United States Industrial Environmental Research EPA-600/2-79-044
Environmental Protection Laboratory February 1979
Agency Research Triangle Park NC 27711
Research and Development
Emission Factors and
Frequency of Leak
Occurrence for Fittings
in Refinery Process Units
-------
RESEARCH REPORTING SERIES
Research reports of the Office of Research and Development, U.S. Environmental
Protection Agency, have been grouped into nine series. These nine broad cate-
gories were established to facilitate further development and application of en-
vironmental technology. Elimination of traditional grouping was consciously
planned to foster technology transfer and a maximum interface in related fields.
The nine series are:
1. Environmental Health Effects Research
2. Environmental Protection Technology
3. Ecological Research
4. Environmental Monitoring
5. Socioeconomic Environmental Studies
6. Scientific and Technical Assessment Reports (STAR)
7- Interagency Energy-Environment Research and Development
8. "Special" Reports
9. Miscellaneous Reports
This report has been assigned to the ENVIRONMENTAL PROTECTION TECH-
NOLOGY series. This series describes research performed to develop and dem-
onstrate instrumentation, equipment, and methodology to repair or prevent en-
vironmental degradation from point and non-point sources of pollution. This work
provides the new or improved technology required for the control and treatment
of pollution sources to meet environmental quality standards.
EPA REVIEW NOTICE
This report has been reviewed by the U.S. Environmental Protection Agency, and
approved for publication. Approval does not signify that the contents necessarily
reflect the views and policy of the Agency, nor does mention of trade names or
commercial products constitute endorsement or recommendation for use.
This document is available to the public through the National Technical Informa-
tion Service, Springfield, Virginia 22161.
-------
EPA-600/2-79-044
February 1979
Emission Factors and Frequency
of Leak Occurrence for Fittings
in Refinery Process Units
by
Robert Wetherold and Lloyd Provost
Radian Corporation
P.O. Box 9948
Austin, Texas 78766
Contract Nos. 68-02-2147 and 68-02-2665
Program Element No. 1AB604
EPA Project Officer: Dale A. Denny
Industrial Environmental Research Laboratory
Office of Energy, Minerals, and Industry
Research Triangle Park, NC 27711
Prepared for
U.S. ENVIRONMENTAL PROTECTION AGENCY
Office of Research and Development
Washington, DC 20460
-------
TABLE OF CONTENTS
Page
List of Figures iv
List of Tables vii
Section 1 - Executive Summary 1
Section 2 - Introduction 4
Section 3 - Program Design and Source Selection 5
Section 4 - Results of Screening and Sampling
Program 11
Appendix A - Emission Factor Development - Statistical
Considerations
References
iii
-------
LIST OF FIGURES
Figure 4-1. Nomograph for Predicting Total Nonmethane Hydrocarbon
Leak Rates from Maximum Screening Values - Valves
and Flanges - Gas/Vapor Streams - Part I 13
Figure 4-2. Nomograph for Predicting Total Nonmethane Hydrocarbon
Leak Rates from Maximum Screening Values - Valves
and Flanges - Gas/Vapor Streams - Part II 14
Figure 4-3. Nomograph for Predicting Total Nonmethane Hydrocarbon
Leak Rates from Maximum Screening Values - Valves
and Flanges -Light Liquid/Two-Phase and Heavy
Liquid Streams Part I 15
Figure 4-4. Nomograph for Predicting Total Nonmethane Hydrocarbon
Leak Rates from Maximum Screening Values - Valves
and Flanges - Light Liquid/Two-Phase and Heavy Liquid
Streams Part II 16
Figure 4-5. Nomograph for Predicting Total Nonmethane Hydrocarbon
Leak Rates from Maximum Screening Values - Pumps, Com-
pressors, Drains, Relief Valves - Part I 17
Figure 4-6. Nomograph for Predicting Total Nonmethane Hydrocarbon
Leak Rates from Maximum Screening Values - Pumps, Com-
pressors, Drains, Relief Valves - Part II 18
Figure 4-7A. Cumulative Distribution of Total Emissions by Screening
Values - Valves - Gas/Vapor Streams 28
Figure 4-7B. Cumulative Distribution of Sources by Screening Values -
Valves - Gas/Vapor Streams 29
Figure 4-8A. Cumulative Distribution of Total Emissions by Screening
Values - Valves - Light Liquid/Two-Phase Streams ... 30
Figure 4-8B. Cumulative Distribution of Sources by Screening Values -
Valves - Light Liquid/Two-Phase Streams 31
Figure 4-9A. Cumulative Distribution of Total Emissions by Screening
Values - Valves - Heavy Liquid Streams 32
Figure 4-9B. Cumulative Distribution of Sources by Screening Values -
Valves - Heavy Liquid Streams 33
iv
-------
Page
Figure 4-10A. Cumulative Distribution of Total Emissions by Screening
Values - Pump Seals - Light Liquid Streams 34
Figure 4-10B. Cumulative Distribution of Sources by Screening Values -
Pump Seals - Light Liquid Streams 35
Figure 4-11A. Cumulative Distribution of Total Emissions by Screening
Values - Pump Seals - Heavy Liquid Streams 36
Figure 4-11B. Cumulative Distribution of Sources by Screening Values -
Pump Seals - Heavy Liquid Streams 37
Figure 4-12A. Cumulative Distribution of Total Emissions by Screening
Values - Compressor Seals - Hydrocarbon Service 38
Figure 4-12B. Cumulative Distribution of Sources by Screening Values -
Compressor Seals - Hydrocarbon Service 39
Figure 4-13A. Cumulative Distribution of Total Emissions by Screening
Values - Compressor Seals - Hydrogen Service 40
Figure 4-13B. Cumulative Distribution of Sources by Screening Values -
Compressor Seals - Hydrogen Service 41
Figure 4-14A. Cumulative Distribution of To'tal Emissions by Screening
Values - Flanges 42
Figure 4-14B. Cumulative Distribution of Sources by Screening Values -
Flanges 43
Figure 4-15A. Cumulative Distribution of Total Emissions by Screening
Values - Drains 44
Figure 4-15B. Cumulative Distribution of Sources by Screening Values -
Drains 45
Figure 4-16A. Cumulative Distribution of Total Emissions by Screening
Values - Relief Valves 46
Figure 4-16B. Cumulative Distribution of Sources by Screening Values -
Relief Valves 47
Figure A-l. Valves and Flanges - Leak Rate/Screening Relationship -
Gas/Vapor Streams A-9
Figure A-2. Valves and Flanges - Leak Rate/Screening Relationship -
Light Liquid/Two-Phase and Heavy Liquid Streams. . . . A-10
-------
Figure A-3. Pumps, Compressors, Drains, Relief Valves - Leak vs
Screening - All Process Streams A-ll
Figure A-4. Distribution of Logio (Max Screening Value)_Valves -
Gas/Vapor Streams A-13
Figure A-5. Cumulative Distribution of Total Emissions by Screening
Values - Valves - Light Liquid/Two-Phase Streams -
Comparison of Confidence Intervals A-17
-------
LIST OF TABLES
Page
Table 1-1. Estimated Vapor Emission Factors for
Nonmethane Hydrocarbons from Selected
Sources 2
Table 3-1. Process Units Sampled in Nine Refineries 6
Table 3-2. Range of Choice Variables for Screened
Baggable Sources 7
Table 3-3. Process Stream Classification by Group 9
Table 4-1. Confidence Intervals for Mean and
Individual Leak Rates - Valves and Flanges -
Gas/Vapor Streams 19
Confidence Intervals for Mean and
Individual Leak Rates - Valves and Flanges - Light
Liquid/Two-Phase and Heavy Liquid Streams 20
Confidence Intervals for Mean and
Individual Leak Rates - Pump Seals, Compressor
Seals, Drains, Relief Valves - All Process
Stream Types 21
Table 4-2. Summary Statistics and Estimated Vapor
Emission Factors for Nonmethane Hydrocarbons
from Baggable Sources 22
Table 4-3. Hypothetical Refinery - Based on ADL Texas
Gulf Cluster Model (330,000 BPCD) 25
Table 4-4. Distribution of Measured Leak
Rates 26
Table 4-5. Percent of Total Mass Emissions Released
by the Upper Ten Percent of Screened
Sources 49
Table A-l. Prediction Equations for Nonmethane Leak Rates
Based on Maximum TLV Screening or Rescreening
Values A-2
vii
-------
SECTION 1
EXECUTIVE SUMMARY
For the past three years, the Environmental Protection Agency has been
sponsoring a petroleum refinery assessment program. As part of this program,
fugitive hydrocarbon emissions have been measured from a number of types of
sources in refineries. This report has been prepared to distribute some of
the data and results obtained thus far in the program. These data should
prove to be of value in the development of state and federal regulations in-
volving fugitive hydrocarbon emissions from sources in petroleum refineries.
The results presented in this report have been developed from data ob-
tained at nine refineries throughout the country. The data for compressor
seals and relief valves represent thirteen refineries. A wide variety of
equipment, process units, and crude oils were included in this study. In
this report, the nonmethane hydrocarbon emissions from selected fugitive
sources are described. These sources include valves, flanges, pump seals,
compressor seals, relief valves, and process drains.
At each refinery, random sample sets of each type of source were selec-
ted, "screened", and if necessary, sampled. The "screening" of sources was
accomplished with portable hydrocarbon detectors. The "screening values"
were defined as the maximum hydrocarbon concentrations detected at the sel-
ected source. Extensive data were taken for each of the selected sources.
One of the important results of this study has been the establishment
of relationships between the screening values and the measured leak rates of
'the various sources. Nomographs were developed which relate the predicted
mean leak rates to the maximum screening values for the various source types.
Confidence intervals for the true mean leak rate are included in the nomo-
graphs .
The emission factors for nonmethane hydrocarbon emissions from selected
types of sources were developed. These emission factors are summarized in
Table 1-1. A very high correlation was found between mass emission rates
from sources and the -type of stream service in which' the sources were em-"
ployed. Except for compressed gases, streams were classified into one of
three"stream groups; (1) gas/vapor streams (2) light liquid/two-phase streams,
and (3) kerosene and heavier liquid streams. Gases passing through compressors
were classified as either hydrogen or hydrocarbon service. It was found that
sources in gas/vapor stream service had higher emission rates than those
sources in heavier stream service. This trend was especially pronounced for
valves and pump seals. Overall emission factors for these two sources are
not presented. The distribution of valves and pump seals within the three
stream groups must be known or estimated to develop applicable overall pro-
cess unit and refinery emission factors for these two source types.
-------
TABLE 1-1. ESTIMATED VAPOR EMISSION FACTORS FOR NONMETHANE EYDROCARBONS FROM SELECTED SOURCES
ESTIMATED
PERCENT
SOURCE TYPE LEAKING
VALVES
Gas/Vapor Streams 29.3
Light Liquid/Two-Phase Streams 36.5
Heavy Liquid Streams 6.7
FLANGES 3.1
PUMP SEALS
Light Liquid Streams 63.8
Heavy Liquid Streams 22.6
COMPRESSOR SEALS
Hydrocarbon Service 70.3
Hydrogen Service 81.2
DRAINS 19.2
RELIEF VALVES 39.2
95% CONFIDENCE
INTERVAL FOR PERCENT
LEAKING
(25.9
(33.6
( 4.6
( 2.2
(58.9
(17.1
(62.9
(71.2
(14. f,
(25.1
, 32.7)
, 39.5)
, 8.9)
, 3.9)
, 68.8)
, 28.1)
, 77.7)
, 88.8)
, 24.0)
, 40.6)
EMISSION 95% CONFIDENCE
FACTOR INTERVAL FOR
ESTIMATE EMISSION FACTOR
(Ib/hr - Source) (Ib/hr - Source)
0.047 (0.027
0.023 (0.016
0.0007 (0.0002
0.00058 (0.0002
0.26 (0.17
0.045 (0.02
0.98 (0.46
0.10 (0.04
0.070 (0.02
0.19 (0.07
, 0.084)
, 0.034)
, 0.002)
, 0.001)
, 0.39)
, 0.11)
, 2.0)
, 0.24)
, n..?0)
, 0.52)
'Leaking sources in this report are defined as sources with screening values
>_200 ppmv or sources with measured leak rates greater than 0.00001 Ibs/hr.
