed to an electrostatic precipitator for particulate
removal. The outlet gas from the precipitator passes through an induced draft fan and is fed into the base
of a 3-meter diameter stack.  Due to the gas conditioning prior  to the precipitator, the  stack conditions were
relatively uniform; stack gas  velocity was about 15 meters/second, stack gas temperature was approximately
250°C, stack gas moisture content ranged around  15 percent and CO2 concentration in the stack gas was
2.0 2.5 percent.

     Four  sampling ports were available. The sample ports were at 90 degrees to each other and were
located 23 meters above grade.  Access to the sample ports was from a square platform around the stack
which was 21.6 meters above grade. The sample ports were located 19.5 meters (6.5 diameters) downstream
of the stack inlet and greater than 6 meters (2 diameters) upstream of the stack outlet. This i equired the use
of 24 traverse points, 12 on each diameter. The sample site configuration is shown in Figure  1.  Figure 2 is
a view of the sampling platform taken from the ground, with the 4 paired trains in place at the corners of
the platform.  Figure  3 shows a crew (technicians  from two independent laboratories) operating a train
during a sampling  run.  Average velocity profiles are shown in Figure 4, and a view of the unit is shown in
Figure 5.

B.   Collaborators

     The collaborators for the Northeast Incinerator test were:

              Name                                                 Organization

     Mr. Mike Taylor                                       Southwest Research  Institute
     Mr. Nolhe Swynnerton                                 San Antonio, Texas
     Mr. Hector Ramos

     Mr. Emil Stewart                                      Entropy Environmentalists, Inc.
     Mr. David Huckabee                                   Research Triangle Park,  N. C.
     Mr. Roy Doster

     Mr. Bill DeWees                                       Pedco-Environmental,
     Mr. John Atkins                                       Cincinnati, Ohio

     Mr. Kim Thompson                                     Commonwealth Laboratory
     Mr. Pete Watson                                       Richmond, Virginia

-------
                          O
o
  1 17m
                1.36m
                                 Side View
5.48m
                                     3.05m
            1.17m—
                                   Ports 0 15m
                                   diameter
                                                         \
                                 Top View

                 FIGURE 1. SAMPLING SITE CONFIGURATION

-------
FIGURE 2. SAMPLING PLATFORM CONFIGURATION

    FIGURE 3. SAMPLING TRAIN OPERATION

-------
            Stack Diameter, Meters
           Profile Across Ports 1-3
            Stack Diameter, Meters
           Profile Across Ports 2-4
FIGURE 4. AVERAGE VELOCITY PROFILES

-------
                    FIGURE 5. METRO DADE COUNTY NORTHEAST INCINERATOR
          Name                                                         Organization

     Mr. Barry Jackson                                     Roy F. Weston, Inc.
     Mr. Jeff O'Neil                                        Westchester, Penn.

     Mr. Al Linero                                         Environmental Science and Engineering, Inc.
     Mr. Gary McRae                                       Gainesville, Florida
     Mr. Bill Wells

     Mr. Joe Wilson                                        Scott Environmental Technology, Inc.
     Mr. Walt Nunes                                       Plumsteadville, Penn.
     Mr. Dave Ochadlik

     Mr. John Dollar                                       Harmon Engineering
     Mr. Wayne Daughtry                                   Auburn, Alabama
     Mr. Jim  Menchey

     Mr. Bob Norton                                       Pacific Environmental Services, Inc.
     Mr. Joe Boyd                                         Santa Monica, California
     Mr. Bob Missen

     Note: Throughout the remainder of this report, the collaborating laboratories are referenced by
     randomly assigned code numbers as Lab  101 through Lab 109.  These code numbers do not
     necessarily correspond to the above ordered listing of collaborators.

The collaborative test was conducted under the general supervision of Dr. Henry F. Hamil of Southwest Re-
search Institute. Dr. Hamil had the overall responsibility for assuring that the test was conducted in accordance
with the collaborative test plan and that the collaborators adhered to the most recent revision of Method 5.

     Dr. William Mitchell, Quality Assurance Branch, EMSL, EPA, was present during the first week of the
test. Mr. Rodney Midgett, Project Officer, Quality Assurance Branch, EPA, and Mr. Ed McCarley, Emission
Measurement  Branch, OAQPS, EPA, were present during the second week of the test.

-------
      Southwest Research Institute was provided by EPA with a list of sixteen laboratories who were current
 or past EPA contractors. These laboratories were considered to have the requisite experience and expertise
 to participate in a Method 5 collaborative test. Requests for quotation along with a scope of work statement
 were submitted to all sixteen laboratories.  Fourteen responses were received, and the eight low bidders were
 accepted for the collaborative test.

 C.    Pretest Calibration Requirements

      Improper calibration of test equipment or changes in calibration factors after laboratory calibration
 and prior to the test can be sources of imprecision in Method 5 determinations.  In order to minimize any
 effect on the test results due to equipment calibration, the collaborators were required to provide Southwest
 Research Institute with their calibration data prior to the start of the test. Meter box calibrations and pilot
 tube calibrations were carried out as specified in the revised methods and in accordance with  the procedures
 specified by Rom.(10) Calibration data for meter boxes and pitot tubes were reported on the forms shown in
 Figure 6 and Figure 7, respectively.

      Prior experience had shown that commercially available nomographs were not always reliable, due to
 misalignment of the various scales on the nomographs. Calibration checks of the nomographs used by the
 collaborators were required, using the procedure of Shigehara.    ^ The nomograph calibration reporting form
 is shown in Figure 8. Each collaborator was required to have a spare nomograph to avoid any delays in the
 test due to damage to a nomograph.

      As a further check on meter box calibrations, and to determine if any calibration changes had occurred
 during shipment, a two-point calibration check of all meter boxes was conducted at the test site prior to the
 start of the test. Southwest Research Institute provided a calibrated dry gas meter to be used as a standard
 during this calibration check. This dry gas meter was calibrated against a spirometer just prior to the test.

      The two calibration  check points were selected  to span the range of Ap expected during the test.  Three
 meter boxes were found to be outside the allowable specifications for dry gas meter calibration  according to
 the revised method (7 =  1.00 ± 0.02).  One meter box was corrected by replacement of the dry gas meter with
 a spare meter. The other two meter boxes were recalibrated in the field, using the Southwest Research
 Institute "standard" dry gas meter. After this recalibration, one meter box  was found to have an orifice
 coefficient (A//) outside the allowable range. The orifice was replaced with  a spare which met the calibration
 specifications.

     Calibration requirements for the pitot tubes specified that the calibration be performed with the com-
 plete pitobe assembly using a 3/8-inch nozzle. All laboratories except one had pitot coefficients in the
 acceptable range (0.85 ± 0.02). The one laboratory whose pitot tube coefficient was outside the allowable
 range used the adjustment  equation described by Shigehara^-^to obtain a proper nomograph  C factor for
 their pitot tube coefficient.

     All collaborators had the required number of acceptable nomographs, checked by the procedure of
 Shigehara(13).

 D.   Conduct of the Test

     The purpose of this collaborative test was to estimate the minimum variability that can be expected
 from Method 5 (revised) when used in accordance with the applicable test methods and procedures specified
 for  the source tested. Additionally, the test was structured  to allow determination of any noticeable effect
in the results of a field test due to spatial/temporal changes  in particulate concentration due to changing
 stack gas flow patterns.

     Previous collaborative tests of Method 5 (3>4>5) involved four collaborating laboratories using con-
ventional Method 5 trains.   Each team sampled all ports sequentially, using either radius or diameter

-------
Calibration Check:  Orifice Meter and Dry Gas Meter




Laboratory	
                              Meter Box
Date
Barometric Pressure, Pf, =.
                              Primary Standard
in Hg
Orifice
Manometer
Setting, AH
inH20
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
2.0
3.0
4.0
Gas Volume
Primary Standard
Vy,
(ft3)












Gas Volume
Dry Gas Meter
vd
(ft3)












Temperatures
Primary
Standard, tw
(°F)












Dry Gas Meter
Inlet
'«/«
(°F)












Outlet
*3°
(°F)












Average
td
(°F)












Time
e
(min)












Average
7













A//@













Calculations
A//
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
2.0
3.0
4.0
7
VwPb(td + 460)
/ Atf \
Vd(Pb+—~)(tw+460)
^ 13. 6/












AW@
0.0317 Af/ frtw + 460)01 2
Pb(td + 460) L Vw J












                                 FIGURE 6. METER BOX CALIBRATION FORM

-------
Calibration Cl e ck:  S Tyi  • Pitot Tube




Reference:  R  vised Meth >d 2 Draft
         Labo' itory
         Pitol  'ube Identi '.cation Number:




         Date:
         Calib  ited by:

1
1
j Run No.
i
' 1
; z
3
i

'A" Tide Calibration
^Pstd
•m H20
n. H2C))




AP(s)
cm H2O
(in. H2O)



Average
CP(S)





Dev.





Run No.
1
2
3
(
"B" Ude Calibration
'^Pstd
m H2Q
i. H,O)
i-.




AP(S)
cm H2O
(in. H20)



Average
CP(S)





Dev.




Dev.   =  C
p(S)   Cp(S)(avg
C_ Difference:  A    - B
  P                avg    a
                                                 (Must be <0. 01)
                                       (Must be < 0.01)
                      FIGURE 7. PITOT TUBE CALIBRATION FORM

-------
Calibration Check:  Nomograph






I ,ihor;iloiv	



D.IIO	.



(I)   (' l-'actor Adjustment



          Probe 1
               Probe 2
                                                                                  Probe 3
I IT,, >OS7orCp <0.83
                                         Qdj =
                  (0.85)2
                                             adj
                                                                                     r2
                                                                                      p
                                                                                   (0.85)2
               (0.85)2








(2}   Accuracy



     (a)   ^-factor line. A//and Ap scales



       Aline                      Set
                                             = C-
                                                  (0.85)2
                                                     (0.85)2
                                             = C(
                                                = C(
                                                                    A// Reading
A/) = 0.001; A// = 0.1
A/>= iO.O; A//= 10.0
A/; = 0 1 ; A// = 1 .0
     (b)   C, /4 and/),, scales
           Dn
   = 0.01
Ap = 1 .0
Ap = 1 .0

Ap = 0.01
                                                      Nomograph 1
                                                 Nomograph 2
                                                                    A// Readings
   Ap
                                                      Nomograph 1
Nomograph 2
20
1 S
1.0
0.7
0.5
0.5
0.4
0.3
0.25
0.2
2500
1500
1000
500
200
0.02
0.8
0.1
3.0
0.9
                          FIGURE 8. NOMOGRAPH CALIBRATION FORM
                                            10

-------
traverses as required by the sampling site configuration. Each team sampled all the sample points, but at
different times. Therefore, any variation in results due to spatial/temporal changes in the particulate con-
centration would be incorporated into the  precision estimates for the method. Also, since the true particu-
late concentration in these tests was not known, it could not be assumed that day-to-day samples obtained
by the laboratories were replicates.  It was  necessary to block the test data before statistical analysis, using
plant operating data as the blocking criterion. However, the plant operating data were only indicators of
changing particulate concentrations and gave no true measure of the concentration itself.

      The present test involved a total of nine collaborating laboratories using paired particulate sampling
trains. The paired particulate sampling trains consist of two mirror-image Method 5 sample boxes on a
common base.  Each half of the train consists of a probe support, heated filter oven, impinger train, icebath
and sample head.  Aside from the base, there is no commonality in the  trains. The pitobe assemblies are
mounted to give a separation of 5.8 cm between the sample nozzles. Probes are mounted so that the pitot
tubes are outboard relative  to each other.  Figure 9 shows a paired particulate sampling train in position for
a sampling run with probes in place and umbilicals connected. The dimensions and glassware layout of the
dual sample box are presented in Figure 10, while overall views are presented in Figure  11.
            tt
                         FIGURE 9. PAIRED PARTICULATE SAMPLING TRAIN
      The individual filter ovens are fitted with circulating fans.  Oven heat is provided by electrical resistance
heaters with solid-state variable power supplies.  Oven temperatures are manually controlled. Oven temperature
and impinger outlet temperature are obtained using bimetallic dial-indicating thermometers.  Probe heat is
controlled using variable transformers. Probe temperature versus transformer settings were calibrated by each
laboratory according to Rom.^0) Stack gas temperature  is  determined by thermocouple readout. Um-
bilicals containing a vacuum line, pitot lines,  and an Orsat sample line were fabricated. All electrical con-
nections to the dual sample boxes are separate from the umbilical.

      Southwest Research Institute provided the paired particulate sampling trains along with special  glass-
ware (filter holders and filter holder-to-impinger train connectors) and umbilicals. Collaborating laboratories
provided their own probes,  pitot tubes, meter boxes, impinger trains, filters, silica gel and other normal
sampling supplies and equipment.  All laboratories except one provided their own gas sampling equipment
to obtain integrated gas samples for Orsat analysis.
                                                11

-------
                                                                            54.6
                                    Top View
                                    Side View
Note:  Dimensions in centimeters.
         FIGURE 10. DUAL SAMPLE BOX-DIMENSIONS AND GLASSWARE LAYOUT
                                       12

-------
O
             r
    !
~i—h
                                                                                                                X
                                                                                                                o
                                                                                                                fO
                                                                                                                u,
                                                                                                                _!
                                                                                                                0-
                                                                                                                S
                                                                                                                
-------
      The test plan called for a 1-day orientation session and equipment calibration check, and a 3-week test.
A total of 15 sampling runs were planned, five runs per week. Sampling was restricted to Monday through
Friday since the incinerator did not operate on weekends.  Four paired particulate sampling trains were used,
with seven laboratories each week. One laboratory operated one paired train, with a single operator running
both meter boxes.  Three independent laboratories, one per week of sampling, were used on this train. The
remaining six laboratories were randomly paired, and each of these laboratories operated one side  of a paired
particulate sampling train. The pairings of these six laboratories were unchanged for the entire three weeks
of the test. This sampling scheme provided for eight separate determinations of particulate concentration on
each run.

      This test plan, utilizing paired  particulate sampling trains, circumvents some of the drawbacks en-
countered in previous Method 5 tests.  Four pairs of samples are obtained on each run, one sample pair taken
by a single laboratory, and three sample pairs taken by three pairs of laboratories.  Each of these pairs of
samples should be free of any effect  of spatial/temporal variation on particulate concentration during the
run (8,9)  Therefore, the paired-laboratory samples allow an estimation of laboratory effect which is not in-
fluenced by  any particulate variation during the run, while the  single-laboratory sample pair allows estimation
of sampling error by using this sample pair as a replicate measurement. Additionally, the use of paired
trains allows concurrent sampling by up to eight laboratories. Previous collaborative tests were restricted
to use of only four laboratories sampling concurrently due to the configuration of the sampling site and
the use of standard Method 5 sampling trains. A detailed discussion of the statistical model and analytical
design for this test is contained in Appendix B.I.

      As stated above, the test plan called for fifteen sampling  runs, five per week for three weeks. Thirteen
runs were made, three the first week, and  five each the second  and third weeks. One run was lost  due to
severe weather conditions, and the other was lost  due to a malfunction in the electrostatic precipitator which
forced a unit shutdown.

      Some problems with equipment were encountered. One  laboratory used a meter box equipped with
Magnahelix™ differential pressure gauges. On the initial velocity traverse, it was observed that a  high Ap
(0.55 in H2O) was obtained on the velocity head gauge compared to Ap valves of around 0.40 inch H20
for the other laboratories. Substitution of an inclined manometer for the Magnahelix™ gauge gave Ap
readings comparable to those obtained by the other laboratories. Another laboratory obtained low Ap
readings (0.3 in. H20). The pilot lines were checked for obstructions or leaks but none were found.
Further investigation led to discovery of a leak in the low-pressure line between the pitot line  quick-con-
nects and the inclined manometer in the meter box. Other problems arose due to lack of familiarity of some
laboratories with the revised version of Method 5  being tested.  Incorrect assembly of the pitobe was noted
in several  cases; errors in the relative positions of the sample nozzle, pitot tube, Orsat sample tube and stack
temperature thermocouple  specified in the revised method to avoid sample nozzle and pitot tube inter-
ferences were observed and corrected.  Use of improper equipment and techniques (plastic wash bottles,
plastic probe wash containers, failure to properly  cap both ends of the probe prior to moving the probe
from the sample site to the  clean up area) were observed and called to the attention of the collaborators
for correction.

      Two operational problems were encountered. The new leak check requirements in the revised method
led to a number of delays in sampling runs. The collaborators chose to leak check through the  sample probes
to avoid removing the probe after each traverse. Numerous leaks were discovered.  Most of these leaks
developed in the front packing gland where the probe liner to sample nozzle seal is effected. Several types
of packing were used by the collaborators, with Teflon ferrules being the most satisfactory. However,
several probe liners were broken in the initial pre-run leak check due to collaborators overtightening the
packing gland in attempts to stop small leaks. No probe liners were broken during the course of a  run.
However, one probe liner was apparently broken after completion of the sampling run and prior to sample
recovery.  A second problem was encountered with the leak check procedure required in the integrated gas
sampling procedure in revised Method 3, which calls for checking the system at 250 mm Hg vacuum. A
                                               14

-------
variety of diaphragm pumps were used by the collaborators, and the vacuum which could be pulled varied
from about 100 mm to about 350 mm Hg. As a result, some leak checks were conducted at less than
250 mm Hg.

      During the test, three laboratories found it necessary to change personnel in their sampling crews. These
changes were made in two cases by having the replacement arrive at the site one day before the departing person
left so that a one-day orientation was available before the replacement  started work on his own. In the
estimation of the test supervisor, the performance of the crews with replacement was as good  as with the
original crews, and in one case, a significant improvement in team performance was observed.

E.    Philosophy of Collaborative Testing

      The concept of collaborative testing followed in the previous tests'-  •* involved conducting the test
in such a manner as  to simulate "real world" testing as closely as possible. "Real world" testing implies
that the results obtained during the test by each collaborator would be the same results obtainable if he were
sampling alone, without outside supervision, and without any additional information from outside sources,
i.e., test supervisor or other collaborators.

      The function of the test supervisor in such a testing scheme is primarily to see that the method is
adhered to as written and that no individual innovations are incorporated into the method by  any collaborator.
During the test program, the  test supervisor observed the collaborators during sampling and sample recovery.
Any deviations from the method as written were pointed out to the collaborator by the test supervisor for
correction. However, if random experimental errors in sampling and sample recovery were observed, no inter-
ference was made by the test supervisor.

      The present test was designed to estimate the minimum error in the revised method. As a result, the
test supervisor played a more active role, in order to minimize random  errors in sampling and sample recovery,
and to assure that the collaborators met the constraints of revised Method 5  and the associated Methods 2
and 3.

      The increased  level of supervision, the increased emphasis on collaborator qualifications, more strin-
gent quality control of pretest equipment calibration and field checks of selected calibration factors were
all implemented to allow estimation of the minimum variability of the  revised Method 5.

F.    Acknowledgements

      Southwest Research Institute wishes to acknowledge the assistance of Dr. William Mitchell, Quality
Assurance Branch, EPA, who made the arrangements for the use of the test site.  Dr. Mitchell also provided
assistance  to the test supervisor during the first week of the test.

      Appreciation is given to the Metro Dade County Public Works Department, Miami, Florida, for use of
their Northeast Municipal Incinerator as a test site. This appreciation is extended especially to Mr. Bud Hill
and his staff at the incinerator for their assistance and cooperation both during site preparation and the con-
duct of the test.
                                                 15

-------
                                Ill, STATISTICAL ANALYSIS
A.   Definitions and Terms

     To facilitate the understanding of this report and the utilization of its findings, this section explains
the statistical terms used in the report. The estimates of the pertinent values and the details of the methods
used to obtain them are contained in subsequent sections.

     We say that an estimator, 6, is unbiased for a parameter, 6, if the expected value of 0 is 0, or in no-
tational form, E(d) = 0.  Let X\, X2, • • ., Xn be a sample of size n from a population of method deter-
minations.  Then we define:
                  n
     (1)   X = -  2-t X{ —the sample mean, an unbiased estimate of the true mean determination. The
              «,-=,

           sample mean estimates the center of the distribution of X.
                     n
            2 _   '   y
     (2)   s =     - Z~, (Xj -• X)2  —the sample variance, an unbiased estimate of the population variance,
               "    / = 1

           a2.  The sample variance estimates the dispersion in the distribution of X.

     (3)   x = \/s2  —the sample standard deviation, a biased^  ^ estimate of a, the population standard
           deviation. The sample standard deviation is an alternative estimate of dispersion.  To remove
           the bias  in s,  a factor of <*„, dependent on the sample size, is  used, so that E (ot.ns) = a.

              a
     (4)   /3 = —  -the  true coefficient of variation.  The sample coefficient of variation, j3, is estimated by

           the ratio of the two unbiased estimates, |3 = —=-.
                                                   j\

     The precision components are estimated for this report by one of two methods. For the majority of
the data, an Analysis of Variance (ANOVA) procedure is used. This is a test for the equality of means of
several groups utilizing the estimated variance components for the various factors in a model.  The ANOVA
is generally summed up in a table which contains the following information.

     (5)   df— degrees of freedom, an indication of the degree of confidence in the estimate. The more
           df, the higher the degree of confidence one has in an estimated variance component.

     (6)   SS—sum of squares. The SS represents the squared deviations from the mean of a particular
           group.

     (7)   MS = SS/df-mean square. The MS is a variance estimate, similar to the s2 estimate above.
           The MS  of a  factor is used both to estimate its variance component and to test for significance.