-------
Emission factors are given for compressor seals in each of the two gas
service classifications. The number of seals in each type of service must
be known to develop overall process unit or refinery emissions.
Valves, because of their number and relatively high emission factor are
the major emission source among the source types discussed in this report.
This conclusion is based on an analysis of a hypothetical refinery coupled
with the emission rates developed herein.
Nomographs have also been prepared which relate the maximum screening
values to the percentage of sources leaking and to the percentage of total
emissions represented by these leaking sources. These nomographs can be
used to predict the potential maximum* reduction of emissions due to mainte-
nance programs. Data have been collected at four additional refineries to
determine the effectiveness of maintenance in reducing fugitive emissions.
These data are not discussed in this report, but will be available in the
final report for this program.
* Maximum, because the percentages from the nomographs can only be eliminated
if maintenance will result in zero emissions for any given source.
-------
SECTION 2
INTRODUCTION
Because of the amount of current activity in the development of state
and federal regulations involving fugitive hydrocarbon emissions from refining
activities, it is imperative to place available current data in the hands of
all concerned parties.
This report has been prepared solely for the purpose of distribution of
data. For that reason, only minimal detail has been provided on the experi-
mental and computational aspects of the program. The discussion of the data
is also brief pending the development of complete sets of all correlation
factors.
The emission factors presented in this report are completely general
(i.e., they are neither region nor refinery specific), and for that reason,
they have a wider applicability than previously generated data.
-------
SECTION 3
PROGRAM DESIGN AND SOURCE SELECTION
The results presented in this report have been developed from data
taken primarily at nine refineries. Data for compressors and relief valves
were also obtained at four additional refineries. The refineries were lo-
cated in major refining areas throughout the country. Large, small, new and
old refineries were sampled. Thus, a wide variety of equipment, process
units, and crude oils were included in this study.
All of the major refinery processing units were studied in this program.
In each refinery, sources in six to nine refinery units were selected for
study. The number of each type of process unit sampled are presented in Table
3-1.
In this report, the emissions from "baggable" sources are presented.
Baggable sources are those that can be completely enclosed to measure their
emission rates. Included in this catagory are valves, flanges, pump seals,
compressor seals, relief valves, and process drains.
Variables which might affect the fugitive emissions from baggable
sources were classified as either choice or correlating parameters. A choice
parameter is defined as a variable that may directly affect fugitive emission.0
and is used in selecting the source distribution. These choice parameters
for the five types of baggable sources are listed in Table 3-2.
All other variables which might affect the level of fugitive emissions
are considered as correlating parameters. Jhe^v^ues^f__aii^BeIect.ed cor-
relatj.ng variables._,ar^_r_ecorded__for each selected source. The number of
'correlating parameters ranges from 17 for pumps to 2 for drains.
Obtaining a statistically significant sample population for all com-
binations of even the choice variables~wbuld require a prohibitively large
number of samples. Therefore, a sampling plan was devised that required a
minimum number of measurements to determine leak rates within the desired
accuracies. By assuming that interactions between the variables were un-
Impdrtant~7~the number of necessary measurements were reduced. A factorial
experimental design procedure was used to select combinations of the vari-
ables so that the relationship of each variable with the leak rates could be
determined. Regression analysis and analysis of variance were used to deter-
mine which vaTiabT^^r"^signiTicantTy7ferated to the_ leakage. The sampling
""plans for~suceessive selected refineries were adjusted according to the re-
sults from the refineries that had been sampled previously.
The distribution of selected sources in the first six refineries en-
compassed the full range of choice and correlating variables. One of the
most important of the correlating parameters is the type ofjyrpcess stream
associated with each of the various selected source types. Hydrocarbon
-------
TABLE 3-1. PROCESS UNITS SAMPLED IN NINE REFINERIES
NUMBER OF
REFINERY PROCESS UNIT SAMPLED UNITS
Atmospheric Distillation 7
Vacuum Distillation 4
Thermal Operations (Coking) 2
Catalytic Cracking 5
Catalytic Reforming 6
Catalytic Hydrocracking 2
Catalytic Hydrorefining 2
Catalytic Hydrotreating 7
Alkylation 6
Aromatics/Isomerization 3
Lube Oil Manufacture 2
Asphalt Manufacture 1
Fuel Gas/Light-Ends Processing 11
LPG 2
Sulfur Recovery 1
Other 3
-------
TABLE 3-2. RANGE OF CHOICE VARIABLES FOR SCREENED BAGGABLE SOURCES
BAGGABLE SOURCE
CHOICE VARIABLE
VARIABLE RANGES
FOR SCREENED SOURCES
Valves
Flanges
Pump Seals
Compressor Seals
Drains
Relief Valves
Pressure
Temperature
Fluid State
Service
Function
Size
Pressure
Temperature
Fluid State
Service
Size
Pressure
Temperature
Capacity
Shaft Motion
Seal Type
Liquid RVP
Pressure
Temperature
Capacity
Shaft Motion
Seal Type
Lubrication Method
Service
Pressure
Temperature
Fluid State
-10 - 3000 psig
-190 - 925°F
Gas, Liquid, Two-Phase
In-line, Open-ended
Block, Throttling, Control
0.5 - 36 inches
-14 - 3000 psig
-30 - 950°F
Gas, Liquid, Two-Phase
Pipe, Exchanger, Vessel,
Orifice
1-54 inches
0 - 3090 psig
0 - 800°F
0 - 100,000 gpm
Centrifugal, Reciprocating
Mechanical Seal, Packed Seal
Complete range
0 - 3000 psig
40 - 300°F
0.06 - 66.0 MMSCFD
Centrifugal, Reciprocating
Packed, Labyrinth, Mechanical
Hydrocarbon lubricant
Active, Wash-up
0 - 1350 psig
40 - 1100°F
Gas, Liquid
-------
stream groups were developed, and all process streams were placed in one of
these classifications. Three major stream groups were defined. The three
groups and the descriptions of the streams within each group are shown in
Table 3-3. The "Gas/Vapor" group contains those hydrocarbon streams which
were completely vaporized at the process conditions. Light hydrocarbon
liquids and two-phase streams are included in the "Light Liquid/Two-Phase"
group. The "Heavy Liquids" group contains those streams which consist primari-
ly of kerosene and heavier hydrocarbon liquids. Compressor seal emissions do not
vary according to these three process streams, instead the emissions are bro-
ken down by a) hydrocarbon service and b) hydrogen service. The stream group
is determined by the most volatile stream component present in a concentra-
tion of 20% or more.
The approximate numbers of each type of source selected for study and
testing in each refinery were:
valves 250-300
flanges 100-750
pump seals 100-125
compressor seals 10- 20
drains 20- 40
relief valves 20- 40
There were normally 500-600 sources studied in each refinery.
The distribution of sources among the process units was determined be-
fore the selection and testing of individual sources was begun. Individual
sources were selected from piping, instrumentation and process flow diagrams
before a refinery processing area was entered. Only those preselected sources
were screened. In this way, bias based on observation of individual sources
was eliminated, and an approximate random sample set within the required
source distribution was achieved.
The screening of sources was accomplished with sensitive portable hydro-
carbon detectors. The principal device used in this study was the J. W.
Bacharach Instrument Co. "TLV Sniffer". The Century Instrument Co. Organic
Vapor Analyzer (Model OVA-108) was used for some screening, but these values
were not included in the correlation calculations which follow. The instru-
ments were calibrated with standard mixtures of hexane in air. The OVA-108
and TLV Sniffer give direct readings of hydrocarbon concentrations in ppm by
volume. In this report, the terms "screening values" and "TLV screening
values" refer to the maximum hydrocarbon concentration detected at selected
baggable sources.
-------
TABLE 3-3. PROCESS STREAM CLASSIFICATION BY GROUP
STREAM GROUP
i
Gas/Vapor
Light Liquids/Two-Phase.
Heavy Liquids
HYDROCARBON STREAM DESCRIPTION2
Ci-C2 Hydrocarbons
Ca-Cit Hydrocarbons
Cs-Ca Hydrocarbons
GIO+ Hydrocarbons
Mixed Molecular Weight Hydrocarbon
Streams
Aromatic Hydrocarbons
Miscellaneous Organic Compounds
Hydrocarbon Streams Containing Ha,
and H20
Cz Hydrocarbons
Cs-Cij Hydrocarbons
C5-C6 Hydrocarbons
Cy-Cg Hydrocarbons
Naphtha
Light Distillate
Aromatic Hydrocarbons (low molecular
weight)
Miscellaneous Streams
Kerosene, Diesel, Heating Oil
Gas Oils
Atmospheric Resid/Vacuum Gas Oil
Vacuum Resid/Asphalt
Aromat ics /Polymer s
Mixed Molecular Weight Streams
Non-distillate Solvents
Miscellaneous Organic Streams
'stream group is determined by the stream conditions within the process lines
LThe most volatile stream component present at a concentration of 20% or more
dete
-------
When screening valves, pumps or compressors, the probe of the hydro-
carbon detector was normally placed as close as possible to the intersection
of the shaft with the sealing device (0 cm). The probe was held at this
location for a minimum of five seconds. The detector reading was recorded.
This was repeated at three other points 90° apart around the shaft. The
maximum reading was used as the sampling criterion.
Flanges were screened by placing the detector probe at 2-inch intervals
all around and right against the outside perimeter of the flange interface.
The maximum detector reading was recorded. Drains were similarly screened.
The detector probe was placed at 2-inch intervals around the perimeter of the
drain. The maximum measured hydrocarbon concentration was recorded.
Relief valves were screened by placing the instrument probe at the valve
"horn" exit. The screening value obtained at that point was used as the
sampling criterion.
The values of all choice variables and correlating parameters were re-
corded at the time of screening. The leak rates from sources with screening
values below 200 ppmv hydrocarbon were considered to be negligible. All
sources with screening values of 200 ppmv or greater were candidate sources
for sampling to measure their leak rates. Time and equipment constraints pre-
vented all candidate sources from being sampled; emissions from the candidate
sources which were not sampled were estimated as described in Appendix A.
At the time of sampling, all sources were rescreened. Values of the recorded
source data were checked and, if necessary, changed to conform to the
conditions existing at the time of sampling.
10
-------
SECTION 4
RESULTS OF SCREENING AND SAMPLING PROGRAM
The results of this program are summarized in this section of the re-
port. The screening results are described and interpreted. The frequency
of leaks from the various baggable sources are presented. The emission fac-
tors which were developed from the sampling results are given.
Relationships Between Screening Readings and Nonmethane Leak Rates
One of the important results of this study has been the establishment
of relationships between concentrations obtained using portable hydrocarbon
detection devices and measured leak rates. The use of these portable de-
vices to estimate leak rates has potential for refinery maintenance programs,
as well as in determining total leak rates for regulatory purposes. The
data analyzed in this section were obtained primarily using a J.W. Bacharach
"TLV Sniffer" calibrated to hexane (by volume). Another portable device,
the Century Instrument Company's Organic Vapor Analyzer (Model OVA-108), was
also used and evaluated. The OVA-108 screening values were found to have
similar, but not identical, correlations with leak rates. Most readings
were obtained by placing the detector probe directly on the sources. How-
ever, for evaluation purposes, some readings were also obtained five centi-
meters from the source. It is possible to relate the various kinds of
screening values to one another. For this report, however, the correlations
are based on "TLV Sniffer" readings at the source.
Screening values were obtained when the source was first located, and
rescreening values were taken at the time each source was sampled. The re-
screening values are generally more highly correlated with leak rates than
are the original screening results. For example, the correlation coefficient
for the correlation of maximum screening (original) values with nonmethane
hydrocarbon leak rates of valves is 0.63. A correlation coefficient of 0.72
is obtained when the maximum rescreening values are correlated with non-
methane hydrocarbon leak rates of valves.
Appendix A contains detailed descriptions of the statistical procedures
used to develop the results presented in this report. The least-square
linear regression equations developed for predicting leak rates from un-
sampled sources in the data base are described. For potential prediction
purposes outside this data base, a statistical analysis of covariance was
done to determine whether different linear equations are required for each
baggable source and stream type. The results of this analysis are presented
in Appendix A. It was found that the source and stream types could be
grouped such that three equations were adequate for predicting leak rates
from screened sources. The three groups are as follows:
11
-------
• Valves and Flanges - Gas/Vapor Streams
• Valves and Flanges - Light Liquid/Two-phase and Heavy Liquid Streams
• Pump Seals, Compressor Seals, Drains, Relief Valves
The equations were used to develop nomographs which relate the predicted
leak rate and maximum screening values for the various source and stream
types. These nomographs are presented in Figures 4-1 through 4-6.
Each nomograph gives the predicted mean leak rate for different values
of maximum TLV screening readings taken directly on the source of the leak.
Ninety percent confidence intervals for the true mean leak rate are also
given. These confidence limits are for the mean leak rate and should not be
confused with confidence intervals for individual leak rates for given
screening values. Table 4-1 compares the 90% confidence intervals for the
mean leak rate and the 90% confidence intervals for individual leaks for
selected screening values.
Emission Factors and Frequency of Leaks
The estimated emission factors for nonmethane hydrocarbon emissions from
six types of sources are summarized in Table 4-2. Overall emission factors,
as well as factors for sources in different process stream services, are in-
cluded. The process stream breakdown for flanges, drains and relief valves
are included for information purposes. Summary statistics are tabulated for
each source type. These include (1) 95% confidence intervals for both the
emission factors and for percent of leaking sources, (2) percentile for in-
dividual leaks, and (3) emission factor percentile. The 95% confidence inter-
val is the interval or range of values expected to include the true emission
factor or true percent of leaking sources, with 95% confidence. For example,
the true emission factor for relief valves (if all relief valves could be
sampled and averaged) should be between 0.06 and 0.52 Ib/hr.
If a random selection of sources are sampled and ranked according to
their emission rates, the range of emission rates excluding the first 1% and
last 1% of the sampled sources is the expected range of leak rates (1st and
99th). As an example, the 1% of relief valves with the lowest emission rates
would emit nothing. The 1% of relief valves with the highest emission rates
are expected to emit more than 4.9 Ib/hr from each valve. Therefore, 98% of
all relief valves would be expected to have leak rates between zero and 4.9
Ib/hr.
The population percentile for emission factors represents the leak rate
percentile at which the emission factor is expected to be located in a ran-
dom sample of a particular source type. For instance, the percentile for
relief valves is 91.9 which means that one could expect about 92% of the leak
rates from relief valves to be less than the emission factor of 0.19 Ib/hr.
About 8% of the relief valve leaks could be expected to be greater than the
emission factor. The fact that most of the emission factor percentiles are
12
-------
(Part I: Screening Values from 0-10,000 ppm)
W
H
O
1
f-J
B
H
U
H
R
0.05 -
0.04
0.03
0.02
0.01
Upper Limit of 90S Confidence
/ Interval for Mean
Mean
Lower Limit of 90S Confidence
/Interval for Mean
Equation for Predicted Mean Ling:
NM Leak Rate * 8.59 do'7) (Max Screening)1"16
I past. Squares Equation IKpri to Develop Chart;
Log(NM Leak)= -7.00 + 1.16 Log (Max Screening)
Correlation Coefficient • 0.71
Standard Error = 0.91 Log(Leak Rate)
Number of Data Pairs = 106
Scale Bias Correction Factor = 8.59
2000
4000
6000
3000 10,000
Maximum Screening Value (ppmv as Hexane)
Using J. W. Bacharach TLV Sniffer at the Source.
Figure 4-1. Valves and Flanges - Gas/Vapor Streams - Part I
Nomograph for Predicting Total Nonmethane Hydrocarbon
Leak Rates from Maximum Screening Values
13
-------
w
I
25
O
CJ
o
Pi
w
g
H
O
M
I
Pi
0.9
0.3
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
(Part II: Screening Values from 0 - 100,000 ppm)
, Upper Limit of 90% Confidence
/ Interval for Mean
Mean
Lower Limit of 90% Confidence
' Interval for Mean
Equation for Predicted Mean Line
NM Leak Rate = 8.59 (10"7) (Max Screening)1'16
Least Squarat FqnaHnn Used t.o Develop Chart;
Log(NM Leak) - -7.00 + 1.16 Log (Max Screening)
Correlation Coefficient - 0.71
Standard Error = 0.914 Log (Leak Rate)
Number of Data Pairs * 106
Scale Bias Correction Factor • 8.59
20,000 40,000 60,000 80,000 100,000
Maximum Screening Value (ppmv as Hexane)
Using J. H. Bacharacn TLV Sniffer at the Source
Figure 4-2. Valves and Flanges - Gas/Vapor Streams - Part II
Nomograph for Predicting Total Nonmethane Hydrocarbon
Leak Rates from Maximum Screening Values
14
-------
(Part I: Screening Values from 0-10,000 ppra)
§
i
O
§
w
H
a
0.07 -
0.06 -
0.05 -
Upper Limit of 90S
/ Confidence Interval for Mean
Mean
Lower Limit of 90*
Confidence Interval for Mean
0.01 -
Equation for Predicted Mean Line:
NM Leak Rate - 4.75 (TO"5) (Max Screening)0'76\
Least Squares Equation used ta Develop Chart!
Log(NM Leak) = -4.80 + 0.76 Log (Max Screening)
Correlation Coefficient = 0.76
Standard Error = 0.648 Log(Leak Rate)
Number of Data Pairs = 147
Scale Bias Correction Factor = 2.997
2000
4000
6000
8000
10,000
Maximum Screening Value (ppmv as Hexane)
Using J. U. Bacharach TLV Sniffer at the Source
Figure 4-3. V.alves and Flanges - Light Liquid/Two-Phase and
Heavy Liquid Streams - Part I
Nomograph for Predicting Total Nonmethane Hydrocarbon
Leak Rates from Maximum Screening Values
15
-------
(Part II: Screening Values from 0 - 100,000 ppm)
M
JA
co
r-t
N_/
w
52:
I
n
p
H
o
H
a
' Upper Limit of 90S
/ Confidence Interval for Mean
Mean
0.25 _
Lower Limit of 90% Confidenes
/^ Interval for Mean
Foliation for Predicted Mean Line:
NM Leak Rate =4.75 (lo~5)(Max Screening) °-7&
Least Squares Equation used to Develop Chart:
Log(NM Leak) =• -4.30 + 0.76 Log (Max Screening)
Correlation Coefficient - 0.76
Standard Error « 0.648 Log (Leak Rate)
Number of Data Pairs « 147
Scale Bias Correction Factor « 2.997
j_
_L
20,000 40,000 60,000 80,000 100,000
Maximum Screening Value (ppmv as Hexane)
Using J. W. Bacharach TLV Sniffer at the Source
Figure 4-4. Valves and Flanges - Light Liquid/Two-Phase and
Heavy Liquid Streams - Part II
Nomograph for Predicting Total Nonmethane Hydrocarbon
Leak Rates from Maximum Screening Values
16
-------
w
55
O
PQ
o
O
53
O
W
H
0.6 -
Q.S -
0.4
0.3
0.2
0.1
(Part I: Screening Values from 0 - 10.000 ppm)
Upper Limit of 90%
Confidence Interval for Mean
Mean
Lower Limit of 90%
Confidence Interval for Mean
Equation for Predicted Mean Line:
NM Leak Rate = 4.9 (10"") (Max Screening)0-73
Least Squares Equation Used to Develop Chart;
Log (NM Leak Rate) = -4.00 + 0.73 Log (Max Screening)
Correlation Coefficient = 0.62
Standard Error = 0.78 Log IQ (NM Leak Rate)
Number of Data Pairs = 168
Scale Bias Correction Factor = 4.90
1000 2000 3000 4000 5000 6000 7000 8000 9000 10,000
Maximum Screening Value (ppmv as Hexane)
Using J. M. Bacharach TLV Sniffer at the Source
Figure 4-5. Pumps, Compressors, Drains, Relief Valves - Part I
Nomograph for Predicting Total Nonmethane Hydrocarbons
Leak Rates from Maximum Screening Values
17
-------
S5
o
PQ
O
§
§
H
O
(Part II: Screening Values from 0 - 100,000 pom)
Upper Limit of 90%
/ Confidence Interval for Mean
Mean
Lower Limit of 90S
Confidence Interval for Mean
Equation for Predicted Mean Line:
NM Leak Rate = 4.9 (10-4) (Max Screening)0-73
Least Squares Equation Used to Develop Chart:
Log (NM Leak Rate) = -4.00 + 0.73 Log (Max Screening)
Correlation Coefficient » 0.62
Standard Error = 0.78 Log IQ (NM Leak Rate)
Number of Data Pairs = 168
Scale Bias Correction Factor = 4.90
20,000
40,000
60,000
30,000
100,000
Maximum Screening Value (ppmv as Hexane)
Using J. W. Bacharach TLV Sniffer at the Source
Figure 4-6. Pumps, Compressors, Drains, Relief Valves - Part II
Nomograph for Predicting Total Nonmethane Hydrocrabon
Leak Rates from Maximum Screening Values
18
-------
TABLE 4-1. CONFIDENCE INTERVALS FOR MEAN AND INDIVIDUAL LEAK RATES
VALVES AND FLANGES - GAS/VAPOR STREAMS
vo
MAXIMUM SCREENING
VALUE (ppmv)
1
200
500
1,000
3,000
5,000
10,000
20,000
50,000
100,000
PREDICTED MEAN LEAK
(Ibs/hr)
io-7
io-5
0.001
0.003
0.009
0.017
0.038
0.084
0.24
0.54
RATE
90% CONFIDENCE INTERVAL FOR:
MEAN LEAK
(Ibs/hr)
(0.000 ,
(0.000 ,
(0.000 ,
(0.001 ,
(0.005 ,
(0.01 ,
(0.02 ,
(0.06 ,
(0.17 ,
(0.36 ,
IO"6 )
0.001)
0.003)
0.005)
0.02 )
0.03 )
0.06 )
0.12 )
0.35 )
0.82 )
INDIVIDUAL LEAKS
Clbs/hr)
(0.000 ,
(0.000 ,
(0.000 ,
(0.000 ,
(0.000 ,
(0.001 ,
(0.001 ,
(0.003 ,
(0.007 ,
(0.02 ,
io-* )
0.02)
0.04)
0.09)
0.31)
0.56)
1.2 )
2.7 )
7.9 )
17.7 )
-------
N>
O
(Cont'd.) TABLE 4-1. CONFIDENCE INTERVALS FOR MEAN AND INDIVIDUAL LEAK RATES
VALVES AND FLANGES - LIGHT LIQUID/TWO-PHASE
AND HEAVY LIQUID STREAMS
MAXIMUM SCREENING
VALUE (ppmv)
1
200
500
1,000
3,000
5,000
10,000
20,000
50,000
100,000
PREDICTED MEAN LEAK RATR
(Ibs/hr)
5 x 10~5
0.003
0.005
0.009
0.021
0.031
0.052
0.088
0.18
0.30
90% CONFIDENCE INTERVAL
MEAN LEAK
(Ibs/hr)
(0.000
(0.001
(0.004
(0.007
(0.02
(0.02
(0.04
(0.07
(0.13
(0.22
, 0.004)
, 0.007)
, 0.01 )
, 0.03 )
, 0.04 )
, 0.06 )
, 0.11 )
, 0.24 )
, 0.42 )
FOR:
INDIVIDUAL LEAKS
(Ibs/hr)
(0.000
(0.000
(0.000
(0.001
(0.002
(0.003
(0.004
(0.007
(0.01
(0.02
, 0.001)
, 0.03 )
, 0.06 )
, 0.11 )
, 0.24 )
, 0.36 )
, 0.61 )
, 1.0 )
, 2.1 )
, 3.6 )
-------
(Cont-'d.) TABLE 4-1. CONFIDENCE INTERVALS FOR MEAN AND INDIVIDUAL LEAK RATES
PUMP SEALS, COMPRESSOR SEALS, DRAINS, RELIEF
VALVES - ALL PROCESS STREAM TYPES
MAXIMUM SCREENING
VALUE (ppmv)
I
200
500
1,000
3,000
5,000
10,000
20,000
50,000
100,000
PREDICTED MEAN LEAK RATE
(lbs/hr)
0.0005
0.023
0.046
0.076
0.17
0.25
0.41
0.68
1.3
2.2
90% CONFIDENCE
MEAN LEAK
Clbs/hr)
(0.000 , 0.002)
(0.01 , 0.04 )
(0.03 , 0.07 )
(0.06 , 0.10 )
(0.13 , 0.22 )
(0.20 , 0.31 )
(0.32 , 0.51 )
(0.52 , 0.88 )
(0.96 , 1.8 )
(1.5 , 3.2 )
INTERVAL FOR:
INDIVIDUAL LEAKS
Olbs/hr)
(0.000 , 0.01)
(0.001 , 0.47)
(0.002 , 0.9i)
(0.003 , 1.5 )
(0.009 , 3.3 )
(0.01 , 4.8 )
(0.02 , 7.9 )
(0.03 , 13. )
(0.07 , 26. )
(0.12 , 43. )
-------
TABLE 4-2. SUMMARY STATISTICS AND ESTIMATED VAPOR EMISSION FACTORS
FOR NONMETHANE HYDROCARBONS FROM BAGGABLE SOURCES
NJ
ts>
SOURCE TYPE
VALVES
Gas/Vapor Streams
Light Liquld/Tvo-Phase Streams
Heavy Liquid Streams
PUMP SEALS
Light Liquid Streams
Heavy Liquid Streams
FLANGES (ALL)
Gas/Vapor Streams
Light Liquid/Two-Phase Streams
Heavy Liquid Streams
COMPRESSOR SEALS
Hydrocarbon Service