     (8)   F—F-ratio. The ratio of a mean square to its appropriate denominator.  A large F-ratio implies
           a difference among means for a term in the ANOVA, and the existence of a variance component
           for that  factor.

     (9)   EMS—expected mean square. The expected value of MS, which allows the  proper F-ratio to be
           formed and variance components to be estimated.
                                                16

-------
      The second technique is referred to as a coefficient of variation approach. In this, the model used is
one in which the standard deviation is assumed to be proportional to the mean value. The  coefficients of
variation, then, are estimated for the components of interest and standard deviations expressed as a
constant times an unknown mean. The estimates take the form of (4) above and are weighted, if necessary,
and averaged into a single value.

      The precision components for this report will  be expressed in terms of standard deviations. The
principal components will be:

      Of, -between-laboratory standard deviation. This represents the total variation in a result, composed
of within-laboratory and laboratory bias components.  The between-laboratory variance can be written as
      a-within-laboratory standard deviation. This represents the ordinary sampling error in replicates
made at the same true level by the same laboratory team.

      OL -laboratory bias standard deviation. This represents the variation that can be expected between
two independent laboratory teams determining results at the same true level, ^. This variation is attributable
to such factors as different operators, equipment and analysts.

B.    Particulate Test Data

      For the statistical analysis, the paired-train laboratories are designated Labs 101,  102 and 103.  Labs
104 and 105 operated on the same paired train as did Labs 106 and 107 and Labs 108 and 109. There were
13 sampling runs made in all, and 104 determined particulate concentrations. These were spot-checked for
their calculation accuracy and adjusted where necessary. The  only values that required  recalculation were
those from laboratories that use a standard computer program to obtain the results of the test.  Since the
revised method specifies that standard temperature be 20°C (68°F), the constants for calculating both
standard metered volume and standard water volume changed slightly.  In these cases, the volumes and par-
ticulate loadings required recalculation. The checked and recalculated data are shown in Table  1 .

                          TABLE 1.  PARTICULATE CONCENTRATION, mg/scm
                                                Labs  •
Run

1
2
3
4
5
6
7
8
9
10
11
12
13
101
A
117.9
100.8
128.5










B
132.6
106.1
135.7










102
A



144.1
103.4
150.9
81.9
101.3





B



148.1
112.3
133.3
81.7
104.2





103
A








101.2
157.3*
169.9
254.5
188.6
B








107.6
158.1
161.3
190.2
189.9
1 r\A
lU*t
189.5
129.9
143.9
155.8
120.3
149.3
101.2
137.2
97.4
158.1
162.8
164.1
185.1
i n<\
i \jj
122.3
79.4
135.0
151.0
124.7
123.3
129.4
114.1
108.1
144.2
124.4
146.7
157.4
1 A£
1UO
139.0
100.1
124.4
142.6
110.7
145.7
91.2
119.9
97.6
146.4
157.6
152.0
159.5
1 f!7
1U /
137.8
87.8
116.7
135.6
107.2
142.1
86.2
116.9
89.2
140.1
155.1
148.1
154.7
1 nR
lUo
135.2
123.0
101.3
141.6
119.1
176.6
99.3
140.2
94.7
136.2
177.4
154.7
163.8
i no
1 \J7
126.2
103.6
107.1
123.5
123.6
159.9
82.2
95.9
90.2
171.8
152.8
137.8
139.1
*Substituted value.
     There is one value in the table that is not an actual method determination, that made by Lab 103 in
Run 10.  The actual reported value was 205.5 mg/scm, but it was noted that the collaborator had detected
a broken probe at the end of the run, and that the material collected on the filter was of an unusual nature.
As a result, the value was eliminated, and the data point replaced using the technique of Yates, as given in
Snedecor and Cochran^  '. While not the most desired option, it is necessitated by the choice of model
for  the ANOVA.
                                                17

-------
      In previous collaborative tests of Method 5, there has been a tendency toward occasional high values
                                                                      19 ^
which have adversely affected the precision estimates obtained. It was noted1  ' that there was generally a
tendency for the acetone wash to contain an unusually high percentage of the total particulate collected.
As a check on the data from this test, the percentages of total particulate collected from the acetone wash
were calculated, and these are shown in Table 2.

                   TABLE 2. PERCENTAGE OF COLLECTED MASS IN ACETONE WASH

Run
1
2
3
4
5
6
7
8
9
10
11
12
13
Labs
101
A
31.0
21.9
20.9










B
33.5
34.7
30.5










102
A



22.1
29.4
34.1
19.7
19.0





B



12.6
21.9
19.1
22.8
17.8





103
A








40.1
45.0*
51.5
72.7
58.1
B








29.0
43.1
38.8
46.0
44.8
104
44.3
50.1
49.2
53.2
50.8
50.9
51.6
50.6
48.5
52.1
47.7
56.1
57.3
105
51.0
59.3
48.8
49.0
47.5
53.5
68.5
51.4
48.4
52.9
65.4
59.5
52.9
106
28.3
42.8
19.9
27.6
13.8
19.7
24.0
27.6
21 5
30.0
21.9
38.2
31.1
107
31.2
30.6
25.5
32.1
24.4
33.6
29.7
31.9
25.8
28.1
27.9
27.2
30.0
108
52.1
50.6
38.6
42.9
45.2
50.3
43.8
34.8
40.2
46.7
45.7
39.5
36 7
109
45.7
32.6
35.5
40.2
51.7
64.6
47.8
32.6
48.2
52.4
40.0
33.7
35.5
*Concentration not used in subsequent analyses
     As can be seen, the percentages are fairly consistent within any single laboratory's data. The amount
of particulate in the acetone wash is related to the length of the probe. Since the pairs of laboratories used
probes of approximately equal length, the laboratories working on the same paired train show similar per-
centages.

     The data are analyzed using a two-way ANOVA, with the run and laboratory terms as factors, and the
interaction  between the two taken to be zero. The model is discussed in detail in Appendix B.I.  The data
are treated  first according to the three separate weeks using 7 laboratories and 8 trains,  then for the entire
13 sampling runs using only the six laboratories who participated in the entire test. Where appropriate,
contrasts among the means of the laboratories are used to determine the sources  that contributed to the
variation in a  Method 5 concentration determination.  The precision of a method result and the conclusions
that can be drawn from the data are discussed below.  In addition, the data are treated as  if they were results
for a compliance test, and corrected to 12% CO2, as called for in the test methods and procedures section
of the Federal Register pertaining to incinerators.'2^ These values are submitted to analysis according  to the
same model as the uncorrected particulate concentrations, and the results discussed in a subsequent section.

C.   Precision of Method  5
     The particulate concentration determinations are used to estimate standard deviations for within-
laboratory and laboratory bias components, and from these, the between-laboratory term is estimated.
The three ANOVA's, one per week of the test, are used to obtain the estimated within-laboratory term.
TABLE 3. ESTIMATED PRECISION BETWEEN
      PAIRED-TRAIN LABORATORIES
Lab
101
102
103
Pooled
SS
148.01
206.71
2125.87
2480.59
df
3
5
5
13
MS
49.34
41.34
425.17
190.81
a
7.03
6.43
20.62
13.81
&,%
5.8
5.2
13.9
10.4
     There are two possible estimates of the within-lab-
oratory variance.  One is obtained from the two trains
being operated by a single laboratory.  A sum of squares
is calculated between the two trains for each run, with
1 df per run as a result. The estimated variances are shown
in Table 3 for the three dual train laboratories.  Testing
for the equality of variance among the three estimates
shows that they can be considered equal at a significance
level of 0.03. This would allow the pooled estimate to
be used.
   18

-------
      Similarly, the run by lab interaction term is an estimate of sampling error under the assumptions of the
 model.  There are three estimates of a2 using the interaction mean square and these are shown in Table 4.
                                                    Using Bartlett's test, the three mean squares are
    TABLE 4.  ESTIMATED WITHIN-LABORATORY
      COMPONENTS FROM INTERACTION TERM
Week
1
2
3
Composite
SS
2,867.98
4.419.77
9,236.25
16,524.00
df
14
28
27
69
MS
204.86
157.85
342.08
239.48
a
14.31
12.86
18.50
15.48
0,%
11.7
10.4
12.4
11.7
tested to see if they are estimating the same true vari-
ance.  The significance level of the test is 0.13, so
the hypothesis of equality is accepted.  This allows
the use of the pooled estimate, which has the greater
df. Comparing the composite mean squares from
Tables 3 and 4, it can be seen that while the second
is larger, the two are close to one another and thus
are probably estimating the same true variance, a2.
                                                         In view of the definition of the within-laboratory
                                                    term, however, the estimate from between the trains
operated by a single laboratory team is selected.  Since both trains were sampling essentially the same gas stream
at the same time, this variability most nearly conforms to the concept of a replicate.  In Appendix B.2, con-
trasts are used to test that the two trains were getting equivalent measurements, and the hypothesis is accepted
for each of the three labs.

      The estimated within-laboratory variance, then is

                                     a2 =MS    . ,
                                     u       pooled
                                        = 190.81

with 13 df. The estimated within-laboratory standard deviation is

                                   a =

                                     = Vl90.81

                                     = 13.81 mg/scm.

and using the overall mean, n , of 132.66, this gives an estimated coefficient of variation of



                                     =  13.81/132.66

                                     =  0.104.

In terms of relative variation, then, replicate measurements by a single laboratory would have a standard deviation
of 10.4% of the mean value. Previous tests in which the concentrations varied over a wider range than in this
one indicated that the percentage variation was a valid model for a Method 5 result.

     The laboratory bias variance is estimated from the labs term in the 3-week, 6 laboratory ANOVA. The
sums of squares for labs is partitioned by means of orthogonal contrasts into SS for the differences between the
single-laboratory pairs. These represent  variation between independent laboratories sampling the same gas
stream at the  same time.   The three SS are

                                      SSj =2117.72

                                     SS2 =  184.66
and
                                      SS3 =  858.13,
                                                 19

-------
each with 1 df. Pooling these gives an estimated mean square of

                                             SS pooled
                                              df pooled

                                             3160.51
                                                3

                                           = 1053.50

Using the EMS of this term, the laboratory bias variance is estimated as

                                    al = 66.36

with 3 df. This gives a laboratory bias standard deviation  of
                                       = V66.36

                                       = 8.15 mg/scm.

      In terms of relative variation, a coefficient of variation may be estimated for the laboratory bias com-
ponent. This gives
                                       = 8.15/132.66

                                       = 0.061

or 6.1% of the mean value.

      The laboratory SS is also partitioned to test whether a difference exists between pairs of laboratories.
There has been speculation that differences among labs result from the fact that the labs sample the points
at different times and that these differences result from a changing particulate loading during the course of
a two-hour run.  If this is the case, there should be a noticeable difference between the pairs of laboratories.

      There was no significant difference among the pairs of trains in the three individual weeks analyses.
For the 6-laboratory analysis, the pairs term was significant, but the mean square partitioned out of the
labs term for pairs is indistinguishable from the mean square for between paired labs shown above. The con-
clusion, then, is that this difference is due to laboratory differences unrelated to the  sampling location of the pairs.

      The lack of a difference among pairs by itself does not assure that there was no effect due to changing par-
ticulate loading. The comparison of the mean values could mask out fluctuations in  the values during the course
of the runs that were related to the order of sampling the four ports. If these gas stream-related fluctuations occur,
then the error term associated with the differences between pairs should be significantly different from the sam-
pling error estimated for  replicate particulate determinations.  The error mean square for pairs is 257.28, with  12
df. Using an F-ratio test  to compare this to the estimated within-laboratory variance, the difference is not signif-
icant. Thus there is no evidence that the differences among laboratories for this test are related to the changing
pattern of particulates flowing in the gas stream, due to differences in the time a particular point is sampled.
Thus, for this test site, spatial/temporal effects in the stack gas stream did not contribute to the variability in
the particulate concentration determinations.
                                                 20

-------
      The between-laboratory variance, a\, thus is composed of only within-laboratory and laboratory bias
terms. The estimate from the above terms is
                                      = 190.81 +66.36

                                      = 257.17,

which gives a between-laboratory standard deviation of
                                    = 16.04mg/scm.

There are 3 df associated with this estimate, due to the laboratory bias term. The estimated coefficient
of variation, then, is
                                   = 16.04/132.66

                                   = 0.121 mg/scm

or 12.1% of the mean value.

D.    Rank Test for Equality of Laboratories

      An alternative technique for determining if all laboratories were obtaining the same concentration is by
applying a rank test to the data. The purpose of this test is to indicate whether there are differences among
the labs where the distributional nature of the determinations is unspecified.

      The test selected is the Friedman'  •* test, which allows for a blocking factor, in this case runs, to be in-
cluded. The data are analyzed first with 7 laboratories for each week of testing, then with 6 laboratories for
the full 13 runs. The details  of the analyses are described in Appendix B.3.

      The ranks of the data are shown in Table 5 for the separate weeks and in Table 6 for the 6 labs through
three weeks.  The significance levels obtained for the three separate weeks indicate that during the first
week there was no perceptible rank ordering, while the second week showed some order tendency and the
third showed a marked tendency.  However, looking at the six labs who participated in all 13 runs, the
probability of obtaining those rank sums due to chance alone is approximately 1 in  1000.

      Lab 104 showed a strong tendency toward giving higher results than the other 6 laboratories, with
6 determinations having  the highest values and no determination being in the lower half of the six values
on a given run.  At the opposite end, Lab 107 has only one determination among the three highest observa-
tions in a run, while Lab 106 tends toward the middle of the 6 on all runs. If these differences are due to
the pattern of the particulates flowing in the gas stream, then the pairs of trains should be similar.  In fact,
however, they are not. In contrast to Lab  104, Lab 105 has a mix of high and low values. The dissimilar
results from Labs 106 and  107 are from a paired train, and the ranks between Labs 108 and 109 appear
unrelated to each other.

      The apparent conclusion, then, is that the differences between labs result not from the pattern of the
gas  flow but from lab-to-lab  differences in such things as the equipment used and the procedures followed
in obtaining a Method 5  result.
                                                 21

-------
 TABLE 5.  RANKS OF CONCENTRATION
DETERMINATIONS  INDIVIDUAL WEEKS


1
2
3
Sums


4
5
6
7
8
Sums


9
10
11
12
13
Sums
Labs
101
6
4
3
13
104
1
1
1
3
105
7
7
2
16

102
3
6
5.5
7
6
27.5
104
1
3
3
2
2
11
105
2
1
7
1
5
16
106
2
5
4
11
107
3
6
5
14
108
4
2
7
13
109
5
3
6
14
Labs
106
4
5
4
4
3
20
107
6
7
5.5
5
4
27.5
108
5
4
1
3
1
14
109
7
2
2
6
7
24
Labs
103
2
2.5
2
1
1
8.5
104
4
2.5
3
2
2
13.5
105
1
5
7
6
5
24
106
3
4
4
4
4
19
107
7
6
5
5
6
29
108
5
7
1
3
3
19
109
6
1
6
7
7
27
 TABLE 6. RANKS OF CONCENTRATION
DETERMINATIONS-THREE WEEKS DATA
Run
1
2
3
4
5
6
7
8
9
10
11
12
13
Sum
Labs
104
1
1
1
1
3
3
2
2
3
2
2
1
1
23
105
6
6
2
2
1
6
1
5
1
4
6
5
4
49
106
2
4
3
3
5
4
4
3
2
3
3
3
3
42
107
3
5
4
5
6
5
5
4
6
5
4
4
5
61
108
4
2
6
4
4
1
3
1
4
6
1
2
2
40
109
5
3
5
6
2
2
6
6
5
1
5
6
6
58
E.   Velocity

     Method 5 calls for the determination of the
average velocity of the gas stream by Method 2 for
calculation of isokinetic variation. The data from
the sampling run are used in the equation:
where

   Cp          pilot tube coefficient

   (\/A/?)aVg  ~ average of the square root of the
               velocity head

   (Tj)avg    -- average stack temperature

   Ps         - stack pressure

   Ms          molecular weight of the stack gas.

The calculated velocities from the six laboratories
that  participated in all  13 runs are shown in Table 7.
The determinations from the  dual trains are not used
in order to facilitate blocking the data. In a previous
report on Method 2,'  ' the precision of the velocity
determination was shown to be proportional to the

  TABLE 7. DETERMINED VELOCITY OF THE GAS
     STREAM, ARRANGED BY BLOCK (m/sec)


Labs
104
105
106
107
108
109
Block 1
1
2
7
8
9
13.9
13.9
13.6
14.3
14.2
14.5
14.4
14.0
14.4
14.3
15.1
14.4
14.8
14.9
14.9
15.4
14.8
146
14.8
15.0
15.3
15.5
15.0
15.7
16.2
14.3
14.7
14.4
14.8
15.3
fl/OC* 2
3
4
5
6
10
11
12
13
14.4
14 2
14.1
14.4
14.5
14.6
14.9
14.8
14.7
14.8
14.3
14.7
15 1
14.9
15.4
15.3
15.8
15.7
15.1
15.4
15.5
15.5
15.9
15.9
16.2
16.0
15.1
15.6
15 5
15.6
16.1
159
16.0
16.5
16.2
15.9
16.2
16.6
16.3
16.5
15.7
16.1
14.9
15.2
149
15.2
15.3
15.0
                                          22

-------
 velocity in the range 13 to 18 m/sec.  In accordance with this, the precision estimates were made using a co-
 efficient of variation approach, estimating coefficients of variation for the between-laboratory, within-laboratory
 and laboratory bias terms. The estimates are obtained in Appendix B.5. The model is
                                           0 =
where
and
(3 — the true coefficient of variation

a — the true standard deviation



ju - true mean velocity.
 For the purposes of the analysis, the runs are grouped into blocks where the mean velocity is approximately
 the same for all the runs.  The blocking criterion is (\/Ap)aVg, which is the principal determinant of the
 velocity.  The sample mean of the (VAp)avg's is taken across runs, and two blocks are formed. The run
 averages and the block to which those runs are assigned are given in Table 8.
  TABLE 8.  BLOCKING CRITERION
           FOR VELOCITY
Run
1
2
3
4
5
6
7
8
9
10
11
12
13
(\/Ap)avg
0.64
0.63
0.65
0.67
0.65
0.65
0.63
0.64
0.63
0.66
0.66
0.67
0.66
Block
1
1
2
2
2
2
1
1
1
2
2
2
2
                                     The blocked data were submitted to the coefficient of variation
                               approach described in detail in Appendix B.4. A mean and a standard
                               deviation are calculated for each run across the 6 trains.  The ratio of
                               the standard deviation  to the mean, multiplied by a correction factor,
                               gives a single estimate of the between-laboratory coefficient of variation,
                               ft,.  Averaging the 13 values thus obtained gives
                                                                  1
                                                                    13
                                                               = 0.043.

                                     There are 5 df associated with this estimate. This gives an estimated
                                     standard deviation of

                                                              Ob = ft,M
                                                                 = (0.043)ju
 or 4.3% of the mean value.
      The within-laboratory coefficient of variation is estimated by obtaining a standard deviation and a
 mean from the velocities obtained by each collaborator with a given block.  Since the blocks are of un-
 equal size, the individual beta values are weighted so that a greater contribution is made to the final estimate
 by the values obtained from the larger numbers of determinations. The weighting procedure is described in
 Appendix B4.

      There are 12 collaborator-block combinations, and the estimated within-laboratory coefficient of
 variation is
                                              12
                                     -
                                     0 = —
                                         12
                                            /= 1
                                                  23

-------
                                      = 0.022

with 66 df. Thus, the estimated within-laboratory standard deviation is

                                         a = (3M

                                          = (0.022)M,

or 2.2% of the mean value.

     From these, the laboratory bias coefficient of variation can be estimated as
                                 = V(0.043)2 -(0.022)2

                                 = V(0.0014)

                                 = 0.037.

This gives an estimated laboratory bias standard deviation of

                                       °L = 0lA<

                                          = (0.037)M,

or 3.7% of the mean value.

           In comparison with the previous report, the estimated coefficients of variation were

                                       fa = 5.0%

                                       0  =3.9%

and

                                       fa = 3.2%

for between-laboratory, within-laboratory and laboratory bias, respectively. These values compare favorably
with those determined in this report, and the conclusion is that a total variation between laboratories of 4 to
5% should be expected.

F.    Stack Gas Analysis

      Test procedures require that EPA Method 3 be used to determine the stack gas composition and the
dry gas molecular weight using an Orsat analyzer or equivalent. All collaborators in this test used Orsat
analyzers on an integrated gas sample.  The teams operating the paired train analyzed a single gas sample for
both trains.  The procedures  for determining C02 and 02 given in paragraph 4.3.3  of revised Method 3
were strictly adhered to.  (Appendix A-2).

      There are four missing  data points in each set of data. One laboratory did  not take  Orsats for the first
four runs but used the results from the laboratory with which they  were paired.  Other values within the
sets require qualification. One laboratory did not bring the necessary equipment with them to take an
integrated sample but performed independent analyses on the gas sample from the  other laboratory team
on that train.  In two other instances, another team's gas sample was used, but independent analyses were

                                                 24

-------
made.  These analytical results represent the use of the method with the exception of the collection of the
gas sample, where minimum variability would be expected. For this reason, they are retained in the data set
for statistical analysis.

     Three variables are submitted to statistical analysis to determine the precision of their determination.
These are the percent carbon dioxide, percent oxygen, and dry gas molecular weight.  The same analysis of
variance model is used for each component of interest to determine its precision components.