Hydrogen Service
DRAINS (ALL)
Light Liquid/Two Phase Streams
Heavy Liquid Streams
RELIEF VALVES (ALL)
Gas/Vapor Streams
Light Liquid/Two Phase Streams
Heavy Liquid Streams
TOTAL
SCREENED1
683
1019
522
470
292
2030
369
616
325
145
85
255
100
107
148
92
28
23
NUMBER
LEAKING2
200
372
35
300
66
62
10
33
6
102
69
49
26
19
58
42
7
8
PERCENT
LEAKING
29.3
36.5
6.7
63.8
22.6
3.1
2.7
5.4
1.9
70.3
81.2
19.2
26.0
17.8
39.2
45.6
25.0
34.8
95% CONFIDENCE
INTERVAL FOR
PERCENT LEAKING
(25.9
(33.6
( 4.6
(58.9
(17.1
( 2.2
( 0.8
( 3.3
( 0.2
(62.9
(71.2
(14.4
(14.6
( 9.5
(25.1
(35.2
(10.7
(16.4
, 32.7)
, 39.5)
, 8.9)
, 68.8)
, 28.1)
, 3.9)
, 4-6)
, 7.4)
, 3.5)
, 77.7)
, 88.8)
, 24.0)
, 37.4)
, 26.1)
, 40.6)
i 56.4)
, 44.9)
, 57.3)
PERCENTILE 95* OOHI
FOR INDIVIDUAL EMISSION INTERS
LEAKS (lb»/hr) FACTOR ESTIMATE EMISSI01
1st 99th (Ibs/hr-source) (Ibs/hr-
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0.61
0.31
0.08
4.0
0.87
0.008
0.008
0.011
0.009
17.0
2.0
1.3
1.9
0.63
4.9
4.0
0.38
0.42
0.047
0.023
0.0007
0.26
0.045
0.00058
0.0005
0.0005
0.0007
0.98
0.10
0.070
0.085
0.029
0.19
0.36
0.013
0.019
(0.027
(0.016
(0.0002
(0.17
(0.02
(0.0002
do'5
(0.0002
(io-s
(0.46
(0.04
CO. 02
(0.02
(0.003
(0.06
(0.10
(0.001
(0.001
?IDENCE
U, FOR EMISSION
) FACTOR FACTOR
-source) PERCENTILE
, 0.084)
, 0.034)
, 0.002)
, 0.39 )
, 0.11 )
, 0.001)
, 0.005)
, 0.001)
, 0.02 )
, 2.0 )
, 0.24 )
, 0.20 )
, 0.32 )
, 0.21 )
, 0.52 )
, 1.30 )
, 0.23.)
, 0.20 )
94.4
89.7
95.8
89.8
88.4
97.4
83.4
86.5
93.3
91.9
*Sone streams could not be accurately classified Into stream category so the totalb are not the sum of the stream groups.
2Leaking sources in this report are defined as sources with screening values greater than or equal to 200 ppav or
with measured leak* greater than 0.00001 IbWhr.
-------
greater than 90% is an indication of the extreme skewness of the leak rate
data.
A very high degree of correlation was found between mass emission rates
and the type of stream service in which the sources were employed. The pro-
cess stream groups are described in Table 3-3 of this report.
It is evident from Table 4-2 that individual sources in gas/vapor stream
service emit more hydrocarbons than comparable sources in heavy liquid- service.
On the average, for example, valves in gas/vapor service emit more than 70
times as much material as valves in heavy liquid service.
Because of the large influence of stream type on source emissions, over-
all emission factors for valves and pump seals are not presented. Due to
the experimental design, the distribution of sampled valves and pump seals
among the stream groups is not representative of the overall distribution
of sources within refineries. Thus, overall emission factors for valves and
pump seals, as determined from the sampling programt are not directly appli-
cable to either entire refineries or individual process units. The stream
group distributions must be known or estimated to develop accurate overall
process unit and refinery emission factors for these two source types.
The emission factors for valves given in Table 4-2 were developed for
in-line pipeline valves. The emission points were around the valve stem
and around the perimeter of the packing gland. A small number of open-
ended valves were also sampled, such as drain valves and sampling valves.
The screening and sampling point for these valves was the open-end of the
pipe downstream from the valve. The emission occured through the valve
seat rather than around the valve stem. The average emission factor for
these open-ended valves was 0.007 Ib/hr with a 95% confidence interval of
0.002 - 0.016 Ib/hr. A total of 129 open-ended valves were screened, and
30 were found to have screening values _>_ 200 ppmv hydrocarbon.
Distribution of Sources within a Refinery
In order to ascertain the relative contribution of the various source
types, the number of sources of each type and stream classification in a
refinery must be known. Some counting of source types was done during.this
sampling program, but these counts were not recorded by process stream
groups. To give an indication of the relative importance-of the various
source types, a hypothetical refinery model has been used. The model is
based on the ADL-Texas Gulf Cluster Model1 processing 330,000 BPCD of crude.
The total number of fittings in the various process units in this hypotheti-
cal model are estimated from counts of fittings made during this program.
Since very little information is available on the number of sources in
the various types of stream service, two cases are presented:
1Kittrell, J.R.,Impact of SOX Emissions Control on Petroleum Refining Industry,
Volume II. Detailed Study Results, EPA/699/2-76/161B, June 1976
23
-------
Case I - The distribution of stream types for valves is weighted toward
liquid streams, the distribution for pump seals is weighted
toward the heavy liquid streams, and the distribution for com-
pressor seals is weighted toward hydrogen service.
Case II - The distribution of stream types for valves is weighted toward
gas service, the distribution for pump seals is weighted
toward light liquid streams, and the distribution of compressor
seals is weighted toward hydrocarbon service.
It should be emphasized that these two cases and the hypothetical refinery
model are being used only for exemplary purposes in this report. An equiva-
lent analysis can be done for any particular situation using appropriate num-
ber of sources and stream classifications in conjunction with the emission
factors in this report.
Table 4-3 shows the number of sources for the above two cases in the
hypothetical refinery and the results of applying the emission factors from
Table 4-2 to these numbers. The emissions thus obtained have been divided by
the total emissions from all source types to give the percent of total non-
methane hydrocarbon emissions attributable to each source category.
For this hypothetical refinery model, it is clear that valves, because of
their number and relatively high emission factors, are the major emission
source among"the source categories included in this report. Depending on the
case, valves are expected to contribute from 61 to 67 percent of the total
emissions from the six source types in this hypothetical refinery.
Although flanges represent the greatest number of any source type in the
hypothetical model, their small emission factor results in an expected con-
tribution of only 6 to 8 percent of total emissions. The contribution from
drains has an expected value of 11 to 15 percent. Pump seals, compressor
seals, and relief valves each have maximum expected contributions of 9 percent
or less, using this model.
Distribution of Leak Rates
Table 4-4 gives the distribution of measured and estimated nonmethane
hydrocarbon emission rates for each source type. It is obvious from this
table that the bulk of the emissions emanate from a small percentage of the
sampled fittings. The table represents an idealized situation for reducing
emissions by selective maintenance of sources. For example, the table indi-
cates that if a particular 3.6% of the valves in gas/vapor service in a re-
finery (those with leak rates greater than 0.1 Ibs/hr) could be repaired
such that the leak rate of each such valve was reduced to zero, then 91.1%
of the mass emissions attributable to valves in gas/vapor service could be
eliminated. The problem in practical implementations of such a repair
program is locating those particular 3.6% of the valves in gas/vapor service
with leak rates greater than 0.1 Ibs/hr without bagging and sampling all
such valves in a refinery.
24
-------
Oi
TABLE 4-3. HYPOTHETICAL REFINERY - BASED ON ADL TEXAS
GULF CLUSTER MODEL (330,000 BPCD)
PERCENT OF TOTAL NONMETHANE HYDROCARBON EMISSIONS
SOURCE TYPE
Valves
Gas/Vapor
Light Liquid/Two-Phase
Heavy Liquid
Pump Seals
Light Liquid
Heavy Liquid
Compressor Seals
Hydrocarbon Service
Hydrogen Service
Flanges
Drains
Relief Valves
(venting to atmosphere )
NUMBER OF SOURCES (PERCENT)
CASE I4 CASE II2
14300
1430
7150
5720
264
106
158
27
14
13
51200
793
64
(10%)
(50%)
(40%)
(40%)
(60%)
(52%)
(48%)
14300
4290
5720
4290
158
106
27
20
7
51200
793
64
(30%)
(40%)
(30%)
(60%)
(40%)
(74%)
(26%)
CASE I
ESTIMATE
61.6
17
43
1
9.1
7
1
. 3.9
3
0
7.7
14.5
3.2
100%
.6
.0
.0
.2
.9
.6
.3
(95/6 CD3
(46.5
(10.1
(29.9
( 0.3
( 6.3
( 4.7
( 0.8
( 2.0
( 1.7
( 0.1
( 2.9
( 4-i
( 1.0
, 86.4)
, 31.4)
, 63.5)
, 3.0)
13.6)
, 10.8)
, 4.5)
, 7.7)
, 7.3)
, 0.8)
, 13.4)
, 41.4)
, 3.9)
CASE II
ESTIMATE
67.3
:40.4
26.3
0.6
9.2
8.2
1.0
4.1
3.9
0.2
5.9
11.1
2.4
100%
(95% CI)S
(48.3
(23.2
(18.3
( 0.2
( 6.3
( 5.4
( 0.4
( 1.9
( 1.8
( 0.01
( 2.2
( 3.2
( 0.8
, 100)
, 72.1)
, 38.9)
, 1.7)
, 13.5)
, 12.3)
, 2.3)
, 8.1)
, 8.0)
, 0.15)
, 10.2)
, 31.7)
, 6.7)
1 Case I - distribution of stream types for valves and pump seals is weighted toward heavier streams.
2 Case II - distribution of stream types for valves and pump seals is weighted toward lighter streams.
3 95% CI - 95% confidence interval for percent of emissions from process fittings attributable
to a particular source type. The width of the Interval is due only to the uncertainty in
emission factors; the percentage of each source type is assumed fixed or known.
" Total is the sum of nonmethane hydrocarbon emissions from the six source types discussed In this
table.
-------
TABLE 4-4. DISTRIBUTION OF MEASURED LEAK RATES1
VALVES
PUMP SEALS
to
Leak Range
(Ib/hr)
>1.0
0.1 -1.0
0.01-0.1
0.001-0.01
0.00001-0.001
>0. 00001
Leak Range
(Ib/hr)
>1.0
0.1-1.0
0.01-0.1
0.001-0.01
0.00001-0.001
>0. 00001
Gas/Vapor
Streams
A B
1.0 68. A
2.6 22.7
8.6 7.9
8.9 0.9
8.1 0.1
29.2%
Light Liquid/
Two-Phase Streams
A B
0.0 13.8
3.3 58.8
11.5 23.7
13.4 3.5
8.1 0.2
100% 36.3%
COMPRESSOR SEALS
Hydrocarbon
Service
A
15.9
33.1
16.6
4.8
2.1
72. 5Z
B
74.2
• 24.3 ,
1.4
0.1
0.0
100%
Hydrogen
Service
A
0
16.5
25.9
24.7
14.1
81.2%
100%
Heavy Liquid
Streams
A
0
0.2
0.9
2.7
2.9
6.7%
FLANGES
B
0.