     The data are divided into three analytical sets, one for each week of the test. This is necessitated by
the fact that the paired train laboratory changed from week to week.  The runs are considered to be repeti-
tions at the same level of each component.  The justification for this was a preliminary analysis which showed
that the runs were an insignificant factor in the model.  The first week has 6 laboratories and three runs, and
the next two weeks have 7 laboratories and 5 runs each.  The first result in week 2 for laboratory  109 was a
replacement value, calculated in the manner described in Snedecor and Cochran.   '  The error term is
reduced by 1 degree of freedom as a result.

     Precision estimates of the laboratory bias and within-laboratory terms are calculated for each week and
then pooled to provide the overall estimates for the method.

     1.    Percent CO2

           The determinations of the percent CO2 made by the collaborators are shown in Table 9. The
ANOVA tables are shown in Appendix B6.  The laboratory term was significant for the second and third
weeks,  but not the first. The consistently low values for C02 by labs  107 and 108 indicate a possible
                                                    leak in the Orsat Analyzers.  As a check on this
                                                    assumption, the combined CO2 + 02 values for
                                                    lab  104 through 109 were examined and not
                                                    found to be significantly different. This indicates
                                                    that the low CO2 values reported were probably
                                                    due to incomplete C02 absorption rather than
                                                    a system  leak.
                                                              The three within-laboratory variance
                                                    estimates were
TABLE 9. PERCENT CO2 DETERMINATIONS


1
2
3


4
5
6
7
8


9
10
11
12
13
Labs
101
2.8*
2.7
3.2
104
2.6
2.3
2.6
105
2.6
2.3
2.2
106*
3.0
2.0
2.5
107
2.7
1.9
2.0
108
2.5
2.4
1.8
109
-t
-t
-t
Labs
102
2.3*
2.7
2.3
2.6
2.7
104
2.5
2.6
2.6
2.2
2.3
105
2.5
2.5
2.6
2.5
2.5
106*
2.6
2.7
2.6
2.7
2.8
107
2.2
2.4
2.0
2.4
2.2
108
1.5
1.5
1.4
1.6
1.7
109
-t
2.6
2.5
2.1
2.4
Labs
103
2.5
2.8
3.1
3.1
2.7
104
2.6
2.4
2.4
2.5
2.5
105
2.4
2.6
2.6
2.5
2.6
106*
2.6
2.7
2.5
1.8
1.9
107
2.2
2.4
2.1
1.8
2.0
108
1.7
1.7
2.0
2.0
2.0
109
2.3
2.6
2.4
2.6
2.5
*Orsat analysis performed on another laboratory's gas
sample.
fOrsat analysis not performed.
                                                   and
                                                                 a] =0.12
                                                                 a\ = 0.02


                                                                 al = 0.05
                                                   with 12, 27 and 28 df, respectively. Pooling these
                                                   gives an estimated within-laboratory variance of
                                                                      J2=0.04
                                                   with 67 df. The within-laboratory standard deviation,
                                                   then is
                                                   the two weeks are
                                                   and
                                                                        = 0.20 percent CO2 .
                                                              The estimated laboratory bias terms for

                                                                          = 0.14
                                                                          =0.09.
                                                25

-------
Combining these gives
                                      aL2=0.12
with 6 df. This results in a laboratory bias standard deviation of
                                         = 0.35 percent CO2.

           The between-laboratory variance is estimated from the above to be

                                      al=?+°i

                                         = 0.04 + 0.12

                                         = 0.16

with 6 df from the laboratory bias term. The between-laboratory standard deviation, then, is
                                         = 0.40 percent CO2.

           In a previous report, ™' the estimated standard deviations were 1.44, 1.06 and 1.78 percent CO2
for within-laboratory, laboratory bias and between-laboratory, respectively. The CO2 levels in that study
were considerably higher, however, ranging from 5 percent to 12 percent for the three test sites studied.
This suggests that the variability may be a function of the concentration of CO2 in the gas stream.

           To investigate this possibility, the between-laboratory components are examined on a relative
basis using the coefficient of variation. For this test, the coefficient of variation is estimated to be

                                      4 == X 100%
                                          X
                                           0.40
                                         = — x  100%
                                           2.4

                                         = 16.7%

For the previous report's data, the overall mean is approximately 9.5 percent CO2 . This gives an estimated
coefficient of variation of

                                      $b = a~ x 100%
                                           X
                                           1.78
                                         = - X  100%
                                           9.5

                                         = 18.7%

close to that for this test. The indication is that if this is the proper model for the CO2 determination, one
could expect  the CO2 variation to be between 15% and 20% of the  mean concentration.
                                                 26

-------
      2.    Percent C>2

           The determinations of 02 content in the stack gas are shown in Table 10. These are analyzed
under the same model as the C02's, and the ANOVA tables and discussion are presented in Appendix B.6.
The laboratory term was significant in all three weeks of testing.
       TABLE 10.  PERCENT O2 DETERMINATIONS


1
2
3

4
5
6
7
8

9
10
11
12
13
Labs
103
17.7*
17.9
17.5
104
17.8
17.7
17.7
105
17.9
18.2
18.3
106*
17.2
18.1
17.6
107
18.1
18.3
17.6
108
18.5
18.6
19.5
109
-t
-t
-t
Labs
102
18.4*
18.1
17.7
17.8
17.6
104
17.7
17.8
17.7
18.1
18.1
105
18.0
18.3
17.8
17.8
18.1
106*
18.0
17.4
17.6
17.4
17.1
107
18.4
18.3
18.7
18.2
18.5
108
19.2
19.2
19.2
19.2
19.1
109
-t
18.2
18.2
18.1
18.0
Labs
103
17.0
17.2
16.7
16.5
18.1
104
17.8
17.5
17.7
17.7
17.9
105
18.2
17.9
18.0
18.0
18.1
106*
17.5
17.4
17.7
18.4
18.3
107
18.5
18.4
18.7
19.2
19.0
108
19.5
19.1
18.6
18.8
18.7
109
17.7
17.8
17.8
17.7
17.8
*Orsat analysis performed on another laboratory's gas sample.
•fOrsat analysis not performed.
                                                                 The within-laboratory variance esti-
                                                      mates from the three cases are

                                                                      o2 =0.12
                                                      and
                                                                      a\ = 0.06
                                                                         =0.13
                                                      with 12, 27 and 28 degrees of freedom, respec-
                                                      tively.  Pooling these estimates gives

                                                                      o2 =0.10

                                                      with 67 df. The estimated within-laboratory
                                                      standard deviation, then, is
                                                                        = 0.32 percent 02.

                                                                 The laboratory bias variance is estimated
                                                      as
                                                      and
a,* =0.14
 *-1
ar 2 = 0.27
 '' 2
a, 2 = 0.38
          '
respectively, from the three weeks.  Combining these gives an estimated laboratory bias term of

                                      6L2 =0.27

with 6 df. The laboratory bias standard deviation is estimated as

                                      OL =V5~27

                                         = 0.52 percent O2 .

           The between-laboratory variance is estimated from the above terms as

                                      af,=o2+aL2

                                         = 0.10 + 0.27

                                         = 0.37.
                                                27

-------
This gives an estimated between-laboratory standard deviation of
                                         = 0.61 percent 02

           In the previous report/6* the standard deviations were estimated to be 1.70, 1.66 and 2.14 per-
cent O2 for the within-laboratory, between-laboratory and laboratory bias terms, respectively.  As in the
case of the C02 results, these are larger than the estimates obtained here. The oxygen concentration at the
previous sites were lower, however. The variability in 02  determination appears to be more nearly related
to the variability in CO2  determination than to the 02 content of the gas stream.  There is an apparent trade-
off between C02 and 02  determinations that causes their precision to be related to one another.  This result
would not be unexpected, since any C02 not absorbed in the first burette in the Orsat Analyser would be
absorbed in the second burette, leading to an increased 02 value.

     3.    Molecular Weight
           The dry gas molecular weight (Md) is determined from the results of the Orsat analysis. The
values obtained for Md in this test are shown in Table  11. In Appendix B.6, the ANOVA tables are pre-
sented and the precision estimates derived.
          TABLE 11.  DRY GAS MOLECULAR WEIGHT
               DETERMINATIONS (gm/gm-mole)

Run
1
2
3

4
5
6
7
8

9
10
11
12
13
Labs
101
29.16*
29.14
29.21
104
29.13
29.08
29.12
105
29.13
29.10
29.08
106*
29.17
29.04
29.10
107
29.16
29.04
29.02
108
29.14
29.13
29.06
109
-t
-t
-t
Labs
102
29.10*
29.16
29.08
29.13
29.14
104
29.11
29.13
29.12
29.07
29.09
105
29.12
29.12
29.13
29.13
29.13
106*
29.14
29.13
29.12
29.13
29.13
107
29.09
29.12
29.07
29.11
29.09
108
29.01
29.00
28.99
29.02
29.04
109
-t
29.14
29.13
28.06
29.10
Labs
103
29.08
29.13
29.16
29.15
29.16
104
29.13
29.09
29.10
29.11
29.12
1 105
29.11
29.13
29.13
29.12
29.14
106*
29.12
29.13
29.10
29.03
29.04
107
29.09
29.12
29.08
29.06
29.08
108
29.05
29.04
2906
29.08
29.07
109
29.08
29.13
29.10
29.12
29.11
*Calculated using Orsat analysis on another laboratory's gas sample.
fOrsat analysis not performed.
and
                                      a,2  =0.0015
                                       L' •>
                                      a,2  = 0.0006.
           The within-laboratory var-
iance is estimated to be

           a\ = 0.0024
                                                               and
                                                                          a\ = 0.0005
                                                                          a\ = 0.0007
                                                               for the three weeks. There are 1 2, 27
                                                               and 28 df associated with these esti-
                                                               mates, respectively.  Pooling these
                                                               gives an overall estimate of

                                                                          a2  =0.0012

                                                               with 67 df.  The estimated within-
                                                               laboratory standard deviation, then, is

                                                                          a  = ^00012

                                                                             = 0.035 gm/gm-mole.

                                                                          The labs term was signifi-
                                                               cant for weeks 2 and 3 of the test. The
                                                               laboratory bias variance estimates are
                                                 28

-------
Combining these gives an estimated laboratory bias variance of

                                      aL2 =0.0011

with 6 df. The estimated laboratory bias standard deviation is

                                      aL  =Vol)On

                                          = 0.033 gm/gm-mole.

           The between-laboratory variance, 62b, is estimated from the above as

                                      a£=a2+a,,2

                                          = 0.0012 + 0.0011

                                          = 0.0023,

with 6 df. The between-laboratory standard deviation is estimated to be
                                         = 0.048 gm/gm-mole

with 6 df.

           The estimates for the previous study1^' were 0.20, 0.14 and 0.24 gm/gm-mole for within-
laboratory, laboratory bias and between-laboratory, respectively. The greater imprecision in the previous
report's results is undoubtedly due to the greater variation in the CO2 and 02 values used to determine Md.

G.    Moisture Fraction

      Included in the revised Method 5 is a provision for the fraction of moisture in the stack gas to be
calculated. The formula is
                                 "WS   y^   + y^


where

     Bws   — moisture fraction

            —volume of water vapor collected, corrected to standard conditions

            - metered volume of gas, corrected to standard conditions.
The moisture fractions determined during the test are shown in Table 12 for the eight trains. These are sub-
mitted to statistical analysis using an ANOVA model. A two-way model without interaction is used to avoid
blocking the runs, and the run-by-train interaction is used for the error term.  The details of the analysis are
given in Appendix B.7.
                                                29

-------
                           TABLE 12. DETERMINED MOISTURE FRACTIONS


1
2
3

4
5
6
7
8

9
10
11
12
13
Labs
101A
0.155
0.151
0.172
101B
0.155
0.156
0.176
104
0.160
0.149
0.165
105
0.117
0.146
0.141
106
0.155
0.148
0.164
Labs
102A
0.169
0.138
0.163
0.144
0.152
102B
0.175
0.142
0.159
0.144
0.156
104
0.169
0.152
0.158
0.177
0.163
105
0.140
0.151
0.157
0.151
0.162
106
0.174
0.152
0.156
0.144
0.150
107
0.145
0.174
0.174
108
0.139
0.156
0.164
109
0.155
0.154
0.173

107
0.150
0.158
0.148
0.156
0.135
108
0.180
0.158
0.158
0.152
0.157
r!09
0.174
0.153
0.162
0.152
0.162
Labs
103A
0.129
0.169
0.172
0.162
0.149
103B
0.135
0.170
0.172
0.164
0.155
104
0.142
0.165
0.165
0.165
0.159
105
0.137
0.158
0.165
0.155
0.147
106
0.130
0.166
0.167
0.156
0.153
107
0.168
0.145
0.172
0.163
0.150
108
0.131
0.168
0.173
0.163
0.154
109
0.136
0.168
0.175
0.162
0.152
           The three weeks are analyzed separately, and the run term is significant in each analysis.  The
     estimated within-laboratory variance, a2, is
and
                                      a? = 0.000070
                                      a! = 0.000086
                                         = 0.000060
from weeks 1, 2 and 3.  There are 14, 28 and 28 df, respectively, associated with these estimates.  Pooling
these terms gives

                                      o2 = 0.000073

and an estimated within-laboratory standard deviation of
                                      0  = v°

                                         = 0.009.

There are 70 df associated with the pooled estimate.

     The trains factor was significant only for the first week's data.  The estimated laboratory bias variance
                                                30

-------
                                      o2L = 0.000064

with 7 df. This gives a laboratory bias standard deviation estimate of

                                      °L =V^

                                        = 0.008.

The between-laboratory variance, a£, is estimated to be

                                      %=?+  °l

                                        = 0.000072 + 0.000064

                                        = 0.000136.

The estimated between-laboratory standard deviation, then, is
                                         = \/0.000136

                                         = 0.012.

There are 7 df associated with this term, from the laboratory bias component.

     In a previous report, the precision components for moisture fraction determination were estimated
to be

                                      a  =0.032

                                      a/ = 0.032

and

                                      6b = 0.045.

In the prior report/6' the absence of several values from the data set necessitated using runs as repetitions,
and undoubtedly caused the error term to be inflated due  to run-to-run variation in stack moisture. The
higher precision estimates also may result from the higher  moisture content of some of the streams sampled
to obtain the prior estimates.

H.   Precision of Particulate Loadings Corrected to 12% COz

     The standard for particulate matter emissions from municipal incinerators is given in terms of
particulate loading corrected to 12 percent CO2.  To obtain this value, the  factor

                                           12
                                          %C02

is applied, where %C02 is obtained from the Method 3 results. Thus the standardized concentration, C', is

                                        C' = k-C
                                                31

-------
where Cis the determined Method 5 concentration. To evaluate the use of these EPA Methods in obtaining
a compliance test result, the collaborators' data are used to obtain corrected concentrations, and  these are
submitted to statistical analysis in the same manner as the uncorrected concentration determinations.  The
corrected concentrations are shown in Table 13, while the statistical analysis is discussed in detail in Appen-
dix BS.The within-laboratory  variance is estimated from the difference between the two trains run by a
single laboratory. The estimates obtained are summarized in Table 14.  Using Bartlett's test, the  three MS's
can be shown to be estimating the same true variance, a2, at a 5% significance level. Thus, the best estimate
of the within-laboratory variance is obtained from the pooled estimate, giving
with 13 df.
                                      a2 = 3072.88
              TABLL 1 3. PARTICULATE CONCENTRATION CORRECTED TO 1 2% CO2, mg/scm

Run

I
2
3
4
5
6
7
8
9
10
11
12
13
Labs
101
A
505 3
448 0
481.9










B
568.3
471.6
508 9










102
A



751.8
459 6
787.3
378 0
450 2





B



772 7
499.1
695.5
377 1
463.1





103
A








485.8
674.1
657.7*
985.2
8382
B








516.5
677.6
624.4
736.3
844.0


874.6
677.7
664.2
747.8
555.2
689.1
552.0
715.8
449.5
790.5
814.0
787.7
888.5


564.5
414.3
736.4
724.8
598.6
569.1
621 1
547.7
518.9
665.5
574.2
704.2
726.5


556.0
600.6
597.1
658.2
492.0
6725
405.3
513.9
433 8
650.7
756.5
1013.3
1007.4


6124
554.5
700.2
739.6
536.0
710.5
431.0
637.6
486.6
700 5
886.3
987.3
928.2
i OK

649.0
615.0
675.3
1132.8
952 8
1513.7
744 8
989 7
668.5
961 4
1064.4
928.2
982.8
1 OQ

605 8
5180
714.0
988 0
570 5
767.5
469.7
479.5
4706
792.9
764.0
636 0
667.7
"^Substituted value
    TABLE 14.  PRECISION ESTIMATION BETWEEN
          PAIRED-TRAIN LABORATORIES
     (Particulate Concentration Corrected to 12% CO2)
The estimated within-laboratory standard deviation,
then, is
Lab
101
!02
103
Pooled
SS
2.62748
5,295 76
32.024 20
39.947.44
df
3
5
5
13
MS
875.83
1,059.15
6.404.84
3.072.88
a
29 59
32.54
80.03
55.43
&,%
5.0
4.9
10.8
82
                                                                      = 55.43 mg/scm, corrected to
                                                                        12%C02.

                                                    The coefficient of variation is estimated using the
overall mean of 677.13, giving,
                                      §=<*//;

                                        = 55.43/677.13

                                        = 0.082
or 8.2 percent of the mean level.
                                                  32

-------
      If the ordinary error term from the ANOVA were to be used, the pooled estimate of a would be

                                         a =96. 52 mg/scm

 corrected to \2'7o CO2 and the estimated coefficient of variation would be
                                     0  =0.143

 or 14.3% of the mean value.  The discrepancy between the two probably results from the fact that, in the
 paired-train laboratories' results, there is no additional variation induced by the Orsat data, since a single
 sample was taken for both trains. From run to run by the same laboratory, however, there is apparently
 some additional error due to the determined C02  concentration.

      The laboratory bias variance is estimated from the contrasts between paired laboratories, as before.
 The pooled laboratory mean square is 174,868.84. Using the EMS of this term, the estimated laboratory
 bias variance is

                                      a\= 13,215.07,

 with 3 df.  The estimated laboratory bias standard deviation, then, is
                                         = 1 14.96 mg/scm, corrected to 12% C02.
The estimated coefficient of variation is
                                        = 114.96/677.13

                                        = 0.170

or 17.0% of the mean value.

     The between-laboratory component, CT|, is estimated from the above as


                                      a\ = a + OL

                                         = 3072.88+ 13,215.07

                                         = 16,287.95.

This gives an estimated between-laboratory standard deviation of
                                         = 127.62 mg/scm, corrected to 12% C02

with 3 df, from the laboratory bias term.  In terms of relative variation, then, the coefficient of variation
is estimated to be
                                                 33

-------
                                       _ 127.62
                                         677.13

                                       = 0.188,

or 18.8f/f of the mean value.

      If the alternative within-laboratory term were to be used, the one which takes the run to run variation
in CO2 determination into account, the between-laboratory components would be

                                      a| = 22,530.29

                                      ab = 150.10

and

                                      &b = 0.222

Thus, the concentrations from lab to lab would be expected to have a standard variation, in relative terms,
of 22.2% of the mean value.

           In comparison to the results for a Method 5 concentration determination, it can be seen that,
using the paired train laboratory's results, the within-laboratory relative variation dropped slightly, while
the laboratory bias increased greatly. Also, there was an increase in the relative variation of the error term
taken from the ANOVA.  This increase is due to variation in the C02 determination from the Orsat data.
The between-laboratory standard deviation for C02 is 0.40 percent C02 by volume. If the true C02 level
were 2.3 percent, then two independent laboratories might  obtain values of 2.1 percent and 2.5 percent,
respectively. For two laboratories which had determined the same concentration, C, the corrected concen-
trations would be

                                            /i;
                                                 1C


and

                                      c-, -  ('-£-
and as a result

                                      C\/C'i = 1.19,

or a 19% difference would be induced.
                                                 34

-------
                       IV.  COMPARISON WITH PREVIOUS STUDIES

     The results from the evaluation of Method 5 can be compared to the results of the three previous
studies.  In these studies, the precision components were estimated in terms of coefficients of variation, and
these are summarized in Table 15, along with the estimated coefficients of variation from this study.
          TABLE 15. COMPARISON OF COEFFICIENTS OF
                  VARIATION FOR METHOD 5
Site
Portland Cement Plant
(after high values excluded)
Power Plant
Houston Incinerator
Dade County Incinerator
Coefficients of Variation, %
Within-
Laboratory
28.4
9.8
31.1
25.3
10.4
Laboratory
Bias
51.0
17.6
19.5
29.3
6.1
Between-
Laboratory
58.4
20.1
36.7
38.7
12.1
                                                                       It is easily seen that the esti-
                                                                 mates for this test are considerably
                                                                 lower than those from the previous
                                                                 studies, and there are several pos-
                                                                 sible reasons for this. First, the
                                                                 purpose of this test was to deter-
                                                                 mine the minimum variability that
                                                                 could be expected with the use of
                                                                 Method 5. To accomplish this, the
                                                                 collaborators were chosen from
laboratories who had a great deal of experience using Method 5, and who had been under contract to EPA
for source testing.  In previous studies, the laboratories were chosen from professional source samplers in
the local area in which the testing was conducted, since they were deemed representative of source samplers
in general.  The calibration checks and rechecks required for this test were not specified in previous tests
except to the extent that they were required in proper use of the method. By these checks, some of the
laboratory-to-laboratory differences could be eliminated, as in the case of the laboratories whose meter
boxes were out of calibration upon arrival at the test site.

      Secondly, in  all previous studies, the authors had urged that the procedural details in the method be
tightened and more well-defined. It was felt, and is still felt, that differences from one crew to another in
the manner of handling these procedural details were the greatest single source of variation between labor-
atories. In  the revised Method 5 shown in Appendix A3, the sample handling and recovery  is defined more
precisely, and teams who follow the new method scrupulously should be able to obtain more  reproducible
results.

     In addition, the statistical analyses of the previous studies' data were hampered by results that could
not be  used because of failure to meet either the minimum sampling volume or isokinetic variation criteria.
In this study, only one data point was deemed invalid, where a broken probe liner and contaminated filter
were noted  after the run.  After substituting for this value, a more valid error estimate was obtainable since
it was no longer necessary to perform some type of posterior blocking of the data to obtain a within-laboratory
term. Also, the laboratory bias term was based upon 6 observations per run, instead of the maximum 4, and
often as few as 2 or 3 per run in the previous tests. This has to give a better estimate, and one which is
affected less by an extremely high or low value.
                                                35

-------
APPENDIX A. REVISED EPA METHODS FOR
 PARTICULATE EMISSION MEASUREMENT
               37

-------
                                        APPENDIX A
A1.  Method 2—Determination of Stack Gas Velocity and Volumetric Flow Rate
      (Type S Pitot Tube)

1.    Principle and Applicability

      1.1   Principle. Stack gas velocity is determined from the gas density and from measurement of the
velocity head using a Type S (Stausscheibe or reverse type) pitot tube.

      1.2   Applicability.  This method should be applied only when specified by the test procedures for
determining compliance with the new source performance standards.
           This procedure is not applicable for direct measurement in cyclonic or swirling gas streams.
(Method 1,  section 2.4 shows how to determine unacceptable flow conditions.) When these conditions
exist, procedures such as the use of flow straightening devices must be employed as necessary to make
accurate flow rate  determinations.  Such procedures are subject  to approval by the Administrator.

2.    Apparatus

      Specifications for the apparatus are given below. Any  apparatus which has been demonstrated to the
Administrator's satisfaction to be capable of meeting the specifications will be considered acceptable for the
purposes of this method.

      2.1   Pitot tube. Type S (Figure 2-1), or equivalent, calibrated according to the procedure in section
4. Other devices may be used when approved by  the Administrator.
          1 PC  2 51 cm
          (075  1.0m.)
                                                               LEAK FREE
                                                             CONNECTIONS
                                              MANOMETER
                         FIGURE 2-1. PITOT TUBE-MANOMETER ASSEMBLY
                                              39

-------
      2.2   Differential pressure gauge.  Inclined manometer, or equivalent device, capable of measuring
velocity head to within 10% of the minimum measured value.  Below a differential pressure of 1.3 mm
(0.05 in.) water gauge, micromanometers with sensitivities of 0.013 mm (0.0005 in.) should be used. How-
ever, micromanometers are not easily adaptable  to field conditions and are not easy to use with pulsating
flow. Thus, methods or other devices acceptable to the Administrator may be used when conditions warrant.

      2.3   Temperature gauge.  Thermocouple, liquid filled bulb thermometer, bimetallic thermometer,
mercury-in-glass thermometer, or othei gauges that are capable of measuring temperature to within 1.5%
of the minimum absolute stack temperature. The temperature gauge shall be attached to the pitot tube
such that the sensor does not touch any metal and its position is adjacent and about 1.90 to 2.54 cm
(0.75 to 1  in.) from the pitot tube openings (see Figure 2-1).  If it can be shown to the satisfaction of the
Administrator that a difference of not more than 1% in the velocity measurement will be introduced, the
temperature gauge need not be attached to the pitot tube.

      2.4   Pressure probe and gauge. Piezometer tube and mercury-or water-filled U-tube manometer
capable of measuring  stack pressure to within 2.5 mm Hg (0.1  in. Hg). The static tap of a standard type
pitot tube or one leg of a Type S pitot tube with the face openings positioned parallel to the gas flow may
also be used as the pressure probe.

      2.5   Barometer. Mercury, aneroid, or other barometers capable of measuring atmospheric pressure
to within 2.5 mm (0.1 in. Hg). In many cases, the barometric reading may be obtained from a nearby
weather bureau station, in which case the station value shall be requested and an adjustment for elevation
differences shall be applied at a rate of minus 2.5 mm Hg (0.1 in. Hg) per 30 m (100 ft) elevation increase.

      2.6   Gas analyzer. To analyze gas composition for determining molecular weight. Use Method 3
or other methods specified by the Administrator for dry molecular weight and use Method 5 or Reference
Method 4 for moisture content.  Other methods may be used when approved by the Administrator.

      2.7   Calibration pitot tube.  Standard type, to calibrate the Type S pitot tube.  The standard  type
pitot tube shall have a known coefficient obtained from the National Bureau of Standards, Route 70 S,
Quince Orchard Road, Gaithersburg, Maryland.  An alternative is to use a Prandtl type pitot tube designed
according to the criteria (given below and illustrated in Figure  2-2) which ensure that its coefficient will be
0.99 ±0.01.
                    	  ____!	
                                           D
                                          T
                                                                    r • 3D
                                                                     STATIC
                                                                      HOLES
                                                              HEMISPHERICAL
                                                                   TIP
                                FIGURE 2-2  STANDARD PITOT TUBE
                                                40

-------
      2.7.1 Hemispherical tip (inlet end of the impact tube).

      2.7.2 Eight diameters of straight run (based on the diameter of the external tube) between the tip
 and the static pressure holes.

      2.7.3 Sixteen diameters between the static pressure holes and the centerline of the external tube,
 following the 90° bend.

      2.7.4 Eight static pressure holes of equal size (approximately 1/32 in. diameter), equally spaced in
 a piezometer ring configuration.

      2.7.5 Ninety-degree bend of relatively large radius (approximately three diameters).

      2.8  Calibration differential pressure gauge. For calibration purposes, inclined manometer, or
 equivalent device, capable of measuring velocity head to within 0.13 mm H20 (0.005 in. H20).

 3.    Procedure

      3.1  Set up the apparatus as shown in Figure 2-1. Make sure all connections are tight and leak free.
 Level and zero the manometer.  Because the manometer level and zero may drift due to vibrations and
 temperature changes, make periodic checks during the sample run. Record all necessary data as shown in
 the example data sheet (Figure 2-3).

      3.2  Measure the velocity head and temperature at the traverse points specified by Method 1.

      3.3  Measure the static pressure in the stack. One reading is usually adequate for all measuring
 points during the test; however, this must be confirmed by randomly moving the pressure probe over
 the cross section to see if there are any significant variations, i.e., greater than about 100 mm H2 0
 (4 in. H20).  If there are significant vaiiations, check the location for disturbances.  If none are found,
 measure  and record the static pressure at each traverse point.

      3.4  Determine the atmospheric pressure.

      3.5  Determine the dry stack gas molecular weight.  For combustion processes, use  Method 3.
 For processes emitting essentially air, an analysis need not be conducted; use a molecular weight of 29.
 For other processes, consult the Administrator.

      3.6  Obtain the moisture content from Method 5 or by using Reference Method 4.

      3.7  Determine the cross sectional area of the stack or duct at the sampling location.

4.     Calibration

      4.1   Pitot tube.

      4.1.1 Calibration set-up.  Calibration shall be done in a flow system having the following essential
design features:

      4.1.1.1   The "flowing gas stream" must be confined to a definite cross sectional area, either circular
or rectangular. The projected area of the pitot tube (area exposed  to the gas stream and perpendicular to
the direction of flow) shall be less than 2% of the duct cross sectional area.  Assuming a 0.95 cm (3/8 in.)
diameter pitot tube and a measuring point  at the  centroid, the diameter of a circular duct must be at least
30.5 cm (12 in.) and the width (shorter side) of a rectangular duct must be  at least 30.5 cm (12 in.). If
the pitot tube is to be calibrated with a particulate probe sheath attached, as in Method 5, larger cross
sections must be employed to meet the 2% criterion.

                                                 41

-------
PI APJ!„_.._
D/UL
	HUM I\IU. __
 SIACK DIAMLTLI! 01! DIMENSIONS, 111(111.)
 BAKOMEmiC PRESSURE, mm Hfj (in. Hg)_
 CROSS SECTIONAL AREA, m2(ft2)	
 OPERATORS 	
PITOTTUBE I.D. WO.
   AVG. COEFFICIENT, Cp = .
   LAST DATE CALIBRATED.
                                       SCHEMATIC OF STACK
                                          CROSS SECTION

Pt No.







i
1










Vtl Hd Ap
mm (m.) H20



















Stack Tc
ts, °C(°F)


















Average
npiMHitire
Ts, °K ("R)



















pa
M
mm Hq (m.Hy)




















/Ap



















alf prelinimary investigation shows that Pg varies no more than 100 mm H20
 (4 in H201, record one reading.
                       FIGURE 2-3. VELOCITY TRAVERSE DATA
                                      42

-------
              The cross sectional area must be constant over a distance of 10 or more duct diameters.  For
a rectangular cross section, use an equivalent diameter calculated from the following equation to determine
the number of duct diameters:

                                                 2L W
                                           Dp =	                                Equation 2-1
                                                (L + W)

where:

     De   = equivalent diameter

     L    = length

      W    = width

     4.1.1.2  To ensure the presence of stable, fully developed flow patterns at the calibration site, or
"test section," the site must be located at least 8 diameters downstream and two diameters upstream from
the nearest disturbances.

     4.1.1.3  The system must have the capacity to generate a test section velocity of at least 600 m/min
(2000 fpm). The velocity must be constant with time, to guarantee the presence of steady flow during
calibration.

     4.1.1.4  Two entry ports, one each for the standard and Type S pitot tubes, shall be cut in the test
section; the standard pitot entry port shall be located slightly downstream of the Type S port, so that the
standard and Type S  impact openings will lie in the same plane during calibration.  To facilitate alignment
of the pitot tubes during calibration, it is advisable that the test section be constructed of plexiglas or some
other transparent material.

     4.1.2 Calibration procedure.  It is recommended that an identification number be assigned to the
pitot tube, and that this number be permanently marked or engraved on the body of the tube; also, one
leg of the tube should be marked "A", and the other, "B". To obtain calibration data for both the "A"
and "B"  sides, proceed as follows:

     4.1.2.1  Clean  and fill the manometer. Inspect and leak check all pitot lines and fittings; repair or
replace if necessary.

     4.1.2.2  Level and zero the manometer. Turn on the fan and allow the flow to stabilize.  Seal with
tape the Type S entry port.

     4.1.2.3  Using  the standard type pitot tube, locate an area where there is little or no velocity variation
over a 5 cm (2 in.) square segment.

     4.1.2.4  Ensure that the manometer is level and zeroed. Then position the standard pitot tube within
the area determined in section 4.1.2.3 and align it so that its  tip is pointing directly into the flow; particular
care should be taken  to avoid "yaw" and "pitch" angles.  Make sure that the entry port surrounding the tube
is properly  sealed.

     4.1.2.5  Read Apst(j and record its value in a data table similar to that shown in Figure 2-4. Remove
the standard pitot tube from the duct; disconnect it from the manometer.

     4.1.2.6  Seal the standard port and open the Type S port. Connect the Type S tube to the manometer.
Check the manometer level and zero. Insert and align the Type S pitot tube so that the "A" side impact
opening is at the  same measuring point (within the area determined in section 4.1.2.3) as was the standard
                                                 43

-------
         PITOTTUBE IDENTIFICATION NUMBER:.

         CALIDRATED BY:	
DATE:

RUM WO.
1
2
3
"A" SIDE CALIBRATION
APstd
cm H20
(in. H20)




AP(s)
cm H20
(in. H20)



AVERAGE
CU(S)





DEV.





RUN WO.
1
2
3
"B"SIDE CALIBRATION"
APstd
cm H20
(in.H20)




AP(s)
cm H20
(in. H20)



AVERAGE
CP(S)





DEV.




            DEV. =   Cp(S)-Cp(S)(avg.)  (MUST BE  <  0.01 )
            Cp  DIFFERENCE:  Aavg - Bavg = _ (MUST BE  <  0.01)

                          FIGURE 24  PITOT TUBE CALIBRATION DATA

tube and is pointing directly into the flow. Make sure that the entry port surrounding the tube is
properly sealed.
     4.1.2.7  Read Aps and record its value in the data table. Remove the Type S tube from the duct
and disconnect the manometer.

     4.1.2.8  Repeat steps 4.1.2.4 through 4.1.2.7 above until three sets of velocity head measurements
yield Cp(S) values (calculated from Equation 2-2) that agree to within 0.01 of their average.

     4.1.2.9  Repeat steps 4.1.2.4 through 4.1.2.8 for the "B" side.

     4.1.3 Calculations.

     4.1.3.1  For each data point, calculate the Type S pilot tube coefficient using the following formula:
                                     C
                                      P(S)
                      Equation 2-2
                                              44

-------
where:

      Cp(S)    = Type S pitot tube coefficient.

      Cp(std)  = Standard pitot tube coefficient; use 0.99 if the coefficient is unknown and the tube is
               designed according to the guidelines in section 2.7.

              = Velocity head measured by the standard pitot tube, cm H20 (in. H20).

              = Velocity head measured by the Type S pitot tube, cm H20 (in. H2O).

      4.1.3.2  Determine the average Cp for side "A" and likewise for side "B"; determine the difference
between these two average values.  Use the pitot tube only if the difference in Cp's is no more than 0.01.
Greater values indicate improperly constructed pitot tubes.

      4.1.3.3  Depending on direction in which the pitot tube  is faced, use corresponding average Cp for
velocity calculations.

      4.1.4 Frequency of Calibration and Maintenance.  Each pitot tube shall be calibrated before initial
use.  Thereafter, if the tube has been significantly damaged by field use, i.e., if impact faces are badly bent
out of shape, cut, nicked, or noticeably misaligned, the tube shall be repaired and recalibrated, or replaced.

      4.2   Temperature gauges. Calibrate dial and liquid filled bulb thermometers against mercury-in-glass
thermometers. New thermocouples need not be calibrated. Calibrate used thermocouples against new ones.
For other devices, check with the Administrator.

      4.3   Barometers. Calibrate against a mercury barometer.

5.    Calculations

      Carry out calculations, retaining at least one extra decimal figure beyond that of the acquired data.
Round off figures after final calculation.

      5.1   Nomenclature.

     A    = Cross sectional area of stack, m2 (ft2).

     BWS   = Water vapor in  the gas stream (from Method 5 or Reference Method 4), proportion by volume.

     Cp    = Pitot tube coefficient, dimensionless.

                                      m I (g/g-mole)(mm Hg)   I ' ' 2
     Kn    = Pitot tube constant, 34.97 — I —	  I       for the metric system and
       y                              seel  (°K)(mmH2O)  J


             85.48 —
                  sec
ft f(lb/lb-mole)(in. Hg)~j1'2
— [ —	 I       for the English system.
;ec     (°R)(in.H2O)  J
     Md   = Molecular weight of stack gas, dry basis (from Method 3 or other approved methods),
            g/g-mole (Ib/lb-mole).

     Ms    = Molecular weight of stack gas, wet basis, g/g-mole (Ib/lb-mole).

           = Md(\ -Bws)+ 18 Bws                                                     Equation 2-3
                                                45

-------
     •^bar  = Atmospheric pressure, mm Hg (in. Hg).

     Pg    = Stack static pressure, mm Hg (in. Hg).

     Ps    = Absolute stack gas pressure, mm Hg (in. Hg).

           = Pfcar + Pg                                                                Equation 2-4

     Pstd  = Standard absolute pressure, 760 mm Hg (29.92 in. Hg).

     Qsd  = Dry volumetric stack gas flow rate corrected to standard conditions, dscm/hr (dscf/hr).

     ts    = Stack temperature, °C (°F).

     Ts    = Absolute stack temperature, °K (°R).

           = 273 + ts for metric                                                       Equation 2-5

           = 460 + ts for English                                                      Equation 2-6

     rstd  = Standard absolute temperature, 293°K (528°R).

     vs    = Average stack gas velocity, m/sec (ft/sec).

     Ap   = Velocity head  of stack gas, mm H2 0 (in. H2 0).

     3600 = Conversion factor, sec/hr.

     1 8    = Molecular weight of water, g/g-mole (Ib/lb-mole).

     5.2   Average stack gas velocity.
                                  vs = Kp Cp (\/Ap)aVgW   s *Vg                       Equation 2-7
Note: Equation 2-7 assumes that TS,PS, andMs do not change appreciably (i.e. >1%) with cross section and
with time. If they do, consult with the Administrator to determine an acceptable procedure.
     5.3   Average stack gas dry volumetric flow rate.
                                = 3600(l-Bws)VsA
nq
H
Equation 2-8
                                                      ^s(avg)

6.   References

     6.1   Mark, L.S., Mechanical Engineer's Handbook, McGraw-Hill Book Co., Inc., New York, N.Y., 1951.

     6.2   Perry, J.H., Chemical Engineers' Handbook, McGraw-Hill Book Co., Inc., New York, N.Y., 1960.

     6.3   Shigehara, R.T., W.F. Todd, and W.S. Smith, Significance of Errors in Stack Sampling Measurements.
Paper presented at the Annual Meeting of the Air Pollution Control Association, St. Louis, Mo., June 14-19, 1970.

     6.4   Standard Method for Sampling Stacks for Particulate Matter. In: 1971 Book of ASTM Standards,
Part 23, Philadelphia, Pa., 1971. ASTM Designation D-2928-71.
                                                46

-------
6.5   Vennard, J.K., Elementary Fluid Mechanics, John Wiley & Sons, Inc., New York, N.Y., 1947.




6.6   ASME, Fluid Meters-Their Theory and Application, ASME, N.Y., 1959.
                                        47

-------
A2.  Method 3—Gas Analysis for Carbon Dioxide, Oxygen, Excess Air and Dry
      Molecular Weight

1.    Principle and Applicability

      1.1   Principle. An integrated or grab gas sample is extracted from a stack and analyzed for percent
carbon dioxide and percent oxygen  using an Orsat analyzer or, for molecular weight determinations, a
Fyrite type combustion gas analyzer.

      1.2   Applicability. This method should be applied only when specified by the test procedures for
determining compliance with the standards of performance for new stationary sources.

2.    Apparatus

      Any apparatus which has been demonstrated to yield results acceptable to the Administrator will be
considered acceptable for the purposes of this method.

      2.1   Grab sample (Figure 3-1).

                      PROBE
                                                  FLEXIBLE TUBING

                                                                       //--
                                                                             TO ANALYZER
             FILTER (GLASSWOOL)
                                    SQUEEZE BULB
                                FIGURE 3-1. GRAB-SAMPLING TRAIN

     2.1.1 Probe.  Stainless steel or borosilicate glass equipped with a filter (either in-stack or out-stack) to
remove particulate matter.

     2.1.2 Pump.  One-way squeeze bulb, or equivalent, to transport gas sample to analyzer.

     2.2   Integrated sample (Figure 3-2).

     2.2.1 Probe.  Stainless steel or borosilicate glass equipped with a filter (either in-stack or out-stack) to
remove particulate matter.

     2.2.2 Condenser. Air-cooled condenser, or equivalent, to remove excess moisture.

     2.2.3 Valve. Needle valve, to adjust sample gas flow rate.

     2.2.4 Pump.  Leak-free, diaphragm type, or equivalent, to transport sample gas to the flexible bag.
Install a small surge tank between the pump and rate meter to eliminate pulsation effect of diaphragm
pump on the rotameter.

     2.2.5 Rate meter. Rotameter, capable of measuring a flow range from 0 to 1.0 liter per minute.
 Mention of trade names or specific products does not constitute endorsement by the Environmental Protection Agency.
                                                48

-------
            PITOTTUBE
                                                              RATE METER
   ._!
1.9cm (0.75m.
                       PROBE
                                    |
              AIR-COOLED
              CONDENSER
 PROBE-PITOTTUBE
          FILTER
       (GLASS WOOL)
                                FIGURE 3-2. INTEGRATED GAS-SAMPLING TRAIN

            2.2.6 Flexible bag. Tedlar,1 or equivalent, with a capacity in the range of 55 to 90 liters. Before each
      field test make sure the bag is leak-free by checking it for leaks. To leak check, connect a water manometer
      and pressurize the bag to 5-10 cm H2 0 (2-4 in. H2O). Allow to stand for 10 minutes.  Any displacement in
      the water manometer indicates a leak. (Note: An alternative leak check method is to pressurize the bag to
      5-10 cm H2O or 2-4 in. H20 and allow to stand overnight. A deflated bag indicates a leak.)

            2.2.7 Pilot tube. Type S, or equivalent, attached to the probe to allow constant monitoring of the
      stack gas velocity so that the sampling flow rate can be regulated proportional to the stack gas velocity.
      The tips of the probe and pitot tube shall be adjacent to each other and the free space between them shall
      be 1.9 cm (0.75 m.). The pitot tube must also meet the criteria specified in Method 2 and be calibrated
      according  to the procedure  in the calibration section of that method.

            2.2.8 Differential  pressure gauge.  Inclined manometer capable of measuring velocity head to within
      10% of the minimum measured value. Below a differential pressure of 1.3 mm (0.05 in.) water gauge,
      micromanometers with sensitivities of 0.013 mm (0.0005 in.) should be used.  However, micromanometers
      are not easily adaptable  to field conditions and are not easy to  use with pulsating flow.  Thus, methods or
      other devices acceptable to  the Administrator may be used when conditions warrant.