33.6
48.0
16.9
1.5
100%
All Stream
Croups
B
0.
75.6
22.5
1.8
0.1
100%
A
0
0.2
0.6
1.4
0.9
3.1%
B
0.0
63.2
30.1
6.0
0.7
100%
Light Liquid
Streams
A
4.0
15.7
22.5
16.4
4.9
63.5%
DRAINS
B
90.5
24.8
4.3
0.4
0.0
100%
All Stream
Groups
A
1.5
4.7
6.7
5.1
1.2
19.2%
B
61.6
33.0
4.9
0.5
0.0
100%
Heavy Liquid
Streams
A
0
5.5
9.6
5.8
1.7
22.6%
RELIEF
B
0.
73.2
25.6
1.2
0.0
100%
VALVES
All Stream
Groups
A
3.4
10.1
14.9
7.4
2.7
38.5%
B
76.0
19.2
4.5
0.3
0.0
100%
A - Percent of total sources screened with sampled leak rates within leak range.
B - Percent of total mass emissions attributable to sources within leak range.
1 - Most sources were bagged and sampled to obtain leak rates; some were estimated using procedures
described in Section 1 of Appendix A.
-------
This is obviously impractical, so the identification of sources requiring
maintenance must occur through some type of screening program.
Distribution of Emissions and Sources Based on Screening Values
A convenient tool both for monitoring hydrocarbon emission sources and
estimating source leak rates is the portable hydrocarbon detector. From the
results of this study, nomographs have been prepared relating hydrocarbon
concentration at the source (screening value) to the percentage of each
source type expected to have screening values above any selected value.
Other nomographs have been prepared relating screening values to the per-
centage of total mass emissions which can be expected from sources with
screening values greater than any given value. (See Appendix A for a dis-
cussion of nomograph development.)
These nomographs for the six source types (and stream groups for valves,
compressors, and pump seals) are presented in Figures 4-7 through 4-16. The
"A" figures relate the percent of total mass emissions for a given source
category to screening values; the "B" figures relate the percent of sources
to screening values. The screening values in these nomographs are the hydro-
carbon concentrations obtained at the source (zero cm) with a Bacharach TLV
Sniffer calibrated with hexane.
Confidence intervals are included on each of these nomographs. The
statistical procedures used to develop these intervals are discussed in
section 4.2 and 4.3 of Appendix A. The confidence intervals for both types
of nomographs indicate how well the cumulative function has been estimated
from the data collected in this program.
The 95 percent confidence intervals for the cumulative percent of sources
can be interpreted as ranges of values which contain the actual percent from
the population of sources studied. Note that these intervals apply to the
entire population of sources (i.e., a composite of all United States refin-
eries), and are not necessarily applicable to a finite number of sources
at any particular refinery. Because of the nature of the function, the
confidence intervals will be approximately valid anytime a random sample of
greater than 100 sources is being considered.
The 90% cqnfidence intervals for the cumulative percent of total emis-
sions function can be interpreted as ranges of values which contain the
actual percent of total emissions function for the entire population of
sources. Again, these intervals describe how well the function has been
estimated for the entire population and are not directly applicable to a
particular refinery situation with a finite number of sources. The variation
of the function for a particular sample of sources is a complex function of
the number of sources. Section 4-3 of Appendix A discusses this issue
further and gives an example of particular confidence bounds for valves in
light liquid/two-phase service.
27
-------
Valves - Gas/Vapor Streams
100
90
80
I 70
£ 60
V)
in
- 50
•W
f «
O
§ 30
fc.
20
10
N Upper Limit of 90S
Confidence Interval
Estimated
Total
Lower Limit of
90% Confidence
X 2345 10 50100 5001000 10,000 100,000 1,000,000
Screening Value (ppmv) (Log,. Scale)
Percent of Total Mass Emissions - indicates the percent of total emissions
attributable to sources with screening values
greater than the selected value.
figure 4-7A. Cumulative Distribution of Total Emissions by
Screening Values - Valves - Gas/Vapor Streams
28
-------
Valves - Gas/Vapor Streams
100
90
80
70
3 60
c.
8 40
30
20
10
•Upper Limit of 952 Confidence Interval
Estimated Percent of Sources
Lower Limit of
the 95S Confidence
Interval
-P
\ . . 7j--r^r-^
1 2345 10 50100 5001000 10,000 lOQ-,000 1,000,000
Screening Value (ppmv) (Log^. scale)
Percent of Sources - Indicates the percent of sources with screening
values greater than the selected source.
Figure 4-7B. Cumulative Distribution of Sources by
Screening Values - Valves - Gas/Vapor
Streams
29
-------
Valves - Light Liquid/Two-Phase Streams
100
90
80
| 70
t/t
Vt
<§ 60
,_ 40
o
8 30
I
20
10
Estimated Percent of
Total Mass Emissions
\\\ Upper Limit of 90X
Lower Limit" of the
90S Confidence Interval A \
1 2345 10 50100 1000 10,000 100,000 1,000,000
Screening Value (ppmv) (Log1Q Scale)
Percent of Total Mass Emissions - Indicates the percent of total emissions
attributable to sources with screening values
greater than the selected value.
Figure 4-8A. Cumulative Distribution of Total Emissions by Screening
Values - Valves - Light Liquid/Two-Phase Streams
30
-------
Valves - Light Liquid/Two-Phase Streams
100
90
80
70
60
50
40
£
30
20
10
Upper Limit of 95Z Confidence Interval
timated Percent of Sources
1 2345 10
50100 1,000 10,000
Screening Value (ppmv) (Log1Q Scaled
,000
Percent of Sources - indicates the percent of sources with screening
values greater than the selected value.
Figure 4-8B. Cumulative Distribution of Sources by Screening
Values - Valves - Light Liquid/Two-Phase Streams
31
-------
100
90
80
Ul
5 70
(A
«l
iS 60
«•
> M
+. 40
o
g 30
I
20
10
0
Valves - Heavy Liquid Streams
\ \ \ Upper Limit of 90S
. \ \Ctmfidence Interval
\ \ \
Estimated Percent of.
Total Mass Emissions*
Lower Limit of the 90S\ \ \
Confidence Interval \ \ \
\\\
n I i
.L I I I i I
1 2 345 10 50 100 5001000 10,000 100,000 1,000,000
Screening Value (ppmv) (Log1Q Scale)
Percent of Total Mass Emissions - indicates the percent of total emissions
attributable to sources with screening values
greater than the selected value.
Figure 4-9A. Cumulative Distribution of Total Emissions by
Screening Values - Valves - Heavy Liquid Streams
32
-------
Valves - Heavy Liquid Streams
§
I/)
-------
I/I
i
100
90 -
80
70
60
50
40
30
20
10
Pump Seals - Light: Liquid Streams
\ ~ —
Up0er Limit of 90%
Interval
Estimated Percent of
Total Mass Emissions
Lower Limit of the -\\
Confidence Interval 90%
I 2345 10 50 100 5001000 10,000 100,000 1,000,000
Screening Value (ppmv) (Log1Q Scale)
Percent of Total Mass Emissions - indicates the percent of total emissions
attributable to sources with screening values
greater than the selected value.
Figure 4-10A. Cumulative Distribution of Total Emissions by
Screening Values - Pump Seals - Light Liquid
Streams
34
-------
Pump Seals - Light Liquid Streams
£
-------
Pump Seals - Heavy Liquid Streams
\ Upper Limit of 90%
\ Confidence Interval
\ \ \
Estimated Percent of v \ \
Confidence Interval 90s
2 3 45 10
50 100 5001000 10,000 100,000 1,000,000
Screening Value (ppmv) (Log Scale)
Percent of Total Mass Emissions - indicates the percent of total emissions
attributable to sources with screening values
greater than the selected value.
Figure 4-11A. Cumulative Distribution of Total Emissions by
Screening Values - Pump Seals - Heavy Liquid
Streams
36
-------
Pump Seals - Heavy Liquid Streams
O
44
100
90
80
70
60
i 50
30
20
10
Limit of 95% Confidence Interval
Estimated Percent of Sources
Lower Limit of the\ x \
the 95S Confidence \ \ \
Interval N
1 2345 10 50100 5001000 10.000 100,000 1,000,000
Screening Value (ppnrv) (Log,. Scale)
Percent of Sources -Indicates the percent of sources with screening
v«1uM greater than the selected value.
Figure 4-11B. Cumulative Distribution of Sources by Screening Values
Pump Seals - Heavy Liquid Streams
37
-------
Compressor Seals - Hydrocarbon Service
VI
c
o
t/l
I
m
o
0)
u
Upper Limit of 90%
Confidence Interval
Estimated Percent of
Total Mass Emissions
Lower Limit of 9055
Confidence Interval
2 345 10
50 100
500 1000
10,000
100,000 1,000,000
Screening Values (ppmv) (Log1Q Scale)
Percent of Total Mass Emissions - indicates th« percent of total emissions
attributable to sources with screening values
greater than the selected value.
Figure 4-12A. Cumulative Distribution of Total Emissions by
Screening Values - Compressor Seals -
Hydrocarbon Service
38
-------
Compressor Seals - Hydrocarbon Service
Upper Limit of 95%
Confidence Interval
Estimated Percent of Sources
Lower Limit of 95*' \
2345 10 50100 1,000 10,000 100,000 1,000,000
Screening Value (ppmv) (Log-,0 Scale)
Percent of Sources -indicates the percent of sources with screening
value* greater than the selected value.
Figure 4-12B.
Cumulative Distribution of Source by
Screening Values - Compressor Seals -
Hydrocarbon Service
39
-------
Compressor Seals - Hydrogen Service
100
90 -
80 1-
ut
e
•£ 70
V)
i
» 60
s 50
o
30
20
10
0
Upper Limit of 90%
Confidence Interval
Estimated Percent of
V~ Total Mass Emissions
Confidence Interval \
2345 10 50100 1000 10,000 100,000 1,000,000
Screening Values (ppmv) (Log 10 Scale)
Percent of Total Mass Emissions - Indicates tin percent of total emissions
attributable to sources with screening values
greater than the selected value.
Figure 4-13A. Cumulative Distribution of Total Emissions
by Screening Values - Compressor Seals -
Hydrogen Service
40
-------
Compressor Seals - Hydrogen Service
M
41
o
ft
o
•u
c
at
100
90
80
70
60
50
40
30
20
10
Upper Limit of 95%
Confidence Interval
Estimated Percent of Sources
Lower Limit of 95%
Confidence Interval
1 2345 10 50100 1000 10,000 100,000 1,000,000
Screening Value (ppmv) (Log1Q Scale)
Percent of Sources -indicates the percent of sources with screening
value* greater than the selected value.
Figure 4-13B. Cumulative Distribution of Sources by
Screening Values - Compressor Seals -
Hydrogen Service
41
-------
Flanges
100
90
80
| 70
IA
in
1 60
1A
VI
£ 50
«0
£ 40
<*-
o
3d
-------
Flanges
s
«1-
o
100
90
80
70
60
50
40
e
0)
I 30
20
10
0
1
Lower Limit of the
the 95% Confidence
Interval
'Estimated Percent of Sources
, ^Upper Limit of 95* Confidence Interval
2345 10 50100 1000 10,000 100,000 1,000,000
Screening Value (ppmv) (Log,- Scale)
Percent of Sources - Indicates the percent of sources with screening
values greater than the selected value.
Figure 4-14B. Cumulative Distribution of Sources by Screening
Values - Flanges
43
-------
Drains
'100
90 f-
80 -
i 70-
»*">
I/I
Irt
1 60
Vt
3. so
r™
10
£ 40
g 30
20
10
\ \ \ Upper Limit of 90S
\ \ XN Confidence '-* ^
\ \ \
\ \ N
. , \
Estimated Percent of \ \ \
Total Mass Emissions \^ \
Lower Limit of the
Confidence Interval 90% \
i i i i 1 i i i i
,\\\
2345 10 50100 500.1000 10,000 100,000
Screening Value (ppmv) (Log^ Scale)
1,000,000
tereent of Total Mass Emissions - indicates the percent of total emissions
attributable to sources with screening values
greater than the selected value.