            2.2.9 Manometer.  About 28 cm (12 in.) water-filled U-tube manometer, or equivalent, to be used
      for the flexible bag  leak  check.
                                                      49

-------
      2.2.10 Vacuum gauge. At least 760 mm Hg (30 in. Hg) gauge, to be used for the sampling train
leak check.

      2.3   Analysis

      2.3.1 Orsat analyzer or Fyrite  type combustion gas analyzer. The latter is used only for molecular
weight determination.

3.    Sampling Procedure

      3.1   Grab sampling.  This procedure is primarily used for, but not limited to, determining molecular
weight. Other uses must first be approved by the Administrator.

      3.1.1 The sampling point in the duct shall be at the centroid of the cross section or at a point no
closer to the walls than 1  m (3.28 ft), unless otherwise specified by the Administrator.

      3.1.2 Set up the equipment as shown in Figure 3-1, making sure all connections are tight and leak-free
by following the procedure in Section 4.

      3.1.3 Place the probe in the stack at the sampling point and then purge the sampling line. Draw a
sample into the analyzer and analyze according to Section 4.

      3.2   Integrated sampling (required when the analytical results will be used to calculate a pollutant
emission rate correction factor).

      3.2.1 Locate the sampling points according to Method 1, except that a minimum of 12 points shall
be used in all cases, unless otherwise specified in an applicable subpart.  For circular stacks of less than or
equal to 0.6 m (2 ft), a minimum of 8 points shall be used.

           Depending on the degree of mixing, more or less points than the above may  be used. As a
general rule, if the sampling location is located 8 diameters (equivalent diameters) downstream from
points of air in-leakages with good mixing (disturbances) in between, one point at the centroid or no
closer than 1 m (3.28 ft) from the stack walls may be used. If considerable variation as evidenced by a
concentration traverse or by review of the ductwork  design and points of air in-leakages, more points in
accordance with Method 1 shall be used. The number of diameters shall be calculated using the linear
distance from the point of air in-leakage and the diameter of the stack at the sampling location.

      3.2.2 Leak check the flexible bag as in Section 2.2.6. Set up the equipment as shown in Figure 3-2.
Just prior to sampling, leak check the train by placing a vacuum gauge at the probe inlet pulling a vacuum
of at least 250 mm Hg (10 in. Hg), plugging the outlet at the quick disconnect, and then turning off the
pump. The vacuum shall remain stable for at least two minutes.  Evacuate the flexible bag. Place the probe
in the stack and then purge the sampling line.  Now, connect the bag and make sure that all connections
are tight and leak free.

      3.2.3 Sample at a rate proportional (within 20% of constant proportionality, or as specified by the
Administrator) to the  stack velocity, traversing all sampling points. Record proportional sampling data
as shown in Figure 3-3. When analytical results will be  used to calculate a pollutant emission rate cor-
rection factor, the sampling MUST span the length of time the pollutant emission rate is being determined,
sampling at each traverse point for equal length of time. Collect at least 30 liters (1 ft3) of sample gas.

      3.2.4 Obtain and analyze at least one integrated flue gas sample during each pollutant emission
rate determination.
                                                  50

-------
TIME




TRAVERSE
PT.




Ap
mm(in.) H20




Q
1pm




AVERAGE
v£lT





% DEV.a





                         '%DEV =
                                        avg
(MUST BE < 20%)
                            FIGURE 3-3. PROPORTIONAL SAMPLING DATA
 4.    Analytical Procedure
      4.1   Leak check for Orsat analyzer.  Moving an Orsat analyzer frequently causes it to leak. There-
 fore, an Orsat analyzer should be thoroughly leak-checked on-site before the flue gas sample is introduced
 into it. The suggested procedure for leak-checking an Orsat analyzer is:

      4.1.1  Bring the liquid level in each pipette up to the reference mark on the capillary tubing and then
 close the pipette stopcock.

      4.1.2  Raise the leveling bulb sufficiently to bring the confining liquid meniscus onto the graduated
 portion of the burette and then close the manifold stopcock.

      4.1.3  Record the meniscus position,

      4.1.4  Observe the meniscus in the burette and the liquid level in the pipette for movement over the
 next  four minutes.

      4.1.5  For the Orsat  analyzer to pass the leak-check, two conditions must be met:

      4.1.5.1  The liquid level in each pipette must not fall below the bottom of the capillary tubing
 during this four-minute interval.

      4.1.5.2  The meniscus in the burette must not change by more than 0.2 ml during this four-minute
 interval. For the results to be valid the Orsat analyzer must pass this leak test before and after the analysis.

      4.1.6  If the analyzer fails the leak-check procedure, all rubber connections and stopcocks should be
 checked until the cause of the leak is identified.  Leaking stopcocks must be disassembled, cleaned and
 regreased. Leaking rubber connections must be replaced. After the analyzer is reassembled, the leak-check
 procedure must be repeated.

      4.2   Determination of stack gas molecular weight. (Orsat leak check described above is optional).
 Within eight hours after the sample is taken, analyze it for percent carbon dioxide and percent oxygen
using  either  an Orsat analyzer or a Fyrite1  type combustion gas analyzer. Determine the percent of the gas
that is nitrogen and carbon monoxide by subtracting the sum of the percent carbon dioxide and percent
oxygen  from 100 percent.

      4.2.1  Grab samples.  Repeat the sampling and analysis until the molecular weight from each of three con-
secutive grab samples differs from their mean by no  more than 0.3 grams/gram mole (0.3 pounds/pound mole).
                                                  51

-------
     4.2.2 Integrated samples.  Repeat the analysis until the molecular weight for three consecutive analyses
differs from their mean by no more than 0.3 grams/gram mole (0.3 pounds/pound mole).

     4.3   Determination of 02, C02, or excess air for calculating pollutant emission rate correction factors.
Note: The Fyrite type combustion gas analyzers are not acceptable for this purpose, unless otherwise shown
to the satisfaction of the Administrator. The results may also be used for determining stack gas molecular
weight.

     4.3.1 Leak check the Orsat analyzer as described in section 4.1. This procedure is mandatory.

     4.3.2 Within four hours after the integrated sample is taken, analyze it for percent carbon dioxide and
percent oxygen using an Orsat analyzer. To ensure complete absorption of these gases make repeated passes
through the absorbing solution until two consecutive readings are the same.  Several passes (3-4) should be
made between readings. (If constant readings cannot be obtained after three consecutive readings, replace
the absorbing solution.) Determine the percent of the gas that is nitrogen and carbon monoxide by subtracting
the sum of the percent carbon dioxide and percent oxygen from 100 percent.

           This procedure assumes that carbon monoxide  concentration is negligible. If appreciable quan-
tities are expected, consult with  the Administrator.

     4.3.3 Repeat the analysis on the integrated sample until each of three consecutive analyses for percent
carbon dioxide and percent oxygen differ by no more  than 0.3 percent by volume when carbon dioxide i.s
greater than 3% and 0.2 percent  by volume when carbon dioxide is less than or equal to 3%.

5.   Calibration

     5.1   Calibrate the pitot tube as specified in Method  2 and the rotameter against a wet test meter.

6.   Calculations

     6.1   Nomenclature

     Mtf   = Dry molecular weight (gram/gram mole).

     %EA = Percent excess air.

     '/rCO2= Peicent carbon dioxide  by volume (dry basis).

     %O2  = Peicent oxygen by volume (dry basis).

     %N   - Percent nitrogen by volume (dry basis).

     0.264= Ratio  of oxygen to nitrogen in air, v/v.

     0.28 = Molecular weight of both nitrogen and CO divided by 100.

     0. ~*>1 = Molecular weight of oxygen  divided by 100.

     0.44 = Molecular weight of carbon dioxide divided by  100.

     6.2   Excess air.  Use equation 3-1 to calculate the percent excess air using the three consecutive
analyses that meet the requirements of section 4.3.3.  Then calculate the average percent excess air.

                                              %O2 (100)
                                     %EA =	                           Equation 3-1
                                            0.264  %N2 - %O2
                                                  52

-------
Note:  The equation above assumes that carbon monoxide concentration is negligible. If" appreciable carbon
monoxide concentrations are expected, consult with the Administrator.

      6.3   Dry molecular weight. Use equation 3-2 to calculate the dry molecular weights using data ob-
tained from sections 4.2.1, 4.2.2, or 4.3.2 and 4.3.3, average the results and report to the nearest 0.1 g/g-mole
(0.1 Ib/lb-mole).

                        Md = 0.44(%CO2) + 0.32(%02) + 0.28(%N2 + %CO)             Equation 3-2

      6.4   Carbon dioxide concentration calculation. Using the three consecutive carbon dioxide analyses
that meet the requirements of section 4.3.3, calculate the average carbon dioxide concentration.

7.    References

      7.1   Altshuller, A. P.  Storage of Gases and Vapors in Plastic Bags, International Journal of Air  and
Water Pollution, 6,75-81(1963).

      7.2   Connor, William D. and J. S. Nader, Air Sampling with Plastic Bags, Journal of the American
Industrial Hygiene Association, 25, 291-297 (1964).

      7.3   "Burrell Manual for Gas Analysts," Seventh Edition (1951), Available from Burrell Corporation,
2223 Fifth  Avenue, Pittsburgh, Pennsylvania  15219.
                                                53

-------
A3.  Method 5—Determination of Particulate Emissions from Stationary Sources

1.    Principle and Applicability

      1.1   Principle. Particulate matter is withdrawn isokinetically from the source and collected on glass
fiber filter maintained at temperatures equal to or less than 120 ± 14°C (248 ± 25°F) or such other
temperature as specified by an applicable subpart of the standards. The particulate mass is determined
gravimetrically after  removal of uncombined water.

      1.2   Applicability. This method is applicable for the determination of particulate emissions from
stationary sources only when specified by the test procedures for determining compliance with new source
performance standards.

2.    Apparatus

      2.1   Sampling train. A  schematic of the sampling train used in this method is shown in Figure 5-1.
      t.9 TO 2.5 cm
      (0.75 TO 1 in.)  TEMPERATURE SENSOR
                                                      IMPINGER TRAIN OPTIONAL, MAY BE REPLACED
                                                           BY AN EQUIVALENT CONDENSER
 1.9 cm (0.75 in.
                                                                        THERMOMETER
                                        HEATED AREA   THERMOMETER
                         PITOTMANOMETER       JMPINGERS              ICE BATH
                                                          BY PASS VALVE
                                  ORIFICE
             REVERSE TYPE
               PITOTTUBE
                        THERMOMETERS
                                                                  MAIN VALVE
                                              V
                                    DRY GAS METER
   ^
AIRTIGHT
  PUMP
                                                                                       CHECK
                                                                                       VALVE
                                                                                        VACUUM
                                                                                         LINE
                            FIGURE 5-1. PARTICULATE-SAMPLING TRAIN.

Commercial models of this train are available.  However, if one desires to build his own, complete con-
struction details are described in APTD-0581; for changes from the APTD-0581 document and for allowable
modifications to Figure 5-1, see the following subsections.

          The operating and maintenance procedures for the sampling train are  described in APTD-0576.
Since correct usage is important in obtaining valid results, all users should read the APTD-0576 document
and adopt the operating and maintenance procedures outlined in it, unless otherwise specified herein.

     2.1.1 Probe nozzle-Stainless steel (316) with sharp, tapered leading edge.  The angle of taper shall be
< 30° and the taper shall be on the outside to preserve a constant internal diameter. The probe nozzle shall
                                                54

-------
be of the button-hook or elbow design, unless otherwise specified by the Administrator.  The wall thickness of
the nozzle shall be less than or equal to that of a 20 gauge tubing, i.e., 0.165 cm (0.065 in.) and the distance
from the tip of the nozzle to the first bend or point of disturbance shall be at least two times the outside nozzle
diameter. The nozzle shall be constructed from seamless stainless steel tubing.  Other configurations and con-
struction material may be used with approval from the Administrator.

           A range of sizes suitable for isokinetic sampling should be available, e.g., 0.32 cm (1/8 in.) up to
1.27 cm (1/2 in.) (or larger if higher volume sampling trains are used) inside diameter (ID) nozzles in increments
of 0.16 cm (1/16 in.). Each nozzle shall  be calibrated according to the procedures outlined in the calibration
section.

      2.1.2 Probe liner-Borosilicate or quartz glass tubing with a heating system capable of maintaining a gas
temperature at the exit end during sampling of no greater than 120 ± 14°C (248 + 25°F) or no greater than
such other temperature as specified by an applicable subpart of the standards.  Since the actual temperature
at the outlet of the probe is not monitored during sampling, probes constructed according to APTD-0581 and
utilizing the calibration curves of APTD-0576 or calibrated according to the procedure outlined in APTD-0576
will be considered as acceptable.

           Borosilicate or quartz glass probe liners shall be used for temperatures up to about 480°C (900°F)
and quartz liners for temperatures up to  about 900°C (1650°F).  Both may be  used at higher temperatures
for short periods of time, but must be approved by the Administrator. The softening temperature for
borosilicate is 820°C (1508°F) and for quartz it is 1500°C (2732°F).

           When length limitations, i.e.  greater than about 2.5 m (8.2 ft), are encountered at temperatures
less than 320°C (608°F), stainless steel (316) or Incoloy 825  (both of seamless tubing), or other materials
as approved by  the Administrator, may be used.  Metal probes for sampling gas streams at temperatures in
excess of 320°C (608°F)  must be approved by the Administrator.

      2.1.3 Pitot tube-Type S, or other  device approved by the Administrator, attached to probe to allow
constant monitoring of the stack gas velocity. The face openings of the pitot tube and the probe nozzle shall
be adjacent and parallel to each other, not necessarily on the same plane, during sampling. The free space
between the nozzle and pitot tube shall be at least 1.9 cm (0.75 in.).  The free  space shall be  set based on a
1.3 cm (0.5 in.) ID nozzle. If the sampling train is designed for sampling at higher flow rates than that
described in APTD-0581, thus necessitating the use of larger sized nozzles, the largest sized nozzle shall be
used to  set the free space.

           The pitot tube must also meet the criteria specified in Method 2 and calibrated according to the
procedure in the calibration section of that method.

      2.1.4 Differential pressure gauge-Inclined manometer capable of measuring velocity head to within
10% of the minimum measured value. Below a differential pressure of 1.3 mm  (0.05 in.) water gauge,
micromanometers with sensitivities of 0.013 mm (0.0005 in.) should be used.  However, micromanometers
are not easily adaptable to field conditions and are not easy to use with pulsating flow. Thus, methods or
other devices acceptable to the Administrator may be used when conditions warrant.

      2.1.5 Filter holder—Borosilicate glass with a glass frit filter support and a silicone rubber gasket. Other
materials of construction  may be used with approval from the Administrator, e.g. if probe liner is stainless
steel, then filter holder may be stainless steel. The holder design shall provide a positive seal against leakage
from the outside or around the filter.

      2.1.6 Filter heating system-Any heating system capable of maintaining a temperature around the
filter holder during sampling of no greater than 120 ±  14°C (248  ± 25°F), or such other temperature as
 Mention of trade names or specific products does not constitute endorsement by the Environmental Protection Agency.
                                                 55

-------
specified by an applicable subpart of the standards. A temperature gauge capable of measuring temperature to
within 3°C (5.4°F) shall be installed such that temperature around the filter holder can be regulated and
monitored during sampling. Heating systems other than shown in APTD-0581 may be used.

      2.1.7 Condenser -Any system that cools the sample gas stream and allows measurement of the water
condensed and moisture leaving the condenser, each to within 1 ml or 1 g. Acceptable means are to measure
the condensed water either gravimetrically or volumetrically and to measure the moisture leaving the con-
denser by (1) monitoring the temperature and pressure at the exit of the condenser and using Dalton's law or
(2) by passing the sample gas stream through a tared silica gel trap with exit gases kept below 20°C (68°F)
and determining the weight gam.

           Note: If "condensible particulate matter" is desired, in addition to moisture content, the following
system shall be used-four impingers connected in series with ground glass, leak free fittings or any similarly
leak free noncontaminating fittings. The first, third, and fourth impingers shall be of the Greenburg-Smith
design, modified  by replacing the tip with a 1.3 cm (1/2 in.) ID glass  tube extending to about 1.3 cm (1/2 in.)
from the bottom of the flask.  The second impinger shall be of the Greenburg-Smith design with the standard
tip. Individual States or control agencies requiring this information shall be contacted as to the sample
recovery and analysis of the impinger contents.

           For purposes of writing the procedure of this method, the system described in the note above
will be used for determining the moisture content of the  stack gas. Modifications (e.g. using flexible con-
nections between the impingers or using materials other than glass) may be used with approval from the
Administrator.

           If means other than silica gel are used to determine the amount of moisture leaving the condenser,
it is recommended that silica gel still be used between the condenser system and pump to prevent moisture
condensation in the pump and metering devices.

           Unless otherwise specified by the Administrator, flexible vacuum lines may be used to connect
the filter holder to the condenser.

      2.1.8 Metering system—Vacuum gauge, leak-free pump, thermometers capable of measuring temperature
to within 3°C (5.4°F), dry gas meter with 2% accuracy, and related equipment, or equivalent, as required to
maintain an isokinetic sampling rate and to determine sample volume. Sampling trains utilizing metering
systems designed for higher flow rates than that described in APTD-0581 or APTD-0576 may be used provided
that the specifications in section 2 of this method are met. When the metering system is  used in conjunction
with a pitot tube, the system shall enable checks of isokinetic rates.

      2.1.9 Barometer—Mercury, aneroid, or other barometers capable of measuring atmospheric pressure to
within 2.5 mm Hg (0.1 in. Hg). In many cases, the barometric reading may be obtained from a nearby weather
bureau station, in which case the station value shall be requested and an adjustment for elevation differences
shall be applied at a rate of minus 2.5 mm Hg (0.1 in. Hg) per 30 m (100 ft) elevation increase

      2.1.10 Gas density determination equipment—Temperature and pressure gauges and gas analyzer as
described in Methods 2 and 3.

      2.1.11 Temperature and  pressure gauges—If Dalton's law is used, to monitor temperature and pressure
at condenser outlet. The temperature gauge shall have an accuracy of 1°C (2°F). The pressure gauge shall
be capable of measuring pressure to within 2.5 mm Hg (0.1 in. Hg).  If silica gel is used in the condenser
system the temperature and pressure must be measured before the silica gel component.

      2.2  Sample recovery.
                                                 56

-------
      2.2.1 Probe liner and probe nozzle brushes-Nylon bristles with stainless steel wire handles. The probe
 brush shall have extensions, at least as long as the probe, of stainless steel, nylon, teflon, or similarly inert
 material.  Both brushes shall be properly sized and shaped to brush out the probe liner and nozzle.

      2.2.2 Glass wash bottles—Two.

      2.2.3 Glass sample storage containers—Chemically resistant, borosilicate narrow mouth glass bottles, for
 acetone washes, 500 ml or 1,000 ml. Screw cap closures shall be teflon rubber-backed liners or of such con-
 struction so as to be leak free and prevent chemical attack from the acetone. Other types of containers must
 be approved by the Administrator.

      2.2.4 Petri dishes—For filter samples, glass or plastic, unless otherwise specified by the Administrator.

      2.2.5 Graduated cylinder and/or balance—To measure condensed water to within 1 ml or 1 g. Graduated
 cylinders shall have subdivisions no greater than 2 ml. Most laboratory balances are capable of weighing to the
 nearest 0.5 g or less. Any of these balances are suitable for use here and in section 2.3.4.

      2.2.6 Plastic storage containers—Air tight containers to store silica gel.

      2.2.7 Funnel and rubber policeman—To aid in transfer of silica gel to container; not necessary if silica
 gel is weighed in the field.

      2.3   Analysis.

      2.3.1 Glass weighing dishes.

      2.3.2 Desiccator.

      2.3.3 Analytical balance-To measure to within 0.1 mg.

      2.3.4 Balance—To measure to within 0.5 g.

      2.3.5 Beakers-250 ml.

      2.3.6 Hygrometer—To measure the relative humidity of the  laboratory environment.

      2.3.7 Temperature gauge—To measure the temperature of the laboratory environment.

 3.    Reagents

      3.1   Sampling

      3.1.1 Filters-Glass fiber filters, without organic binder exhibiting at least 99.95% efficiency (< 0.05%
penetration) on 0.3 micron dioctyl phthalate smoke particles. The filter efficiency test shall be conducted
in accordance with ASTM standard method D 2986-71.  Test data from the supplier's quality control program
is sufficient for this purpose.

      3.1.2 Silica gel-Indicating type, 6-16 mesh. If previously used, dry at  175°C (350°F) for 2 hours.  New
silica gel may be used as received.

      3.1.3 Water-When analysis of the material caught in the impingers is required, distilled water shall be
used. Run blanks prior to field use to eliminate a high blank on test samples.

      3.1.4 Crushed ice.
                                                  57

-------
      3.1.5 Stopcock grease—Acetone insoluble, heat stable silicone grease. This is not necessary if screw-on
connectors with teflon sleeves, or similar, are used.

      3,2   Sample recovery.

      3.2.1 Acetone—Reagent grade, < 0.001% residue, in glass bottles. Acetone from metal containers
generally has a high residue blank and should not be used. Sometimes, suppliers transfer acetone to glass
bottles from metal containers.  Thus, acetone blanks shall be run prior to field use and only acetone with low
blank values (< 0.001%) shall be used.