Figjire 4-15A. Cumulative Distribution of Total Emissions by
Screening Values - Drains
44
-------
Drains
g
in
<*-
o
+*
|
-------
Relief Valves
\ Upper Limit of 90X
\fr*Confidence Interval
Estimated Percent of
Total Mass Emissions
Lower Limit of 90%
Confidence Interval
2345 10 50 100 1000 10,000
Screening Value (ppmv) (Login Scale)
1,000,000
Percent of Total Mass Emissions - Indicates the percent of total emissions
attributable to sources with screening values
greater than the selected value.
Figure 4-16A. Cumulative Distribution of Total Emissions
of Screening Values - Relief Valves
46
-------
Relief Valves
100
90
80
70
VI
£ 60
3
O
t 50
S 40
-------
The nomographs must be used with caution. As discussed earlier, the
correlation between screening values and actual leak rates is imperfect.
Because of this, values obtained from the nomographs for percent of total
sources and percent of total emissions at a specific screening value will
not exactly match similar values in Table 4-4, if one converts screening
values to leak rates. In most cases, the nomographs will indicate a higher
percentage of sources being responsible for a given percentage of total
emissions than Table 4-4. In this sense, the nomographs are conservative
(i'.e., they will identify more sources than necessary to achieve a given
level of reduction on total emissions). In a practical sense, however, un-
less every source with a screening value exceeding a specific level is
bagged and sampled, there is no better method than screening for identi-
fying sources for maintenance.
The nomographs can be used to evaluate the potential effectiveness of
maintaining and repairing sources for reducing emissions. For example,
approximately 5% of valves in gas/vapor stream service can be expected to
have screening values above 50,000 ppmv (Figure 4-7B) . However, these 5%
of the valves are responsible for an estimated 92% of the mass emissions
(Figure 4-7A). Similarly, for a screening value of 10,000 ppmv, the percent
of sources and percent of emissions are 10%, and 98%, respectively. (See
example lines on Figures 4-7A and 4-7B).
Analyses, using the nomographs, can also be done for other sources and
process streams. For example, Table 4-5 shows the percent of emissions
for various sources and process streams when the upper 10% of screened
sources are considered. Confidence intervals are also shown. Table 4-5
is presented only to illustrate the use of the nomographs and to emphasize the
fact that a small fraction of the sources within any one source category
account for the majority of emissions in that category. There is no intent
here to prejudge that a reasonable level of control is 10% of sources, or
any other specific number. Ultimately, the decision regarding reasonable
control will be based on reducing overall emissions. Therefore, percentage
reduction goals for each source category may be different. The variations
in source emissions from a hypothetical refinery (Table 4-3) further illu-
strates the. need to develop different emission reduction goals for various
source categories.
In summary, Figures 4-7 through 4-16 present continuous distributions
of the percent of emissions and sources versus specific screening values.
These figures can be used to estimate the reduction in emissions which could
ideally occur if a given percentage of leaking sources were repaired. It is
emphasized that these figures represent an amalgamation of data from nine
refineries (thirteen for compressor seals and relief valves) and do not
represent any single refinery. Therefore, these results must be used with
caution when analyzing a specific refinery or process unit.
48
-------
VO
TABLE 4-5. PERCENT OF TOTAL MASS EMISSIONS RELEASED BY THE
UPPER1 TEN PERCENT OF SCREENED SOURCES
SOURCE
VALVES
Gas/Vapor
Light Liquid/ Two-Phase
Heavy Liquid
PUMP SEALS
Light Liquid
Heavy Liquid
COMPRESSOR SEALS
Hydrocarbon Service
Hydrogen Service
FLANGES
DRAINS
RELIEF VALVES
MINIMUM SCREENING
VALUE (ppmv)
•
11,000
12,000
120
44,000
1,500
50,000
120,000
17
1,200
4,500
95% CONFIDENCE INTERVA1
FOR PERCENT OF SOURCES
( 7
( 7
( 5
( 7
( 5
( 4
(3
( 7
( 4
( 4
, 13 )
, 12 )
, 15 )
, 13 )
, 15 )
, 18 )
, 15 )
, 13 )
, 15 )
, 18 )
PERCENT OF TOTAL EMISSIONS
I.
MEAN
98
82
77
68
60
47
41
98
78
80
90% CONFIDENCE
INTERVAL
( 97
( 78
( 69
( 63
( 47
( 31
( 23
( 97
( 68
( 65
, 99 )
. 86 )
, 85 )
, 75 )
, 78 )
, 64 )
, 57 )
, 99 )
, 90 )
, 87 )
1 The upper ten percent of screened sources is defined as the ten percent of sources
having the highest screening values.
-------
Appendix A. Emission Factor Development - Statistical Considerations
Because of the high degree of skewness in the distribution of nonmethane
leak rates from baggable sources, conventional statistics are inadequate for
efficient estimation of emission factors and their variances. In addition to
the skewness, a large percentage of the sources studied were considered "non-
leaking • These sources affect the emission factor and therefore must be
considered in developing estimates for these factors. Another statistical
problem which had to be addressed in developing the emission factors was the
estimation of leak rates for sources which screened greater than or equal to
200 ppmv but were not sampled.
A.I Estimating Emissions for Nonsampled Sources
Due to time and equipment constraints, it was not always possible to
sample all sources that screened greater than 200 ppmv. At the fifth re-
finery, a sampling strategy was developed to reduce the sampling workload.
All sources screening greater than 10,000 ppmv were sampled, but only one-
fourth of the valves and pumps with screening values between 200 and 10,000
ppmv were sampled. In order not to bias the distribution of leaking sources,
it was necessary to develop estimated values for all sources screening
greater than 200 ppmv and not sampled. The number of sources sampled and
estimated for each source type is shown in the following table:
Baggable Source Total Sources Sources Sources to
Type Sampled or Screened Sampled be Estimated
>200 ppmv
Valves 627 474 153
Pump Seals 382 281 101
Compressor Seals
Hydrocarbon Service 102 83 19
Hydrogen Service 69 60 9
Flanges 62 43 19
Drains 49 28 21
Relief Valves 58 31 27
Least-squares regression analyses were done for each device type, re-
gressing the logarithm of the nonmethane leak rate on the logarithm of the
maximum screening reading. Both the original screening value and rescreen-
ing values (taken closer to the time of sampling for leak rate) were evalua-
ted and a "best" equation was selected for each device as summarized in
Table A-l.
Using the equations in Table A-l, predicted log-nonmethane leak rates
were computed for each source not sampled with a screening value greater than
or equal to 200 ppmv. Leak rates (lb/hr) were then computed using
leak rate = explO [Log leak + z (standard error of estimate)],
A-l
-------
TABLE A-l. PREDICTION EQUATIONS FOR NONMETHANE LEAK RATES
BASED ON MAXIMUM TLV SCREENING OR RESCREENING VALUES
T
S3
SOURCE TYPE
Valves
Pump Seals
Compressor Seals:
Hydrocarbon Service
Hydrogen Service
Flanges
Drains
Relief Valves
LEAST - SQUARES EQUATION
LOG(NMLK)=
LOG(NMLK)=
LOG(NMLK)=
LOG(NMLK)=
LOG(NMLK)=
LOG(NMLK)=
LOG(NMLK)=
-5
-4
-4
-3
-5
-5
-4
.41 +
.64 +
.77 +
.66 +
.11 +
.02 +
.47 +
0.88
0.89
0.92
0.44
0.84
1.16
0.87
LOG(MKTLV-RS)
LOG(MXTLV-RS)
LOG(MXTLV-RS)
LOG(MXTLV- S)
LOG(MXTLV- S)
LOG(MXTLV- S)
LOG(MXTLV-RS)
NUMBER
OF DATA
PAIRS
177
171
48
44
47
60
53
CORRELATION
COEFFICIENT
(r)
0
0
0
0
0
0
0
.78
.68
.58
.36
.74
.72
.78
STANDARD
ERROR OF
ESTIMATE
0
0
0
0
0
0
0
.736
.820
.791
.884
.535
.807
.637
NMLK - Nonmethane leak rate (Ib/hr)
MXTLV- S - Maximum value - original screening, (ppmv)
MXTLV-RS - Maximum value - rescreening (ppmv)
LOG - Logarithm, base 10
-------
where z is a random number from a standard-normal distribution. The use of
the random number is an attempt to yield a predicted distribution of leak
rates which would approximate the distribution if all sources were sampled.
No bias correction factor is needed in going from the Log to arithmetic scale
since the mean leak rate is not being predicted. These predicted leak rates
were used in further analyses and development of emission factors.
Because the true leak rate/screening relationship is unknown, there is a
potential bias introduced when these predicted leak rates are used in devel-
oping emission factors. The potential bias is proportioned to the standard
error of the estimates divided by the number of data pairs used to develop
the equation. The impact of the bias on emission factors depends on the per-
cent of sources leaking. The potential bias was found to be less than one
percent for all source types except compressor seals and relief valves. The
potential biases were estimated to be ±9.6% of the emission factor for
relief valves, ±17.5% for compressor seals in hydrocarbon service, and ±7.1%
for compressor seals in hydrogen service. These potential biases were taken
into consideration, in developing confidence intervals 'as discussed in
Section A.3.
A.2 Statistical Distribution Models for Leak Rates
A lognormal distribution was used to model the distribution of leaking
sources. This distribution has the property that when the original data are
transformed by taking natural logarithms, the transformed data will follow a
normal distribution. The lognormal distribution is often appropriate when
the standard error of an individual value is proportional to the magnitude of
the value. The form of the lognormal distribution is as follows:
r_ (In x - y)2-i
f(x) = "PL 2cT J for 0>x>°°
X0/21T
= 0 for x £ 0
2
Mean = exp[yH j ^
Variance = exp[2y + 2a2] - exp[2y + a2]
In order to develop estimates for emission factors, the non-leaking
sources Cleak rate assumed equal to zero) also had to be modeled. A mixed
distribution, specifically a lognormal distribution with a discrete probab-
ility mass at zero, was used for this purpose. Letting p equal the fraction
of nonleaking sources in the population, this distribution has the following
form:
A-3
-------
fT \ r (In x -
(1"P)
f (x) = - 2 for 0
XCT/2TT
= p for x = 0
=0 for x < 0
2
Mean = (1-p) exp[y + — ^ ]
Variance = (1-p) [exp(2y + a2)] [exp(a2) - (1-p)]
Efficient estimates of the mean and variance of the population modeled by
this mixed distribution have been developed [Finney (1941), Aitchison (1955)],
These estimates are as follows:
The best, unbiased estimator of the population mean emission rate is
m = [ (1 - J) exp [ (x) ] g (f )
and the best, unbiased estimator of the population variance of the emission
rates is
v = [ (1 - ) exp(2x) [g(2s2) - (1 -
where n = number of sources screened,
r = number of sources screened <200 ppm or with
measured leak <10~5 Ibs/hr
m =s n_r = number of "leaking" sources,
g(t) = infinite series
_ (m-l)t (m-1)3 t2 (m-1)5 t3
m m^2! (m+1) "*" m33! (m+1) (m+3)
x = average of the logarithm of leaking sources
n-r
In (nonme thane leaks) /(.n-r)
s2 = variance of the logarithm of leaking sources
n-r
=T1 fin (nonmethane leaks) - x]2/(n-r^l)-
i
A-4
-------
The mean and variance formulas hold whenever there is more than one leaking
source (n-r >1) . When only one leaking source is identified, the following
estimates are appropriate:
2
mean = — and variance = ^!—
= —
n
n
where x, is the single measured leak. If no leaks are found (r=n), then the
best estimate for both the mean and variance is zero.
Computer programs were developed for these estimators and the estimator
for the mean was used for all emission factors presented in this publication.
Finney (1941) showed that this estimator is more than twice as efficient as
the arithmetic mean for data distributed similarly to the leak rates from
baggable sources.
Since data distributed lognormally can be transformed to a normal dis-
tribution by taking natural logarithms of the data, the distribution
assumptions can be tested by examining distributions of the Log leak rates.
Histograms displaying these distributions were constructed for all important
source type and process stream classifications. The data for most sources
appeared to adequately approximate a normal distribution. The compressor
seals data from hydrocarbon service and the heavy stream data for pump
seals both appeared skewed to the left. Compressor seals with sampled leak
rates less than 10 3 were considered as negligible (zero) to minimize this
skewness.
To statistically test the assumption of a normal distribution for the
log-leak rates, skewness and kurtosis statistics were computed for each data
group and tested for departures from their expected values of zero in a nor-
mal distribution. The following table summarizes these statistics.
Source Type/
Stream Group
Valves (all)
Gas/Vapor Streams
Light Liquids/Two-Phase
Heavy Liquids
Pump Seals (All)
Light Liquids
Heavy Liquids
Compressor Seals
Hydrocarbon Service
Hydrocarbon Service
Flanges
Drains
Relief Valves
Number of
Leaking Sources
627
200
372
35
382
300
66
99
69
62
49
42
Skewness
0.02
0.17
-0.
0.
15
37
-0.07
0.01
-0.77*
-0.65*
-0.29
0.39
-0.04
-0.22
Kurtosis
-0.03
-0.36
-0.25
-0.73
-0.30
-0.38
0.06
0.37
0.69
0.20
-0.47
-0.26
* probability <.10 given a normal distribution.
A-5
-------
Only two of the twelve cases indicated significant lack of normality,
confirming the conclusions from the histograms. It was decided that these
two cases did not merit fitting of an alternative distribution at this time.
These distributions will be addressed in the final report when other potential
correlating variables are studied.
A.3 Confidence Intervals for Percent Sources Leaking and for Emission Factors
^ •
Confidence intervals for the percent of leaking sources- were computed
using the Binomial Distribution. The Binomial is used to model data when a
random sample is selected and each item is classified into one of two cate-
gories (leaking or non-leaking here). Exact confidence limits (level 1-a)
for the estimate of percent leaking can be obtained by iteration solving for
Jji, ? | P x (1 - P )n~1 - a for the lower limit and for P in
U
f°r the
2
i-o
where n = number of sources screened and k = number of leaking sources.
Tables of these solutions, available for most cases, were used to develop
95% confidence intervals reported in this publication and for computing 97.5%
confidence intervals which were used in developing confidence intervals for
emission factors. 97.5% was selected so that 95% confidence intervals for
emission factors would result when the estimated percent leaking was com-
bined with the estimated mean leak rate (.0.975 x 0.975 - 0.95)
Patterson (1966) described how confidence intervals for the mean from a
lognormal distribution can be computed using estimators developed by Finney
(1941). 97.5% confidence intervals were computed for the average, y", of the
transformed data, y = In (.leak) , using
CT = lower limit = y - 2.24 [ s2/(n-r) ]* and
LI
h
0^ = upper limit = y + 2.24 [ s2/(n-r) P
where s2 is the variance of the transformed data and n-r is the number of
leaking sources. Then, following Patterson's arguments, confidence intervals
for the mean leak rate can be computed using:
lower limit = exp [CL]g(.s2/2) and
upper limit = exp [Cu]gCs2/2)
where g(t) is the series given in Section A. 2.
A-6
-------
To obtain 95% confidence limits for the emission factors, the confidence
limits for the percent leaking and for the mean leak rate were combined as
follows:
lower 95% limit for emission factor = PT (CT)
ij J_i
upper 95% limit for emission factor = P (C )
These confidence intervals are conservative in the sense that 95% is a
lower bound for the confidence coefficient for the intervals. The confidence
intervals should be interpreted as follows:
When we state that the true emission factor falls within the
limits computed as described above, we expect to be correct
at least 95% of the time.
The confidence intervals consider random sampling variation and random
test error, with no adjustments for potential bias in the sampling and ana-
lytical methods. Standards sampled and analyzed at most refineries have
not indicated any significant bias at this time, but some results have shown
low recoveries ( <100%) . A bias factor, if significant, will be computed
when all standards data becomes available. To account for potential low
bias at this time, the upper confidence limits were increased 5% as follows:
adjusted upper 95% confidence limit = P (Cu) + .05 (PU(CU))
In addition to the above adjustment for all source types, the upper and lower
confidence intervals for compressor seals and relief valves were adjusted to
compensate for the potential bias due to estimating leak rates as discussed
in Section A.I. The limits were expanded an additional 9-6% for relief valves,
17.5% for compressor seals in hydrocarbon service and 7.1% for compressor
seals in hydrogen service.
A.4 Development of Nomographs
Three types of nomographs were developed as part of this report:
(1) Predicting mean leak rate from screening values
(.2) cumulative percent of sources for increasing screening values
03) cumulative percent of total emissions for increasing screening
values.
This section describes how these nomographs were constructed.
A.4.1 Predicting Mean Leak Rate from Screening Values
Section A.I describes least-square linear regression equations developed
for predicting leaks from non-sampled sources in the data base with screening
values greater than 200 ppmv. For prediction purposes outside the data base,
a statistical analysis of covariance was done to determine if different
A-7
-------
equations were required for the various source types and stream groupings.
The results of this analysis showed that one equation using the maximum
rescreening value was adequate for all stream types for pumps, compressors,
drains, and relief valves. This does not imply that the leak rate versus
screening relationship is identical for these devices, but that the differ-
ences are small relative to the random scatter of the data. For valves
and flanges, separate equations were required for the gas/vapor streams
and the light liquid/tw-phase and heavy liquid streams.
The resulting equations are as follows:
I. Valves and Flanges - Gas/Vapor Streams
LOGio (.NMLEAK) = -7.0 + 1.16 LOGio (Max-rescreening)
correlation =0.71 number of data pairs = 106
standard error of estimate = 0.91 LOGio (NMLEAK)
95% confidence interval for intercept (-7.9, -5.9)
95% confidence interval for slope CO.94, 1.4)
scale bias correction factor =8.59
II. Valves and Flanges - Light Liquid/Two-Phase and Heavy Liquid Streams
LOGio (NMLEAK) - -4.8 + 0.76 LOGio (Max-rescreening)
correlation = 0.76 number of data pairs = 147
standard error of estimate =0.65 LOGio (NMLEAK)
95% confidence interval for intercept (-5.2, -4.3)
95% confidence interval for slope (0.66, 0.87)
scale bias correction factor = 3.00
III. Pump Seals, Compressor Seals, Drains, Relief Valves
LOGio (NMLEAK) = -4.0 + 0.73 LOGio (Max-Rescreening)
correlation = 0.62 number of data pairs = 168
standard error of estimate = 0.82 LOGio CNMLEAK)
95% confidence interval for intercept (-4.5, -3.4)
95% confidence interval for slope (0.59, 0.87)
scale bias correction factor =4.90
The data used to develop these equations are shown in Figures A-l through
A-3. These equations were used to develop the three sets of nomographs shown
in Figures 4-1 to 4-6 of this report. Although the equations were developed
on a logarithmic scale, the nomographs are shown on an arithmetic, scale for
ease in reading and interpolation. Predicting the arithmetic mean leak rate
for a given screening value is similar to predicting the mean from a lognormal
distribution as discussed in Section A.2. The mean value for a given
screening value on the nomograph was computed as follows.:
mean = expio I B0 + Bj LOGi o (screening)] g( Ln/2)
= (10) ° (screening value) l (scale bias correction factor)
A-8
-------
FIGURE A-l. VALVES AND FLANGES - LEAK RATE/SCREENING RELATIONSHIP
Gas/Vapor Streams
LEGEND: A * 1 DBS, B - 2 DBS, ETC.
N
0
N
M
E
T
H
A
N
E
L
E
A
K
R
A
T
E
L
a
G
L
B
S
/
H
R
C
0. B +
C
C
C
0. 0 +
E
C
[
-o. a +
c
c
c
-1. 6 +
t
c
c
-2. 4 +
r
c
c
-3. 2 +
C
C A A
t A
-4. O + A
C A
C
C A
-4. 8 *
[
I
C
-5. 6 +
C
A
D
G
B
A A AA F
B
A A C
A H
A B E
A AA C
A AC
A A D
A A A A A
C
A A
AA B B
A A A
B A
A
•A
A A
AA
A A
A A
_, + —J _J a. _ J — _ __J 4. j._ 1
1.8 2.1 2.4 2.7 3.0 3.3 3.6 3.9 4.2 4.5 4.8 5.1 5.4
MAXIMUM SCREENING VALUE AT THE SOURCE LOG (PPMV)
-------
FIGURE A-2. VALVES AND FLANGES - LEAK RATE/SCREENING RELATIONSHIP
Light Liquid/Two-Phase Streams
LEGEND: A = 1 DBS, B = 3 DBS, ETC.
N
O
N
M
E
T
H
A
N
E
L
E
> A
,L K
!—*
R
A
T
E
L
0
G
L
B
S
/
H
R
C
0. 8 +
t
t
C
0. 0 +
C
C
C
-0. B +
C
C
C
-1. 6 +
C
t
t
-2. 4 +
C
t
C
-3. 2 +
r
c
c
-4. 0 +
c
c
c
-4. 8 +
r
r
c
-5. 6 +
A
B
A A
A A B
A A
A B D
AA ED
B A A B A BB
A A B BB A E
A A B BA AA BA
A A B AAA A
A A B A A
A A B AA
A A
A A A A
C A B
A AA A
A
A
A A
B
F
A A
A
A C
A AD
A B
AA C
B
A A AA
AAA
A A
A
0. O 0.4 0.8 1.2 1.6 2.0 2.4 2.8 3.2 3.6 4. O 4.4 4.8 5.2
MAXIMUM SCREENING VALUE AT THE SOURCE LOG (PPMV)
-------
FIGURE A-3. PUMPS, COMPRESSORS, DRAINS, RELIEF VALVES LEAK VS SCREENING
All Process Streams
LEGEND: A = 1 DBS, B = 2 DBS. ETC.
N
0
N
M
E
T
H
A
N
E
L
E
A
K
R
A
r
E
L
0
G
L
B
S
H
R
C
1.6 +
C
C
I
0. 8 +
C
C
C
0,0 +
C
C
C
-O. 8 +
C
C
C
-t.fe +
r
c
t
-2. 4 +
t A B
L
C A
-3. 2 +
C A
C
C
-4. 0 +
C
C
-4. 8 +•
t
1 j
X
A
B
B B C
B
B A AD
A A A A AD A
A AA C
A A AA B C A A
B AA A A D
A BA A C A AA
AAA AABCAABAAAA D
A ABB BAAB
A BA AAA A AAA
A A A AB A AA AC AA A B B A A
AA AABA A A
A A AA A BAA AAA A
A A BB A C A A
A A A A B AA
A A D A
A B AA A
A A
A A
A A A
A
1 4- H h h + 4. 1. 4. j
A
B
C
C
G
M
D
H
G
A A F
BBB A H
AB AA F
B F
C B
B
B
MAXIMUM SCREENING VALUE AT THE SOURCE LOG (PPMV)
-------
where B0 - Log regression intercept
BI - Log regression slope
SET - standard error of estimate in natural Log scale
g It)- series described in Section A. 2.
90% confidence intervals for the predicted mean leak for a given screening
value were computed in a similar manner to the confidence intervals for the
mean leak rate as described in Section A.2.
A.4.2 Cumulative Percent of Sources for Increasing Screening Values
The nomographs in Figures 4-7B to 4-16B of the report contain the
estimated cumulative distribution of Log screening values. The nomographs
show 100% minus the cumulative percent, or the estimated percent of sources
which would have screening values greater than any particular screening value.
These cumulative distribution functions were estimated by fitting a lognormal
distribution, as described in Section A.2, to the screening data and then
generating the cumulative distribution.
There was some difficulty in fitting the lognormal distribution to the
screening values. Figure A.4 shows a typical histogram of Log screening
values for valves from gas streams. The histogram appears to approximate
a normal distribution adequately up to 10,000 ppm (4.0 on LOGio scale).
The spike at 10,000 ppm was due to the inability of the screening device
to measure beyond 10,000 without a dilution probe. The dilution probe was
used in only a few cases in the screening process during this program.
To overcome the bias caused by this spike, only Log screening values
less than 4.0 were used to estimate the parameters of this distribution.
Formulas from "censored" normal distribution theory (Cohen, 1959) were then
used to arrive at unbiased estimates of the entire distribution. These es-
timates were used to generate the cumulative distribution function for each
source type/process stream grouping.
Confidence intervals for these cumulative functions were obtained using
the Binomial Distribution as in Section A.3. The 95% confidence interval
for individual probabilities were approximated using
e ± 1.96 tp ci-&) /nf2
where p is the estimated cumulative percent and n is the number of screening
values for each particular source type and stream group.