      3.3   Analysis.

      3.3.1 Acetone—Same as 3.2.1.

      3.3.2 Desiccant-Anhydrous calcium sulfate, indicating type.

4.    Procedure

      4.1   Sampling.  The sampling shall be conducted by competent personnel experienced with  this test
procedure.

      4.1.1 Pretest preparation. All the components shall be maintained and calibrated according to the pro-
cedure described in APTD-0576, unless otherwise specified herein.

           Weigh approximately 2CO-300 g of silica gel in air tight containers to the nearest 0.5 g.  Record
the total weight, both  silica gel and container, on the container. More silica gel may be used but care should
be taken during sampling that it is not entrained and carried out from the impinger.  As an alternative, the
silica gel may be weighed directly in the impinger or its sampling holder just prior to  the train assembly.

           Check filters visually against light for irregularities and flaws or pinhole leaks. Label a filter of
proper diameter on the back side near the edge using numbering machine ink. As an alternative, label the
shipping container (glass or plastic petri dishes) and keep  the filter in this container at all times except during
sampling and weighing.

           Desiccate the filters at 20 ± 5.6°C (68 ± 10°F) and ambient pressure for at least 24 hours and
weigh at 6 or more hour intervals to a constant weight, i.e., < 0.5 mg change from previous weighing, and
record results to the nearest 0.1 mg. During each weighing the filter must not be exposed to the laboratory
atmosphere for a period greater than 2 minutes and a relative humidity above 50%.

      4.1.2 Preliminary determinations.  Select the sampling site and the minimum number of sampling points
according to Method 1 or as specified by the Administrator.  Determine the stack pressure, temperature, and
the range of velocity heads using Method 2 and moisture  content using Approximation Method 4 or its
alternatives for the purpose of making isokmetic sampling rate calculations. Estimates may be used.  However,
final results will be based on actual  measurements made during the test.

           Select a nozzle size based  on the range of velocity heads such that it is not necessary to change the
nozzle size in order to maintain isokinetic sampling rates. During the run,  do not change the nozzle size.
Ensure that the differential pressure gauge is capable of measuring the minimum velocity head value to within
10%, or as specified by the Administrator.

      Select a suitable probe liner and probe length such  that all traverse points can be sampled. Consider
sampling from opposite sides for large stacks to reduce the length of probes.
                                                 58

-------
           Select a total sampling time greater than or equal to the minimum total sampling time specified in
 the test procedures for the specific industry such that the sampling time per point is not less than 2 min. or
 some greater time interval as specified by the Administrator and the sample volume that will be taken will ex-
 ceed the required minimum total gas sample volume specified in the test procedures for the specific industry.
 The latter is based on an approximate average sampling rate.  Note also that the minimum total sample volume
 is corrected to standard conditions.

           It is recommended that 1/2 or an integral number of minutes be sampled at each point in order to
 avoid timekeeping errors.

           In some circumstances, e.g. batch cycles, it may be necessary to sample for shorter times at the
 traverse points and to obtain smaller gas sample volumes. In these cases, the Administrator's approval must
 first be obtained

      4.1.3 Preparation of collection train. During preparation and assembly of the sampling  train, keep all
 openings where contamination can occur covered until just prior to assembly or until sampling is about to
 begin.

           Place 100 ml of water in each of the first two impingers, leave the third impinger empty, and place
 approximately 200-300 g or more, if necessary, of preweighed  silica gel in the fourth impinger.  Record the
 weight of the silica gel and container to the nearest 0.5 g. Place the container in a clean place  for later use  in
 the sample recovery.

           Using a tweezer or clean disposable surgical gloves, place the labeled (identified) and weighed filter
 in the filter holder.  Be sure that the filter is properly centered  and the gasket properly placed so as to not
 allow the sample gas stream to circumvent the filter. Check filter for  tears after assembly is completed.

           When glass liners are used, install selected nozzle using a Viton A O-ring when stack temperatures
 are less than 260°C (500°F) or an asbestos  string gasket when temperatures are higher. The Viton A O-ring
 and asbestos string gasket are installed as a seal where the nozzle is connected to a glass liner.  See APTD-0576
 for details. When metal liners are used, install the nozzle as above or by a  leak free direct mechanical
 connection. Mark probe with heat restraint tape or  by some other method to denote the proper distance
 into the stack or duct for each sampling point.

           Unless otherwise specified by the Administrator, attach a temperature probe to the metal sheath
 of the sampling probe so that  the sensor extends beyond the probe tip and does not touch any metal. Its
 position  should be about 1.9 to 2.54 cm (0.75 to 1 in.) from the pilot tube and probe nozzle to avoid
 interference with the gas flow.

           Set up the train as in Figure 5-1, using, if necessary, a very light coat of sihcone grease on all ground
 glass joints, greasing only the outer portion (see APTD-0576) to avoid possibility of contamination by the
 sihcone grease.  With approval from the Administrator, a glass cyclone may be used between the probe and
 filter holder.

           Place crushed ice around the impingers.

      4.1.4 Leak check procedure—After the sampling train has been  assembled, turn on and set the filter and
 probe heating system to the power required to reach a temperature of 120 ±  14°C (248 + 25°F) or such other
 temperature as specified by an applicable subpart of the standards for the leak check.  (If water condensation
 is not a problem the probe and/or filter heating system need not be used.) Allow time for the temperature to
 stabilize. If a Viton A O-ring or other leak free connection is used in assembling the probe nozzle to the probe
liner, leak check the train at the sampling site by plugging the nozzle and pulling a 380 mm Hg (15 in. Hg)
vacuum.  (Note:  A lower vacuum may be used provided that it is not exceeded during the test.) If an asbestos
 string is used, do not connect the probe to the train  during the  leak check.  Instead, leak check the train as
                                                 59

-------
above by first plugging the inlet to the filter holder.  Then connect the probe to the train and leak check at
about 25 mm Hg (1 in. Hg) vacuum.  A leakage rate in excess of 4% of the average sampling rate of 0.00057
m3/min. (0.02 cfm), whichever is less, is unacceptable in either case.

           The following leak check instructions for the sampling train described in APTD-0576 and
APTD-0581 may be helpful.  Start the pump with by-pass valve fully open and coarse adjust valve completely
closed.  Partially open the coarse adjust valve and slowly close the by-pass valve until 380 mm Hg (15 in. Hg)
vacuum is reached. Do not reverse direction of by-pass valve.  This will cause water to back up into the filter
holder.  If 380 mm Hg (15 in. Hg) is exceeded, either leak check at this higher vacuum or end the leak check
as shown below and start over.

           When  the leak check is completed, first slowly remove the plug from the inlet to the probe or filter
holder and immediately turn off the vacuum pump.  This prevents the water in the impingers from being
forced backward into the filter holder and silica gel from being entrained  backward into the third impinger.

           Leak checks shall be conducted as described whenever the train is disengaged, e.g. for silica gel
or filter changes during the test, prior to each  test run, and at the completion of each test  run. If leaks are
found to be in excess of the acceptable rate, the test will be considered invalid. To reduce lost time due to
leakage  occurrences, it is recommended that leak checks be conducted between port changes.

     4.1.5 Particulate tram operation—During the sampling run, isokinetic sampling rate to within  10%, or as
specified by the Administrator, of true isokinetic and the temperature around the filter of no greater than
120 ± 14°C (248 ± 25°F),  , or as specified by  an applicable subpart of the standards, shall be maintained.

           For each run, record the data required on the example data sheet shown in Figure 5-2. Be sure
to record the initial dry gas meter reading.  Record the dry gas meter readings at the beginning and end of
each sampling  time increment, when changes in  flow rates are made, and when sampling is halted. Take  other
data point readings at least once at each sample  point during each time increment and additional readings
when significant changes (20% variation in velocity head readings) necessitate  additional adjustments in flow
rate. Level and zero the manometer.

           Clean  the portholes prior to the test run to minimize chance of sampling the deposited material.
To begin sampling, remove the nozzle cap, verify that the filter and probe are  up to temperature, and that the
pilot tube and probe are properly positioned.  Position the nozzle at the first traverse point with the tip
pointing directly into the gas stream. Immediately start the pump and adjust  the flow to isokinetic conditions.
Nomographs are available for sampling trains using type S pitot tubes with 0.85 ± 0.02 coefficient and when
sampling in air or  a stack gas with equivalent density (molecular weight equal  to 29 ± 4), which aid in the
rapid adjustment of the isokinetic sampling rate without excessive computations.  APTD-0576 details the
procedure for using these nomographs. If Cp  andAf^ are outside the above stated ranges,  do not use the
nomograph unless appropriate steps are taken to compensate for the deviations.

           When  the stack is under significant negative stack pressure (height of impinger stem), take care
to close the coarse adjust valve before inserting the probe into the stack to avoid water backing into the
filter holder.  If necessary, the pump may be turned on with the coarse adjust valve closed.

           When  the probe is in position, block off the openings around  the probe and porthole to prevent
unrepresentative dilution of the gas stream.

           Traverse the stack cross section, as required by Method 1 or as specified by the Administrator,
being careful not to bump  the probe nozzle into the stack walls when sampling near the walls or when
removing or inserting the probe through the portholes to minimize chance of extracting deposited material.

           During the test run, make periodic adjustments to keep the temperature around the filter holder
at the proper temperature and add more ice and, if necessary, salt to maintain a temperature of less than
                                                 60

-------
 CC
 r>
 <
 cc
 CD


 4.
                      o   K
                      z   a     I     1

                      i   a   i   I
CO
LJJ
CC
a.

CJ

cc
o
CC
-
                     uj  uj   CJ   *—
                     h-  t-   <   O
                     UJ  UJ   ^*-   ^
uj 2£ cc
cc O uj
^coOccW-
<«?S?°
CC ^ > T» Q- —
UJ u. < If 2 U
i O 2 Q ~o
2 i —r h—
i- O <
u _}
cc
UJ UJ
35 _
° ^ ^
i 2 °_
CC UJ (j
SI o
—1 UJ
u7 >-
UJ
CC
»~ cc
< t;
cc *~
£ uj
1
-1
UJ &
it cc
S 3
^5
t/j
<
u
*~ rr
_J 0
h- —
O U
o

u_
h- 0
UJ —
z o
0
UJ
CL LU —
S S ^
< D -^
(/) _j
_ Q m
3 > £
O
^ ^ (/3 UJ
-S ^ 00 CJ ^
-J ^ (— 1 LLJ
y, 0= * ^ ^
" S U cc y
a: u. < o *
0. u.
Q
O O
Psi CN
r r
1 1

;- 0
(- >N
5 § -3.S
3| §
LLJ
CC
ID
<_><
°_
>—
o
UJ
(J CC „. —
- 3 I °>
S"1 z
»_ UJ * —
o^ cc t r.
CL
'O
2
-J
cu
<
C/)



c
1 c
s
Z
0 ce
CL LLJ
vi S
11 D
a; i
<
cc
»—





























































































































































































































































































































ci
>

d>
5
















-j
c
^






en
<















UJ
O
<
UJ
>
<
                                                                                                                                                                          <
                                                                                                                                                                          Q
                                                                                                                                                                          Q
                                                                                                                                                                          _
                                                                                                                                                                          O

                                                                                                                                                                          H
                                                                                    61

-------
20°C (68°F) at the condenser/silica gel outlet to avoid excessive moisture losses.  Also, periodically check the
level and zero of the manometer.

           If the pressure drop across the filter becomes too high making isokinetic sampling difficult to
maintain, the filter may be replaced in the midst of a sample run.  It is recommended that another complete
filter assembly be used rather than attempting to change the filter itself. After the new filter or filter
assembly is installed conduct a leak check.  The particulate weight shall include the summation of all filter
assembly catches.

           A single train shall  be used for the entire sample run, except for filter and silica gel changes. However,
if approved by the Administrator, two or more  trains may be used for a single test run when there are two or
more ducts or sampling ports.  The results shall be the total of all sampling train catches.

           At the end of the sample run, turn off the pump, remove the probe and nozzle from the stack,
and record the final dry gas meter reading.  Perform a leak check at a vacuum equal to or greater than the
maximum reached during sampling. Calculate percent isokinetic (see calculation section) to determine
whether another test run should be made. If there is difficulty in maintaining isokinetic rates due to source
conditions, consult with the Administrator for possible variance on the isokinetic rates.

     4.2   Sample recovery. Proper cleanup procedure begins as soon as the probe is removed from the stack
at the end of the sampling period.

           When the probe can be safely handled, wipe off all external particulate matter near the tip of the
probe nozzle and place a cap over it to prevent losing or gaining particulate matter. Do not cap off the probe
tip tightly while the sampling train is cooling down as this would create a  vacuum in the filter holder, thus
drawing water from the impingers into the fiber.

           Before moving the  sample train to the cleanup site, remove the probe from the sample train, wipe
off the silicone  grease, and cap the open outlet of the probe. Be careful not to lose any condensate, if present.
Wipe off the silicone grease from the filter inlet where the probe was fastened and cap it. Remove the
umbilical cord from the last impinger and cap the impinger.  If a flexible line is used between the first impinger
or condenser and the filter holder, disconnect the line at the filter holder  and let any condensed water or
liquid drain into the impingers  or condenser. After wiping off the silicone grease, cap off the filter holder
outlet and impinger inlet.  Either ground glass stoppers or plastic caps or serum caps may be used to close
these openings

           Transfer the probe and filter-impinger assembly to the cleanup area. This area should be clean
and protected from the wind so that the chances of contaminating or losing the sample will be minimized.

           Save a portion of the acetone used for cleanup as a blank. Place about 200 ml of this acetone in
a glass sample container labeled "acetone blank."

           Inspect the train prior to and during disassembly and note any abnormal conditions.  Treat the
samples as follows:

           Container No. 1. Carefully remove the filter from the filter holder and place in its identified
petri dish container.  Use a pair of tweezers and/or clean disposable surgical gloves to handle the filter. If
it is necessary to fold the filter, do so such that the particulate cake is inside the fold. Quantitatively remove
any particulate  matter and/or filter which adheres to the filter holder by carefully using a dry nylon bristle
brush and/or a sharp-edged blade and place into this container.  Seal the container.

           Container No. 2. Taking care to see that dust on the outside  of the probe or other exterior
surfaces does not get into the sample, quantitatively recover particulate matter or any condensate from the
                                                  62

-------
probe noz/de, probe fitting, probe liner, and front half of the filter holder by washing these components with
acetone and placing the wash into a glass container in the following manner.

           Distilled water may be used when approved by the Administrator or shall be used when specified
by the Administrator.  In these cases, save  a water blank and follow Administrator's directions on analysis.

           Carefully remove the probe nozzle and clean the inside surface  by rinsing with acetone from a
wash bottle and brushing with a nylon bristle  brush. Brush until acetone rinse shows no visible particles,
after which make a final rinse of the inside surface with acetone.

           Brush and rinse with acetone the inside parts of the Swagelok fitting in a similar way until no
visible particles remain.

           Rinse the probe liner with acetone by tilting the probe and squirting acetone into its upper end,
while rotating the probe so that all inside surfaces will be rinsed with acetone.  Let the acetone drain from
the lower end into the  sample container. A funnel may be used to aid in transferring liquid washes to the
container. Follow the  acetone rinse with a probe brush. Hold the probe in an inclined position, squirt
acetone into  the upper end as the probe brush is being pushed with a twisting action through the probe, hold
a sample container underneath the lower end  of the probe, and catch any acetone and particulate matter
which is brushed from  the probe. Run the brush through the probe three times or more until no visible
particulate matter is carried out with the acetone or remains in the probe liner on visual inspection. With
stainless steel or other metal probes, run the brush through in the above prescribed manner at least six times
since metal probes have small crevices in which particulate matter can be entrapped.  Rinse the brush with
acetone and quantitatively collect these washings in the sample container.  After the brushing make a final
acetone rinse of the probe as described above.

           It is recommended that two people be used to clean the probe  to minimize losing the sample.
Between sampling runs, keep brushes clean and protected from contamination.

           After ensuring that all joints are wiped clean of silicone grease,  clean the inside of the front half
of the  filter holder by rubbing the surfaces with a nylon bristle brush and rinsing with acetone. Rinse each
surface three times or more if needed to remove visible particulate. Make a final rinse of the brush and filter
holder. After all acetone washings and particulate matter are collected in the sample container, tighten the
lid on the sample container so that acetone will not leak out when it is shipped to the laboratory. Mark the
height of  the fluid level to determine whether or not leakage occurred during transport  Label container to
clearly identify its contents.

           Container No. 3.  Note color of indicating silica gel to  determine if it has been completely spent
and make a notation of its condition.  Transfer the silica gel from the fourth impinger to  the original container
and seal.  A funnel may make it easier to pour the silica gel without spilling. A rubber policeman may be used
as an aid in removing the  silica gel from the impinger.  It is not necessary to remove  the small amount of dust
particles that may adhere to the walls and are  difficult to remove   Since  the gain in weight is to be used for
moisture calculations, do not  use any water or other liquids to transfer the  silica gel.  If a balance is available
in the field, follow the  procedure under analysis

           Impinger water  Treat the impingers or condenser as follows: Make a notation of  any color or
film  in the liquid catch. Measure the liquid which is in the first three impingers to within ± 1 ml by using a
graduated cylinder or, if available, to within ± 0.5 g by using a balance.  Record the volume or  weight of
liquid present.  This information is required to calculate the moisture content of the  effluent gas.

           If analysis of  the impinger catch is not required, discard the liquid after measuring  and recording
the volume or weight.  If  analysis of the impinger catch is required, leave the impingers intact to transfer the
liquid, cap off the inlet, and pour the liquid through the outlet into the graduated cylinder or into a sample
container  after its weight  has been determined.
                                                  63

-------
            If a different type of condenser is used, measure the amount of moisture condensed either
 volumetrically  or gravimetrically.

      4.3   Analysis.  Record the data required on the example sheet shown in Figure 5-3.  Handle each sample
 container as follows:
                     Plant.
                     Date_
                     Run No
                     Relative Humidity.
                    Amount liquid lost during transport  	
                     Acetone blank volume, ml	
                     Acetone wash volume, ml	
                     Acetone blank concentration, mg/mg (equation 5-4).
                     Acetone wash blank, mg (equation 5-5)	
CONTAINER
NUMBER
1
2
TOTAL
WEIGHT OF PARTICULATE COLLECTED.
mg
FINAL WEIGHT


]]^>
-------
           Container No. 2. Note level of liquid in container and confirm on analysis sheet whether or not
leakage occurred during transport. Measure the liquid in this container either volumetrically to ± 1 ml or
gravnnetrically to ± 0.5 g. Transfer the contents to a tared 250 ml beaker, and evaporate to dryness at ambient
temperature and pressure. Desiccate for 24 hours and weigh to a constant weight.  Report the results to the
nearest 0.1 mg.

           Container No. 3. Weigh the spent silica gel to the nearest 0.5 g using a balance. This step may be
conducted in the field.

           "Acetone Blank" Container.  Measure acetone in this container either volumetrically or gravimetrically.
Transfer the acetone  to a tared 250 ml beaker and evaporate to dryness at ambient temperature and pressure.
Desiccate  for 24 hours and weigh to a constant weight. Report the results to the nearest 0.1 mg.

5.    Calibration

      Maintain a laboratory log of all calibrations.

      5.1   Probe nozzle. Using a micrometer, measure the inside diameter of the nozzle to the nearest
0.025 mm (0.001 in.).  Make 3 separate measurements using different diameters each time and obtain the
average of the measurements. The difference between the high and low numbers shall not exceed 0.1 mm
(0.004 in.).

              When  nozzles become nicked,  dented, or corroded, they shall be reshaped, sharpened, and
recalibrated before use.

           Each nozzle shall be permanently and uniquely identified.

      5.2   Pitot tube.  The pitot tube shall be calibrated according to the procedure outlined in Method 2.

      5.3   Dry gas meter and orifice meter.  Both meters shall be calibrated according to the  procedure outlined
in APTD-0576. When diaphragm pumps with by-pass valves are used, check for proper metering system design
by calibrating the dry gas meter at an additional flow rate of 0.0057 m3/min. (0.2 cfm) with the by-pass valve
fully opened and then with it fully closed. If there is more than ± 2% difference in flow rates  when compared to
the fully closed position of the by-pass valve,  the system is not designed properly and must be corrected.

      5.4   Probe heater calibration. The probe heating  system shall be calibrated according to the procedure
contained in APTD-0576. Probes constructed according to APTD-0581 need not be calibrated if the calibration
curves in APTD-0576 are used.

      5.5   Temperature gauges.  Calibrate dial and liquid filled bulb thermometers against mercury-in-glass
thermometers. New thermocouples need not be calibrated.  Calibrate used thermocouples against new ones.
For other  devices, check with the Administrator.

6.    Calculations

      Carry out calculations, retaining at least one extra decimal figure beyond that of the acquired data.  Round
off figures after final calculation.

      6.1    Nomenclature

           An   = Cross sectional area of nozzle, m2 (ft2)

           BWS  ~ Water vapor in the gas stream, proportion by volume
                                                  65

-------
Ca    = Acetone blank residue concentration, mg/mg

cs    = Concentration of particulate matter in stack gas, dry basis, corrected to standard conditions,
        g/dscm (g/dscf)

/     = Percent of isokinetic sampling

mn   = Total amount of particulate matter collected, mg.

Mw   = Molecular weight of water, 18 g/g-mole (18 Ib/lb-mole)

ma   = Mass of residue of acetone after evaporation, mg

/"bar  = Barometric pressure at the sampling site, mm Hg (in. Hg)

Ps    = Absolute stack gas pressure, mm Hg (in. Hg)

Pstd   = Standard absolute pressure,  760 mm Hg (29.92 in. Hg)

R    = Ideal gas constant, 0.06236  mm Hg-m3/°K-g-mole (21.83 in. Hg-ft3/°R-lb-mole)

Tm   = Absolute average dry gas meter temperature (see Figure 5-2), °K (°R)

Ts    - Absolute average stack gas temperature (see Figure 5-2), °K  (°R)

rstd   = Standard absolute temperature, 293°K (528°R)

Va    = Volume of acetone blank, ml

Vaw  ~ Volume of acetone used in wash, ml

YIC   - Total volume of liquid collected in impingers and silica gel (see Figure 5-3, ml).

Vm   = Volume of gas sample as measured by dry gas meter, dcm (dcf)

      ~ Volume of gas sample measured by the dry gas meter corrected to standard conditions, dscm (dscf).

      = Volume of water vapor in the gas sample corrected to standard conditions, scm (scf).

vs    = Stack gas velocity, calculated by Method 2, Equation 2-7 using data obtained from Method 5,
        m/sec (ft/sec)

Wa   = Weight of residue in acetone wash, mg

A//   = Average pressure differential across the orifice (see Figure 5-2), meter, mm H20 (in. H20)

pa    = Density of acetone, mg/ml (see label on bottle)

pw   = Density of water, 1 g/ml (0.00220 Ib/ml)

9     = Total sampling time, min.

13.6  = Specific gravity of mercury
                                            66

-------
      60   -  sec/nun




      100  =  Conversion to percent




      6.2   Average dry gas meter temperature and average orifice pressure drop. See data sheet (Figure 5-2).




      6.3   Dry gas volume.  Correct the sample volume measured by the dry gas meter to standard conditions


(20°C, 760mm Hg or 68°F, 29.92 in. Hg) by using Equation 5-1.
                           - V
                                                 A//



                                                13.6
               = K Vn
where:
      A' = 0.3855 °K/mm Hg for metric units




        = 17.65 °R/in. Hg for English units




      6.4   Volume of water vapor.
  V
- Vlc\
                                                         -K- V
                                                        ]~K Vlc
where
      K = 0.001 34 m3/ml for metric units




        = 0.0472 ft3 /ml for English units




      6.5   Moisture content
      6.6   Acetone blank concentration.
                                          ^m(std) + ^w(std)
                                                 mn
     6.7   Acetone wash blank.
                                                VaPa
                                          a ~ Ca Vaw pa

Equation 5-1
Equation 5-2
                                                                                     Equation 5-3
                                                                                     Equation 5-4
                                              Equation 5-5
     6.8   Total particulate weight.  Determine the total particulate catch from the sum of the weights


obtained from containers 1 and 2 less the acetone blank (see Figure 5-3).
     6.9   Particulate concentration.
                                 cs = (0.001 g/mg)(mn/Fm(std))
                                              Equation 5-6
                                                 67

-------
     6.10  Conversion factors:

                          From               To               Multiply by
scf
g/ft3
g/ft3
g/ft3
m3
gr/ft3
lb/ft3
g/m3
0.0283
15.4
2.205 X 10^:
35.34
     6.11  Isokinetic variation.

     6.11.1 Calculations from raw data.
                            100 Ts [K Vlc + (Vm/Tm)(Pbar + Aff/13.6)]
                         / = 	                Equation 5-7
                                          6Q8vsPsAn

where:

     A' = 0.00346 mm Hg-m3/ml-°K for metric units

       = 0.00267 in. Hg-ft3/ml-°R for English units

     6.11.2 Calculations from intermediate values.

                                        T, Fm^tH^trt 100
                                     TstdvseAnPs60(l-Bws)

                                           TS ^m(std)
                                                                                    Equation 5-8
where:

     K = 4.323 for metric units

       = 0.0944 for English units

     6.12  Acceptable results.  If 90% 
-------
      60   =  sec/mm


      100  =  Conversion to percent


      6.2   Average dry gas meter temperature and average orifice pressure drop.  See data sheet (Figure 5-2).



      6.3   Dry gas volume.  Correct the sample volume measured by the dry gas meter to standard conditions

(20°C, 760mmHgor68°F, 29.92 in. Hg) by using Equation 5-1.
                           ~ V
                           - vm\ —— ,
                                                 AH
where.
      A' = 0.3855 °K/mm Hg for metric units


        = 17.65 °R/in. Hg for English units


      6.4   Volume of water vapor.
                                       — TV  /
                                    d) - Vic  —
                                            MWI\ /'std
where
     A' = 0.00134 m3/ml for metric units


        = 0.0472 ft3 /ml for English units


     6.5   Moisture content.
                                                ^w(std)
                                          ^m(std) + ^w(std)
     6 6   Acetone blank concentration.
                                                 mn
                                                VaPa
     6.7   Acetone wash blank.
                                                                      A///13.
                                                                     T
>]
Equation 5-1
Equation 5-2
                                                                                     Equation 5-3
                                                                                     Equation 5-4
                                                                                     Equation 5-5
     6.8   Total particulate weight.  Determine the total particulate catch from the sum of the weights

obtained from containers 1 and 2 less the acetone blank (see Figure 5-3).
     6.9   Particulate concentration.
                                   = (0.001 g/mg)(m«/Fm(std))
Equation 5-6
                                                 67

-------
     6.10  Conversion factors:





6.11
6.11.


From
scf
g/ft3
g/ft3
g/ft3
Isokinetic variation.
, 1 Calculations from raw data.
100 Ts [K V

To
m3
gr/ft3
lb/ft3
g/m3


'lc+(VmlTm)(Pm-
609 vsPsAn
Multiply by
0.0283
15.4
2.205 X 10"3
35.34


fAtf/13.6)]

where:
     K = 0.00346 mm Hg-m3/ml-°K for metric units

       = 0.00267 in. Hg-ft3/ml-°R for English units

     6.11.2 Calculations from intermediate values.
                                         
-------
 B.1  Statistical Model for Particulate Concentration Data

      The preliminary test plan for the collaborative test included a detailed discussion of the plan for the
 statistical analysis of the data in order to insure that the components of interest could be validly estimated.
 The sampling plan called for 5 runs each week with 4 paired trains per run.  One laboratory team (console
 operator and two technicians) would operate a paired train alone, while 6 laboratories would independently
 operate a single tram of the paired tram system. To insure that the results of the paired-train laboratory
 were indicative of the results one could expect of other laboratories, three independent teams were used as
 the paired-train laboratory, one for each week of the test. The remaining laboratories were to be paired, and
 the pairings would remain constant during the course of the test.

      The analytical method to be used was an analysis of variance according to the model

                                        YJJ = fj. + p,  + yf +  e,j
 where

      Y,j — determined concentration from laboratory / in run i
      ju   — overall mean
      Pi  — effect due to run /

      jf  — effect due to lab /

 and

      e,/ — random error associated with  YJJ.

      Laboratories and runs are assumed to be random effects. The model is an additive or no-interaction
 model.  The term for the interaction between runs and laboratories is then used as the error term in the
 ANOVA.

      The data are investigated in two forms.  First the results  of each week's runs are submitted to the
 ANOVA procedure.  There  are 8 trains for each of these analyses, and a proposed 5 runs, for a total of 40
 concentration determinations in each week. The single-week data are principally the  source of the within-
 laboratory or sampling error term. There are two potential sources for this  term.

      The usual estimate of sampling error comes from the run by train interaction term. In addition, a
 preliminary  test of the paired particulate sampling train using a single laboratory team demonstrated  that the
 two  concentrations  thus obtained could be considered replicates.  An error variance of

                                           k    2
                                     s2 =  y^   y   (Y- - Y  Y
                                         I = 1  / = 1
where
                  + r/2)

      k  — number of runs

can be calculated if the paired-train laboratory can be shown to be obtaining replicate samples.  This estimate
has k df associated with it, one from each sampling run.

      If the laboratory term is significant, the laboratory SS can be partitioned into SS for various hypotheses
of interest. This is done by the use of orthogonal contrasts among the means, as described in Snedecor and
Cochran.'  ' A contrast is of the form
                                               73

-------
                                            I =  1
where
     Xj - zth sample mean

     Q —constant
and
   X>,.=0.
Two contrasts,  Y^  QJ, and
              / = 1           i=l
                                    , are said to be orthogonal if
                                                Qd«=0.
                                           / = 1
If the contrasts are chosen to be orthogonal, then they can be used to partition the SS for a particular term.
The hypothesis to be tested by a contrast is

                                        H0:  £  C,X, = 0,

and the SS due to the hypothesis are calculated as
 with 1 df for each contrast.

      For this study, contrasts are included which satisfy the questions of interest.  If the values for the
 paired-train laboratory are to be used as replicates, then it must be shown that they are estimating the same
 mean level. The differences between the paired laboratories on a single train are to be used to estimate the
 laboratory bias variance, and a contrast is used for each of these pairs. The differences among pairs are also
 investigated, so that a contrast for this term is included.  The contrasts used for each analysis are summarized
 in Table Bl.
     TABLE Bl. CONTRASTS AMONG LABORATORY MEANS
Contrasts
Laboratory Mean
101A
101B 104
105
106 1C
)7 108
109
Week 1
1
2
1
1
Contrasts
-1 0
1 1
0
1
0
-1
0 0
1 -1
0
-1
Laboratory Mean
104
105

106 | 107 |
108 | 109
3 Week
1
2
3
4
1
0
0
2
0
0
2
0
1
0
-1
0
-1
0
-1
0
0
1
-1
0
0
-1
-1
                                                                     For the single week's analysis, the
                                                               first contrast tests for the equality of
                                                               means between the two results from
                                                               the paired-laboratory train. The second
                                                               is an investigation of any differences
                                                               among pairs of laboratories.  The con-
                                                               trasts are shown in Table B1  for week 1,
                                                               and are identical for the second and
                                                               third week, with the substitution of the
                                                               other laboratories running the paired-
                                                               laboratory train.  For the three  week
                                                               analysis, contrasts 1, 2 and 3 are tests
                                                               for equality between paired laboratories,
                                                               while the fourth is another test  for
                                                               differences among pairs.
                                               74

-------
     Along with the contrasts among the means, the error SS can be partitioned in a similar manner to pro-
vide a separate error term for each contrast.  This is done to insure that a proper divisor is used in the F-tests,
for the case where the error variance may not be homogeneous for all terms. The error term for each con-
trast is obtained by applying the contrasts separately to each run, then obtaining a sums of squares from
these run contrasts.  These subdivisions of the error term each have (r - 1) df associated with them, where r
is the number of runs.

     The results of these analyses are summarized in ANOVA tables, from which variance components are
estimated and significance is determined for  the factors of interest. These are shown in subsequent appendices
for both the particulate concentration from Method 5, and the corrected concentration, as pertains to
incinerators.

B.2  Precision Estimation and Tests of Hypotheses for Method 5

     The particulate concentrations shown in Table  1 are submitted to  statistical analyses according to the
model in Appendix B. 1.  The three weeks of results are analyzed separately first, using the concentration from
all 8 trains. The ANOVA tables generated from these analyses are shown in Table B2.

                TABLE  B2. ANOVA FOR METHOD 5 CONCENTRATION DETERMINATIONS,
                                        INDIVIDUAL WEEKS
Week
1
Mean = 121 83






2
Mean = 123 19


3
Mean = 148.64






Source
Runs
Labs
1 vs 2
(l-4)vs(5-8)
Error


Total
Runs
Labs
Error
Total
Runs
Labs
1 vs 2
(1-4) vs (5-8)
Error


Total
df
2
7
1
1
14
2
2
23
4
7
28
39
4
7
1
1
27
4
4
38
SS
4,610.25
4,068.19
123.40
593.79
2,867.98
16.47
1,530.85
11,546.42
15,511.33
2,193.13
4,419.17
22,123.63
26,964.76
6,398.31
414.74
2.373 14
9,236.55
684.45
4,259.95
42,599.62
MS
2,305.13
581.17
123.40
593.79
204.86
8.24
715.43

3,877.83
313.30
157.85

6,741.19
914.04
414.74
2,373.14
342.08
171.11
1,064.99

F
11.25*
2.84*
14.98f

-------
      In each week's data, the error term has an EMS of a2, the within-laboratory variance.  Testing these by
Bartlett's test gives a significance level of 0.13, so the hypothesis of equality is accepted. These may also be
pooled, then, to give

                                            a2p = 239.48

with 69 df. The estimated standard deviations for the two cases are 13.81 and  15.48, respectively, which
indicates that either could serve as a valid estimate of a2.

      The second hypothesis tested is one dealing with the pairs of trains. The  four paired trains are split
into two groups, and if a difference among pairs exists, it should be reflected in this contrast. However, in
both the first and the third week's data, there was no detectable difference among the pairs.

      In Table B3, the ANOVA for the 6 laboratories through 13 runs is summarized. This analysis is used
to estimate the laboratory bias term for the method. The laboratory term for the analysis is shown to be
significant, which indicates that not all trains have the same true mean. Contrasts among the laboratory
means are then used to determine significance and estimate the  variance components.

                TABLE B3.  ANOVA FOR METHOD 5 CONCENTRATION DETERMINATIONS,
                                         3-WEEK ANALYSIS
Source
Runs
Labs
4vs5
6vs 7
8vs9
(4,5) vs (6,7) vs(8,9)
Error
4vs5
6 vs 7
8vs9
(4,5)vs(6,7)vs(8,9)
Total
df
12
5
1
1
1
1
59
12
12
12
12
76
SS
36,593.44
4,436.18
2,117.72
184.66
858.13
1,175.46
11,158.60
600.86
8.20
341.86
3,087.34
52,188.22
MS
3,049.55
88724
2,117.72
184.66
858.13
1,175.46
189.13
50.07
0.68
28.49
257.28

F
16.40*
4.77*
42.30*
271.56*
30 12*
4 57*






EMS
_
a2 + 13 CT?




a1





*Significant at 5% level.
     The three contrasts related to the laboratory bias term are those comparing laboratories operating on
the same paired train. In each case, F-ratios calculated using the error terms calculated for the contrast show
that significant differences do exist.  These SS are then pooled to give laboratory bias SS of SS/ = 3160.5 1,
with 3  df. The laboratory bias MS, then, isMSL = 1053.50, and has expected value of a2 + 13 a/.  Sub-
stituting the estimated sampling error term gives
                                          = 1053.50- 190.81
                                                    13
                                          = 66.36,
with 3 df.
                                              76

-------
      The pairs term is significant in this analysis, but the SS due to pairs is 1175.46, which is close to the SS
for labs.  The indication is that this is simply another estimate of the laboratory bias component.  In addition,
if spatial/temporal changes in the gas flow are the cause of the differences, the error term for the contrast
should be large, but this is not the case. The error MS for pairs is 257.28.  Using an F-ratio to test for equality,
a test statistic of Fc = 1.35 is calculated. The critical value from a table of the F-distribution with 12 and
13 dfisF(12,  13)= 1.47, using a significance level of 25%.  Since Fc does not exceed  1.47, no difference can
be shown even at that level.

B.3  Rank Test for Equality of Laboratories

      In order to  test whether the separate laboratories obtain equivalent results, the concentrations are sub-
mitted to analysis using a nonparametric analysis of variance technique.  The purpose  of this test is to deter-
mine whether there is an ordering of laboratories independent of concentration and, if so, to what it can be
attributed.

      The test chosen is the Friedman test, a two-way procedure which allows for a blocking factor, in  this
case runs. The data are ranked across runs, and then summed for each laboratory. These rank sums,/?,,,
form the basis for the test. If all laboratories are equivalent, the RJJ should be approximately equal. The test
statistic is
                                           k
                                  12     ^
                           Xr =	  X, RU  ~ 3N(k +  1)
                               Nk(k  + 1) /T!   '

where

R,j  — rank sum for laboratory j

N   — number of runs

k   — number of labs

This value is then compared to a table of the chi-square distribution with k — 1 df to determine significance

      The test is applied separately to each week's data, using the six results from the  labs running a single
train, plus the average of the  two results from the dual train. The value of y$ along with its df and significance
level is shown in Table B4 for these data. In addition, the test is applied to the data from the six laboratories
that completed all 13 runs, and these results included in the table
     TABLE B4. SUMMARY OF
         FRIEDMAN TEST
                                         The results from the separate weeks are inconclusive, with one
                                    insignificant, one marginally significant and one significant result.
                                    The 6-laboratory test, however, shows that the ordering is significant
                                    at a level  of 0.001, showing a strong tendency for the labs to be
                                    ordered relative to one another, irrespective of position or pairing.
                                    The rank  sums which are most nearly equal are the pairs Lab 107
                                    and Lab 109, and Lab  106 and Lab 108, which do not represent
                                    the pairing of the laboratories with respect to the trains. The ap-
                                    parent conclusion, then, is that this phenomenon is unrelated to
                                    the stack  flow, but rather reflects differences in such things as
                                    sample recovery techniques and equipment operation.

B.4 Weighted Coefficient of Variation Estimates

     The technique used for obtaining estimates of the coefficients of variation of interest is to use a  linear
combination of the individual beta values obtained. The linear combination used will be of the form
Test
Week 1
Week 2
Week 3
6 Labs
Xr2
7.71
11 21
13.82
21.04
df
6
6
6
5
Significance
Level
0.26
0.08
0.03
0001
                                             77

-------
                                               k / =
or

                                              i JL
where /?/ is thejth coefficient of variation estimate, k is the total number of estimates, and oiy is a weight
applied to thejth estimate.  The weighted values are used if the coefficient of variation estimates are based
on unequal sample sizes.

     As previously discussed, the individual estimate of j3 is obtained as
for a sample of size n. The factor an is defined as^   '
                                         «„ = ./—
where F represents the standard gamma function.  Thus for k equal-size samples, the best estimator is simply
the arithmetic mean of the individual beta values/1 ^'. However, if the sample sizes are unequal, weighting
is more desirable in that it provides for more contribution from those values derived from larger samples.
There is more variability in the beta values obtained from the smaller samples, as can be seen by inspecting
the variance of the estimator. We have that
                                         Var(j3) = Var
                                                         x
                                                          -
                                               _  2


for normally distributed samples/ ' and true coefficient of variation, (3.  Rewriting this expression, we have


                                                         2/32)


and all terms are constant except for o£  and n.  Thus, the magnitude of the variance changes with respect
to the factor a^/n.  Now, since an decreases as n increases, the factor a^/n must decrease as n increases, and
the variance is reduced.

     The weights, to/, are determined according to the technique used in weighted least squares analysis1-8-*,
which gives a minimum variance estimate of the parameter. The  individual weight, w,-, is computed as the
inverse of the variance of the estimate, ft, and then standardized. Weights are said to be standardized when
                                              78

-------
To standardize, the weights are divided by the average of the inverse variances for all the estimates. Thus, we


can write
where
                                            u, -
and
                                        u=- y  —r-
                                            fr/Ti  VarOJ/)




      Now, from the above expressions we can determine u,, u andco, for the beta estimates. For any

estimate, |3,,



                                                  1
for sample size n,, and
                                                Var(ft)
                                               j32 (1 +2|32)
                                                      1
                                                   Var(j3;)




                                                    k
                                               =-;
                                                 fc=i a«2(1 + 2'32)
Thus, the ;th weight, co,-, is
                                             79

-------
                                       |/(l+2/32)
                                    1  f  "L
                                    f   i
The estimated coefficient of variation is
                                   A:.
 EKrijC


~T~
                                              —

                                              a2.
                                         /= 1  "/
                                               '
                                                  ft
                                          a
                                    / =  1   "/

B.5  Precision Estimation for Velocity Determination




      In a previous report, it was shown that the precision of the determination of velocity varies according


to the actual velocity, and that a valid model is of the form,
and
                                             80

-------
 for the within- and between-laboratory components, respectively. Thus, to provide estimates of the standard
 deviations of interest, the coefficients of variation, j3 and ft,, are estimated and the standard deviations expressed
 as percentages of an unknown mean value.

      The technique for obtaining estimated coefficients of variation is discussed in Appendix B.4. The between-
 laboratory coefficient of variation, )3/,, is estimated as
                                              1
                                                  13
                                           = ___ V1      ii
                                             13 ^    " 3c-
                                             i-> / = i     •*;
 where
and
 Xf — the sample mean of the ;tn run, across laboratories

 s, — sample standard deviation of the /th run



 a,, - finite correction factor used to correct for the bias in s.
 Since all the estimates are based upon 6 determinations, no weighting procedure is used, and an. = an = 1.0509.
 The means and standard deviations for the runs are shown in Table B5.  These values are substituted into the
                                          above equation to obtain
     TABLE B5. RUN DATA SUMMARY
                 (m/sec)
Run
1
2
7
8
9
3
4
5
6
10
11
12
13
Mean
14.39
15 06
1462
14.65
14 80
15.45
15.64
15 53
14.94
15.19
15.26
15.58
15 39
Standard
Deviation
0.52
0.80
055
046
0.50
073
0.53
0.87
0.73
0.59
058
063
0 70
Coefficient of
Variation
00534
0.0375
0.0358
00340
00316
0.0473
0.0561
0.0491
0.0387
0.0382
00456
00340
0.0408
                                                                    j3 = 0.043.

                                          There are 5 df associated with this estimate from the 6 laboratories.
                                          The estimated between-laboratory standard deviation, then, is
                                                                        = (0.043)
                                          or 4.3% of the mean velocity.
                                               The within-laboratory coefficient of variation, (3, is estimated
                                          from the values obtained from each collaborator in a given block.
                                          There are 12 of these collaborator-block combinations, which
                                          gives
                                                                    1
                                                                       12
                                                                               sl
                                                                               x.
where
and
Sj  — standard deviation for collaborator-block/

X] — mean of collaborator-block/ results



cj, — weight applied to the/* estimate.
                                                 81

-------
The weighting procedure is used since the two blocks are of unequal size, and is discussed in Appendix B.4.
The weights incorporate the finite correction factor for the bias in s, and give more weight to the values
obtained from the larger blocks.