Assuming that the sources screened approximate a random sample from
the population of a particular source type, these confidence intervals can
be interpreted as follows:
When we state that the true percent of sources in the population which
have screening values greater than any selected screening value lies within
the confidence bounds, we expect to be correct about 95% of the time.
A-12
-------
w
S3
i.
i.
i.
0.125
0.375
0.625
0.875
1.125
,375
,625
,875
2.125
2.375
2.625
2.875
3.125
3.375
3.625
3.875
4.125
4.375
4.625
4.875
0.374
0.624
0.874
1.124
1.374
1.624
1.874
2.124
2.374
2.624
2.874
3.124
3.374
3.624
3.874
4.124
4.374
4.624
4.874
5.124
20
30 40
NUMBER OF VALVES
50
60
70
80
90
Figure A-4.
DISTRIBUTION OF LOG 10 (MAX SCREENING VALUE)
Valves - Gas/Vapor Streams
-------
Note that these limits apply to the entire population for a source type and
are not necessarily applicable when addressing a particular situation con-
cerning a small number (say less than 100) of sources.
The estimated cumulative distribution functions were compared with the
sample distribution function and appeared to fit the data for each case
except compressor seals. Discrepancies were found at the 10,000 ppmv
screening value, in almost all cases, but this was to be expected since
the sample function had a big jump at this point.
For compressor seals, censoring the data at 10,000 ppmv eliminated
64% of the observations, so the lognormal parameters were reestimated
using all the data as recorded. These estimated resulted in a "better"
agreement between the sample and estimates distribution function, and
were therefore used to generate the cumulative distribution function for
compressor seals in both types of service.
A.4.3 Cumulative Percent of Total Emissions for Increasing Screening Values
The nomographs in Figures 4-7A to 4-16A of this report contain a
function estimating the cumulative percent of the total emissions attributable
to each particular source type/stream group as a function of increasing Log
screening values. As before, 100% minus the cumulative function is shown
so that the percent of total emissions attributable to sources with
screening values above any selected screening value can be determined.
This cumulative function was estimated by integrating the leak/screening
regression relationship over a lognormal distribution of screening values.
This function has the following form:
exp [-(In x -u)2J dx
2a2
where So - selected upper screening value for integration
C - Log /arithmetic scale bias correction factor
BO - Log 10 regression intercept term
BI - Logio regression slope term
u - mean of the Log (screening values)
a2 - variance of the Loge (screening values)
x - screening values over which the integration is being done
CF - cumulative function described above in Ibs/hr.
The form of the cumulative function can be simplified by algebraic
reduction and change of variables to obtain.
[- "
L
= C(10)B° exp " - (" + Bia)2 1 LOG (S0) - u -
2a2
A-14
-------
where $ is the cumulative function of a standard normal distribution.
This function was used in developing the cumulative emissions function
shown on the nomographs. The censored distribution parameter estimates
described in Section A.4.2 were used for the lognormal distribution para-
meters in each case except compressor seals. The Log/Log least-squares
regression estimates described in Section A.4.1 were used for the scale
bias correction factor and for B0 and BI.
The scale for the above cumulative function is in Ibs/hr. To obtain
a cumulative percent function, the number obtained in Ibs/hr at each
screening value was divided by the value of the function at a screening
value of one-million ppmv. This forced the cumulative function to 1.0 at
one-million ppmv. These scaled values were then subtracted from 1.0 and
multiplied by 100.0 to obtain the functions shown on the nomograph.
The estimated cumulative emissions functions were compared with the
sample functions and found to adequately approximate the data in most
cases. Again, the biggest discrepancies were near the 10,000 ppmv screening
value where the sample function has a big jump. This area is more critical
for this function than the cumulative distribution function since most of
the emissions are attributable to sources with screening values greater
than 10,000 ppmv. It is important to note that very little screening data
is available with screening values greater than 10,000 ppmv. Thus, this
portion of the curve is based on extrapolations using models developed
from screening values less than 10,000 ppmv.
This cumulative function is a very complex nonlinear function of five
sample statistics:
j
(1) the intercept and slope from the regression of Log leak rate
on Log screening value,
(2) a bias correction factor used when converting the logarithmic
to the linear scale, and
(3) the mean and variance of the natural logarithm of the screening
values.
Due to the complexity of this function, it was not possible to derive a
closed-form analytical expression for the confidence intervals. Thus, a
Monte-Carlo computer method was used to generate the confidence intervals.
This method involved regenerating the cumulative function 400 times.
Each time, the data collected in the project (the exact number of sources
with screening values greater than zero) were regenerated, except with an
independent set of random variations. The distributional properties of
the leak rate and screening data were used in computing the required random
numbers.
A-15
-------
For each of the 400 trials, sample estimates of the five parameters required
to compute the cumulative function were computed. Then these estimates were used
to generate a new cumulative function. The five percent lower result and the
five percent upper result from the 400 trials for any given screening value
were then selected as the 90% confidence limits for the population cumulative
function. These approximated 90% confidence limits can be interpreted as
follows:
When we state that the true percent of total emissions, for
the population of sources, attributable to sources with screening
values greater than a selected valve, is within the confidence
bounds, we expect to be correct about 90% of the time.
Since these confidence limits address the uncertainty in the cumulative fun-
ction for the entire sampled population of a particular source type, they are
not necessarily applicable to a finite sample of sources in a particular sit-
uation. The variation of this function depends on the number of sources in
a complex manner, so it is not possible to draw a general conclusion for the
effect of sample size. Monte Carlo simulation techniques can be used to
approximate intervals for a finite random sample of a particular source type.
As an example, Figure A-5 shows the confidence intervals for the
cumulative percent of emissions functions for valves in light liquid/
two-phase service. Intervals are shown which are applicable to a random
sample of 100 valves and a random sample of 1000 valves. Also included
are the confidence intervals for the entire population (from Figure 4-8A).
As can be seen, the intervals applicable to a finite number of sources are
significantly wider than those for the population.
These intervals for finite populations were also developed using simu-
lation techniques. Four hundred Monte Carlo trials generating 100 sets of
data and 400 trials generating 1000 sets of data were run. In each of the
trials, the generated sample was ranked according to screening values and a
sample cumulative leak rate function computed. Each sample function was
scaled by dividing by the total leak rate generated. Then the five percent
lower result and the five percent upper result from the 400 trials for any
given screening value were selected as the 90% confidence bounds.
These confidence intervals can be interpreted as follows:
When we state that the cumulative percent of total emission
function, which would be generated from a random sample of "m"
sources, will fall within the confidence bounds, we expect to be
correct about 90% of the time.
where "m" is the number of randomly selected sources in a particular situation.
A-16
-------
Valves - Light Liquid/Two-Phase Streams
100
90
80
70
VI
! 60
LU
£ 50
i° 40
<«_
o
c
01
\
- Estimated Percent of \
Total Mass Emissions
---- 90% Confidence Interval \
\ \\\
for Percent of Emissions \ x \ V
from Total Population of \ \ \\\
Valves (n = ») \ \ \\\
30
(n = »)
•90% Confidence Interval
for Percent of Emissions
in a Random Sample of
1000 Valves
90% Confidence Interval for
Percent of Emissions in a
Random Sample of 100 Valves
OJ
°- 20
10
0
LJ I
1 2345 10 50100 1000 10,000 100,000 1,000,000
Screening Value (ppmv) (Log1Q Scale)
Figure A-5. Cumulative Distribution of Total Emissions by
Screening Values - Valves - Light Liquid/Two-
Phase Streams - Comparison of Confidence
Intervals
A-17
-------
REFERENCES
Aitchison, J., "On the Distribution of a Positive Random Variable Having a
Discrete Probability Mass at the Origin", American Statistical
Association Journal. 50. C9), 1955, 901-908.
Cohen, A.C.,Jr. "Simplified Estimators for the Normal Distribution when
Samples are Singly Censored or Truncated", Technometrics 1,
(1959) 217-237.
Finney, D. J., "On the Distribution of a Variate Whose Logarithm is Normally
Distributed", Journal of the Royal Statistical Society. Series B,
7 (1941), 155-161.
Patterson, R. L., "Difficulties Involved in the Estimation of a Population
Mean Using Transformed Sample Data:, Technometrics 8, No. 3, (1966),
535-537.
A-18
-------
TECHNICAL REPORT DATA
Iflease read Instructions on the reverse before ci
EPA- 600/2 -79-044
2.
3. RECIPIENT'S ACCESSION NO.
SUBTITLE
Emission Factors and Frequency of Leak Occurrence
for Fittings in Refinery Process Units
5. REPORT DATE
February 1979
6. PERFORMING ORGANIZATION CODE
8. PERFORMING ORGANIZATION REPORT NO.
Robert Wether old and Lloyd Provost
*. • -- ORGANIZATION NAME AND ADDRESS
Radian Corporation
P.O. Box 9948
Austin, Texas 78766
10. PROGRAM ELEMENT NO.
1AB604
11. CONTRACT/GRANT NO.
68-02-2147 and -2665
12. SPONSORING AGENCY NAME AND ADDRESS
EPA, Office of Research and Development
Industrial Environmental Research Laboratory
Research Triangle Park, NC 27711
13. TYPE OF REPORT AND PERIOD COVERED
Interim: 3/76 - 3/79
14. SPONSORING AGENCY CODE
EPA/600/13
15'SUPPLEMENTARYNOTESIERL-RTP project officer is Dale A. Denny, MD-62, 919/541-2547
i6. ABSTRACT Tne report gives results of sampling fugitive emissions at nine integrated
oil refineries throughout the U.S. The petroleum refining industry is a significant
source of atmospheric hydrocarbon (HC) emissions in the U.S. Each refinery has a
large number of potential emission sources, both controlled (e.g. , stacks and vents)
and uncontrolled (e.g. , leaks). HC emission data were collected for valves, flanges,
pump and compressor seals, pressure relief valves, and process drains. The sam-
pling techniques are presented. Potential leaking components were initially screened
using portable HC detectors; screened devices which indicate signifleant emissions
were then subjected to fully qualitative and quantitative sampling and analysis. For
the nine refineries, 5680 sources were screened, 1250 of which were sampled and
analyzed. Data on non-methane HC emission rates are presented for each fugitive
source, with statistics on data variability. Information on frequency of leaks is also
provided. These data show that most HC emissions from fugitive sources occur due
to a relatively few leaking components. Gas/vapor streams contribute a proportion-
ately greater amount of emissions than the light and heavy liquid streams.
17.
KEY WORDS AND DOCUMENT ANALYSIS
a.
DESCRIPTORS
b.lDENTIFIERS/OPEN ENDED TERMS
c. cos AT I Field/Group
Pollution Sampling
Petroleum Refining Analyzing
Hydrocarbons
Leakage
Flue Gases
Vents
Pollution Control
Stationary Sources
Fugitive Emissions
Nonmethane Hydro-
carbons
13B
13H
07C
14B
21B
13A
8. DISTRIBUTION STATEMENT
Unlimited
19. SECURITY CLASS (ThisReport)
Unclassified
21. NO. OF PAGES
78
20. SECURITY CLASS (Thispage)
Unclassified
22. PRICE
EPA Form 2220-1 (9-73)
A-19
-------
ACKNOWLEDGMENT
Mr. C. D. Smith and Dr. D. D. Rosebrook of Radian Corporation are
also co-authors of this report.
The authors wish to acknowledge the assistance of Dr. Dale Denny,
Dr. I. A. Jefcoat, and Dr. Bruce Tichenor of the Environmental Protection
Agency under whose guidance this program has been carried out.
We wish to thank the members of the Ad Hoc Advisory Panel of the
American Petroleum Institute. Their assistance in the formulation of the
program and their advice during its duration are greatly appreciated. The
authors especially wish to thank Mr. Edward P. Crockett of the American
Petroleum Institute for his considerable efforts in support of the project.
We are grateful to Mr. Herbert W. Bruch of the National Petroleum
Refiners Association for his substantial assistance in this program.
The data upon which this report is based were obtained at a number of
refineries throughout the country. The assistance and exceptional cooperation
of the staffs of these refineries is gratefully acknowledged.
It would be impossible to individually thank everyone on the Radian
staff who participated in this program. Their outstanding attitude and
dedication have made this project successful.
ii
------- |