     The means and standard deviations are shown in Table B6, for the  12 collaborator-block combinations.
    TABLE B6. COLLABORATOR-BLOCK
             DATA SUMMARY
                 (m/sec)
Collaborator
Mean
Standard
Deviation
Coefficient of
Variation
Block 1
Lab 104
Lab 105
Lab 106
Lab 107
Lab 108
Lab 109
13.97
14 30
14.79
14.92
15.55
14 70
028
0.17
0.27
0.29
0.48
0.37
2.02
1.22
1.83
1.91
3.07
2.53
Block 2
Lab 104
Lab 105
Lab 106
Lab 107
Lab 108
Lab 109
1447
14.88
15.60
15.74
16.26
15.28
0.28
0.37
0.29
0.34
0.24
0.40
1.93
2.46
1 87
2.19
1.48
2.62
Substituting into the above equation gives an estimated
coefficient of variation of

                       |3 = 0.022

with 60 df. The within-laboratory standard deviation
is estimated as
                                                                       = (0.022) ju

                                                 or 2.2% of the mean value.

                                                      The laboratory bias coefficient of variation is
                                                 estimated from the above as
                                                                  = V(0.043)2 - (0.022)2

                                                                  = 0.037.
The laboratory bias standard deviation, then, is
                                            = (0.037)Ai

or 3.7% of the mean value. There are 5 df associated with this estimate from the between-laboratory term.

B.6 Precision Estimation for Method 3

     a.    Model

           The Orsat data and dry gas molecular weight from the 9 laboratories that participated in the test
are submitted to an ANOVA each according to the same model.  The three weeks are treated separately
according to the model
where

      Yt] — determination by collaborator ,- in runy

     7j  — effect due to laboratory/
                                                82

-------
 and
         — random error associated with
 All effects are assumed to be random. The first week there were 3 runs by 6 laboratories and in the next two
 weeks there were 5 runs by 7 laboratories.  In the second week, one laboratory did not make a determination,
 and a value was substituted according to the formula given in Snedecor and Cochran/   '  This is compensated
 for by reducing by 1  the degrees of freedom for error. The model assumes runs to be replicate determinations,
 since  a preliminary test showed the runs factor insignificant in all cases.

      b.    CO2

           The determinations of the CO2 level made by the collaborators are shown in Table 9 in the body of
 the report. Performing the analysis of variance according to the model in a. above, three ANOVA tables are
 generated, and these are shown in Table B7.
 TABLE B7. ANOVA FOR CO, DETERMINATION
Sou ice
df
SS
MS
F
EMS
Week I
Labs
lirror
Total
5
12
17
0 97
1.45
2 43
0 19
0 12

1 58f

a2 + 3a?
a2

Week 2
Labs
Erroi
Total
6
27
33
4.21
0 65
4.86
0.70
002

35.00*

a2 + 5a[
a2

Week 3
Labs
Error
Total
6
28
34
2.96
1 41
4 37
0.49
005

9 80*

a2 + Sa2,

fNot significant.
*Sigmficant at 5% level.
           The error mean square  has an expected value
(EMS) of a2 , the true within-laboratory variance.  The esti-
mated within-laboratory variance, then, is obtained by
pooling the sums of squares and df for error into a single
estimate. The estimated within-laboratory variance is
                                                                      ""
                                                                    _
                                                                        pooled
                                                                      ^/pooled

                                                                      [1.45 + 0.65+ 1.41]
                                                                         12 + 27 + 28
                                                                    _2.51
                                                                      67

                                                                    = 0.04
                                              with 67 df. The within-laboratory standard deviation is esti-
                                              mated to be
                                                                 a = \/a
                                                                   - 0.20 percent CO2  by volume
           The laboratory term is significant for weeks 2 and 3. For each, the expected value of the mean
square for laboratories is a2 + 5ai, which gives an estimated value of

                                          -2   MSL - 
-------
                                               0.68
                                              = 0.14

and

                                              0.49 - 0.05
                                              0.44
                                            = 0.09,

each with 6 df from the 7 laboratories. The combined estimate is taken to be the average of the two values,
since each is based on the same number of df. This gives
                                          = - (0.14 + 0.09)
                                         = 0.12

with 6 df. The estimated laboratory bias standard deviation is
                                              = V/0~l2

                                              = 0.35 percent of CO2.

           The between-laboratory variance is estimated from the above to be

                                         °b = o2 + ol

                                            = 0.04 + 0.12

                                            = 0.16.

This gives an estimated between-laboratory standard deviation of
                                       = X/0~l6

                                       = 0.40 percent CO2.

There are 6 df associated with this estimate, due to the laboratory bias term.
                                                 84

-------
      c.
02
           The oxygen determinations from Table 10 are submitted to the ANOVA procedure outlined in
 section a. above.  The resultant tables are shown in Table B8.
   TABLE B8. ANOVA FOR O2 DETERMINATION
Source
df
SS
MS
F
EMS
Week 1
Labs
Error
Total
5
12
17
3.19
1 45
4 64
0 64
0.12

5 33*

a2 + 3o2L
CT2

Week 2
Labs
Error
Total
6
27
33
8 52
1.43
995
1.42
0.06

23.67*

a2 + 5o2L
a2

Week 3
Labs
Error
Total
6
28
34
12.12
3.53
15.65
2.02
0.13

15.54*


a2 4- 5(7?
CT2

*Significant at 5% level.
                                              The expected mean square of the error term is
                                         a2, the within-laboratory variance.  Pooling the three
                                         estimates obtained gives,

                                                    -2 _   pooled

                                                         ^/pooled

                                                       _[1.45 + 1.43 + 3.53]
                                                             12T27 + 28

                                                       = 6.41/67

                                                       = 0.10

                                         with 67 df. This gives an estimated within-laboratory
                                         standard deviation of
                                                                        = 0.32.

           The laboratory term in the ANOVA is significant in all three sets of data. The EMS of the lab
term is a2 + kaj , where h = 3, 5, 5, respectively, for the three weeks.  The three estimates obtained are

                                           OL, =0.14

                                           61, =0.27

and

                                           ff!3 =0.38

with 5, 6, and 6 df, respectively. Pooling these estimates gives

                                      5(0.14)+ 6(0.27)+ 6(0.38)
                                 ol =	

                                    = 0.27.

There are 6 df associated with this estimate.  The estimated laboratory bias standard deviation, then is
                                               = 0.52.

                                                 85

-------
           Combining the above estimates gives an estimated between-laboratory variance ot

                                        al = d2 + 61

                                           = 0.10 +0.27

                                           = 0.37

with 6 df. The between-laboratory standard deviation is estimated as
                                        = V0.37~
                                        = 0.61 percent O2
     d.    Md
           The dry gas molecular weights calculated by the collaborators are shown in Table 1 1 for the
three weeks of the test. Using the model in section a. above, the ANOVA tables are generated and the
ANOVA are shown in Table B9.
    TABLE B9. ANOVA FOR Md DETERMINATION
Source
df
SS
MS
F
EMS
Week I
Labs
Error
Total
5
12
17
0.0150
00291
0.0441
0.0030
0 0024

1.25f

a2 + 3o2L
a2

Week 2
Labs
Error
Total
6
27
33
0.0493
0.0136
0.0629
0.0082
0.0005

16.40*


a2 + 5a2
a2

Week 3
Labs
Error
Total
6
28
34
00209
0.0189
0.0398
0.0035
0.0007

5.00*

a2 + 5a2L

tlnsignificant.
*Significant at 5% level.
           The EMS of the error term is a2 ,the
within-laboratory variance.  The mean squares for
error from the three weeks are combined into a
single estimate by pooling the sums of squares and
df.  The result is

        -2 _   pooled
            ^/pooled

          _ (0.0291 + 0.0136 +  0.0189)
                    12 + 27 + 28

          _ 0.0825
             67

          = 0.0012

with 67 df. The estimated within-laboratory
standard deviation is
                                       = 0.035 gm/gm-mole.

           The laboratory term is significant for the second and third weeks of the test. The EMS for each
of these terms is a2 + 5 a!, and the two estimates of the laboratory bias variance are calculated as
                                               =0.0015

-------
and

                                          6?  =0.0006.
                                           ** 3

Since both of these estimates have 6 df, the combined estimate is merely the average of the two, or

                                     al =- (0.0015+ 0.0006)


                                        = 0.0011

with 6 df. The estimated laboratory bias standard deviation, then, is
                                         = Vo.oon

                                         = 0.033 gm/gm-mole.

           The between-laboratory term may be estimated from the above to be

                                     al = a2 + dl

                                        = 0.0012 + 0.0011

                                        = 0.0023

with 6 df from the laboratory bias term. The estimated between-laboratory standard deviation is
                                        = 0.048 gm/gm-mole

B.7  Precision Estimation for Moisture Fraction Determination

      The moisture fractions determined from the eight sampling trains are shown in Table  12. These are
submitted to an ANOVA procedure according to the model

                                      Y,, = M + Pi + T) + e,j

where

     M  — overall mean

     Pi  - effect due to run /

     T,  — effect due to tram/

and

     e,j — random error associated with Yy.

The model is a random effects model. The interaction between laboratories and runs is assumed to be zero,
and the interaction term, then, is used as the error term in the analysis. This assumption is warranted, since
the change in stack moisture content should have no effect on a laboratory's performance

                                              87

-------
      The ANOVA tables for each separate week are shown in Table BIO.  The EMS of the error term is
 a2, the within-laboratory variance. There are three estimates of a2, and these are pooled to give

     TABLE BIO. ANOVA FOR Bws DETERMINATION           „  /
Source
df
SS
MS
F
EMS
Week 1
Runs
Trams
Error
Total
2
7
14
23
0.001406
0.001832
0 000982
0.004220
0.000703
0.000262
0.000070
10.02*
3.73*
o2 + 3a2
a* L
Week 2
Runs
Trains
Error
Total
4
7
28
39
0.001233
0.000866
0.002417
0.004516
0.000308
0.000123
0.000086
3.57*
1.43f
CT2 + So]
a2 L
Week 3
Runs
Trains
Error
Total
4
7
28
39
0.004782
0.000240
0.001676
0.006698



*Significant at 5% level.
f Insignificant.
                                                              ^/pooled

                                                            _ (0.000982 + 0.002417 + 0.001676)
                                                                         14 + 28 + 28

                                                            _ 0.005075
                                                                 70

                                                            = 0.000073

                                                      with 70 df. The estimated within-laboratory
                                                      standard deviation is
                                                                     a=\/0.000073

                                                                      = 0.009.

                                                           The trains term is significant only for the
                                                      first week's data. The EMS of the trains com-
                                                      ponent is a2 + 301.  Solving for a£ gives

                                                                    .,   MSL - a2
                                                                       = 0.000064

with 7 df from the 8 trains.  This gives an estimated laboratory bias standard deviation of
                                              = 0.008.

      The between-laboratory variance is estimated from the above as
                                       = 0.000073 + 0.000064

                                       = 0.000137.

The between-laboratory standard deviation is estimated to be

                                           ah = -v/aI

                                              = 0.012.

There are 7 df associated with this estimate, taken from the trains term.

B.8   Precision  Estimation and Tests of Hypotheses for Particulate Concentration Corrected
      to12%CO2

      The corrected particulate concentrations shown in Table 13 are submitted to statistical analysis
according to the model in Appendix B.I.  Each week's results are analyzed separately first, with 8 trains

-------
 (7 laboratories) per run. These ANOVA provide the estimates of the within-laboratory term, and the three
 tables generated are shown in Table Bl 1.
   TABLE Bl 1. ANOVA FOR PARTICULATE CONCENTRATION
        CORRECTED TO 12% CO2, INDIVIDUAL WEEKS
Source
df
SS
MS
F
EMS
Week I (Mean = 596.40)
Runs
Labs
Error
Total
2
7
14
23
42,945.57
134,463.36
76,262.73
253,671.66
21.472.79
19,209.05
5,447.34
3.94*
3.53*
a2 + 3o*
a> L
Week 2 (Mean = 659.03)
Runs
Labs
Error
Total
4
7
28
39
637,229.77
1,006,676.51
309,433.13
1,953,339.41
159,307.44
143,810.93
11,051.18
14.42*
13.01*
a' + 5aL
Week 3 (Mean = 743.66)
Runs
Labs
Error
Total
4
7
27
38
660,077.51
283,688.30
257,054.05
1,200,819.86
165,019.38
40,526.90
9,520.52
17.33*
4.26*

*Significant at 5% level
and
                                          a\ = 5447.34

                                          a\ = 11051.18
                                          a\ =9520.52
     The differences between paired-lab-
oratory trains was shown in Appendix B.2
to be insignificant, and the two concentra-
tions in a given run were determined to be
replicate Method 5 determinations.  Using
the formula in Appendix B.I , a pooled
within-laboratory term can be estimated,
as
        = E E
         i =  i / = i
                                                                                     F<--)2
                                                            for the 13 runs. Substituting the values
                                                            from Table  13 into this formula gives

                                                                         ^ = 3072.88

                                                            with  13 df.  This estimates a2 ,the within-
                                                            laboratory variance.  However, there is no
                                                            effect in these results from the determination
                                                            of CO2 in the gas stream by Method 3.

                                                                 A second estimate of a2 is obtainable
                                                            directly from the ANOVA table.  The EMS
                                                            of the error  term in the ANOVA tables is
                                                            a2. Thus we have,
from the three weeks. Using Bartlett's test, the three estimates above can be shown to be estimating the same
true variance. The test statistic is X2= 2.09 with 2 df, and has a significance level of 0.35 associated with it.
Thus the pooled estimate is used,

                                          a2  =9315.22

with 69 df. This would give an estimated standard deviation of
                                           = 96.52
for the within-laboratory term.
     The data from the 6 laboratories who participated in all 13 sampling runs are used to obtain the
estimated laboratory bias component. The ANOVA table for the 3 week analysis is shown in Table B12.
The laboratory term is significant, which allows contrasts among the means to be tested.
                                              89

-------
               TABLE B12. ANOVA FOR PARTICULATE CONCENTRATION CORRECTED TO
                                1 2% CO,, THREE-WEEK ANALYSIS
Source
Runs
Labs
4 vs5
6 vs 7
8 vs 9
Error
4 vs 5
6 vs 7
8 vs9
Total
df
12
5
1
1
1
59
12
12
12
76
SS
1,152,948.09
771,566.39
59,219.56
1 1,779.33
453,607.62
977,989 77
16,410.11
3,646.97
39,072 88
2,902,504.25
MS
96,079.01
154,313.28
. 59,219.56
11,779.33
453,607.62
16,576.10
1,36751
303.91
3,256.07

F
5 80*
9.31*
43.30*
38.76*
139 31*





EMS
_
a2 + 13o[



o2




*Significant at 5% level.
      The three contrasts between paired laboratories provide a total SS of 524,606.51, with 3 df. The
mean square, then, is

                                    MSL =524,606.51/3

                                         = 174,868.84.

The EMS of this term is a2 + 13 al, so that

                                    _MSL -a2
                                         -


                                    _ 174,868.84 -3072.88
                                              13

                                    = 13,215.07

with 3 df. This gives an estimated standard deviation of

                                         aL = 114.96

for the laboratory bias term.
                                           90

-------
                                   LIST OF REFERENCES
1.    Cramer, H., Mathematical Methods of Statistics, Princeton University Press, New Jersey, 1946.

2.    Environmental Protection Agency, "Standards of Performance for New Stationary Sources," Federal
     Register, Vol. 36, No. 247, December 23, 197J, pp. 24876-24893.

3.    Hamil, H. F. and Camann, D. E., "Collaborative Study of Method for the Determination of Particulate
     Matter Emissions from Stationary  Sources (Portland Cement Plants)." Southwest Research Institute
     report for Environmental Protection Agency, EPA-650/4-74-029, May 1974.

4.    Hamil, H. F. and Thomas, R. E., "Collaborative  Study of Method for the Determination of Particulate
     Matter Emissions from Stationary  Sources (Fossil Fuel-Fired Steam Generators)." Southwest Research
     Institute report for Environmental Protection Agency, June 30, 1974.

5.    Hamil, H. F. and Thomas, R. E., "Collaborative  Study of Method for the Determination of Particulate
     Matter Emissions from Stationary  Sources (Municipal Incinerators)." Southwest Research Institute
     report for Environmental Protection Agency, July 1, 1974.

6.    Hamil, H. F. and Thomas, R. E., "Collaborative  Study of Method for Stack Gas Analysis and
     Determination of Moisture Fraction with Use of Method 5." Southwest Research Institute report
     for Environmental Protection Agency, EPA-650/4-74-026, June 1974.

7.    Hamil, H. F. and Thomas, R. E., "Collaborative  Study of Method for Determination of Stack Gas
     Velocity and Volumetric Flow Rate in Conjunction with EPA Method 5." Southwest Research
     Institute report for Environmental Protection Agency, EPA-650/4-74-033, September 1974.

8.    Mitchell, W. J. and Midgett, M. R., "Means to Evaluate Performance of Stationary Source Methods,"
     Environmental Science and Technology, Vol 10, Number 1, January, 1974.

9.    Mitchell, W. J. and Midgett, M. R., "Method for  Obtaining Replicate Particulate Samples from
     Stationary Sources," Environmental Protection Agency, Environmental Monitoring Series,
     EPA-650/4-75-025,June 1975.

10.   Rom, Jerome J., "Maintenance, Calibration, and Operation of Isokinetic Source Sampling Equipment,"
     Environmental Protection Agency, APTD-0576, March  1972.

11.   Searle, S. R., Linear Models, Wiley, New York, 1971.

12.   Siegel, S. Nonparametric Statistics: For the Behavioral Science, McGraw-Hill Book Company, Inc,
     New York, 1956.

13.   Shigehara, R. T., "Adjustments in the EPA Nomograph for Different Pilot Tube Coefficients and
     Dry Molecular Weights." Stack Sampling News,  Vol. 2, No. 4, October, 1974, pp. 4-11.

14.   Snedecor, G. W. and Cochran, W. G., Statistical Methods, Iowa State University Press, Ames,  1967.

15.   Ziegler, R. K., "Estimators of Coefficients of Variation  Using k Samples." Technometrics, Vol. 15,
     No. 2, May 1973, pp. 409-414.
                                                91

-------
                                   TECHNICAL REPORT DATA
                            (Please read Instructions on the reverse before completing)
 1 REPORT NO
  EPA-600/4-76-014
 4. TITLE AND SUBTITLE
 COLLABORATIVE STUDY OF  PARTICULATE EMISSIONS MEASUREMENT
 BY EPA METHODS 2, 3, AND  5  USING PAIRED PARTICULATE
 SAMPLING TRAINS  (MUNICIPAL  INCINERATORS).	
             6. PERFORMING ORGANIZATION CODE
                                                           3. RECIPIENT'S ACCESSION-NO.
                                                           5. REPORT DATE
                                                             MARCH 1976
 7 AUTHOR(S)
                                                           8. PERFORMING ORGANIZATION REPORT NO.
 Henry F. Hamil and Richard  E.  Thomas
 9. PERFORMING ORGANIZATION NAME AND ADDRESS
 Southwest Research Institute
 8500 Culebra Road
 San Antonio, Texas  78284
              10. PROGRAM ELEMENT NO.

                1HA327
              11. CONTRACT/GRANT NO.

                68-02-0626
 12. SPONSORING AGENCY NAME AND ADDRESS
 Environmental Monitoring  and  Support Laboratory
 U.S.  Environmental Protection Agency
 Office of Research and Development
 Research Triangle Park, North Carolina  27711
              13. TYPE OF REPORT AND PERIOD COVERED
                Contract	
              14. SPONSORING AGENCY CODE
                EPA-ORD
 15. SUPPLEMENTARY NOTES
 16. ABSTRACT

        This report represents  the results of statistical  analyses of data from  a  colla-
 borative test using paired particulate sampling trains.   The purposes of the  test were
 to  estimate the minimum variability that can be expected  with the use of Method 5 and
 to  determine the effect of spatial/temporal changes  in  the  gas flow on a Method 5 re-
 sult.   The paired train consists  of two mirror-image Method 5 trains in a single  box,
 and allows two independent laboratories to obtain simultaneous particulate concentra-
 tion data with probe nozzles only 5.8 cm apart.  The report deals with Method 5,  and
 also Method 2 (Velocity) and Method 3 (Stack Gas Analysis), which are called  for  in
 the use of Method 5.  In addition,  the particulate concentrations are converted to the
 applicable compliance test result for the source tested,  and these are also analyzed.
 The latest in-house revisions  of  the EPA methods were used  in this test, and  the  re-
 sults  contained here are applicable to these revisions.
                                KEY WORDS AND DOCUMENT ANALYSIS
                  DESCRIPTORS
Air  Pollution
Flue  Dust
                                              b.IDENTIFIERS/OPEN ENDED TERMS  C.  COSATI Held/Group
 Source  sampling
 Particulates
 Collaborative testing
 Methods  standardization
13B
 3. DISTRIBUTION STATEMENT
Release  Unlimited
                                              19. SECURITY CLASS (This Report)
                                                Unclassified
                           21. NO. OF PAGES

                                100
20. SECURITY CLASS (This page!
  Unclassified
                                                                         22. PRICE
EPA Form 2220-1 (9-73)

-------