Quality assurance for emissions analysis systems 1
EPA has established a repository of 37 gaseous organic compounds for use in source test performance audits. So far, eight have been evaluated statistically R.K.M. Jayrnty Corette B. Parker Clifford F. Decker William F. Gutknecht Research Triangle Institute Research Triangle Park, N.C. 27709 Darryl 1. von Lehmden Joseph E. Knoll Environmental Monitoring Sysfems Laboratory V.S. Environmental Protection Agency Research Triangle Park, N.C. 2771 I The application of quality assurance practices is having a significant, positive impact on the quality of environmental data being generated. One such practice is the performance audit, which involves one laboratory providing “unknown” or “blind samples to another laboratory for analysis. A performance audit conducted simultaneously with a source emission sample analysis provides an assessment of analysis accuracy. This accuracy assessment is important for data use in setting standards and for compliance testing after standards have been promulgated. When it has been demonstrated that specific organic gases are stable enough to be used as performance audit materials in the 5-20ppm concentration range, a performance audit will be included as an integral part of the promulgated test method. Two EPA test methods currently include the requirement for a performance audit during source test analyses-Method 106 for vinyl chloride and Method I IO for benzene (1). 0013-936X/83/0916-0257A$01.50/0
L
/’
GCcalibrstion. This chemisr is obtaining dara by rhe “glass bulb” rrchnique Under a contract to the EPA, the Research Triangle Institute (RTI) is developing an extensive repository of gaseous compounds for use in audits, as designated by EPA project officers. The RTI repository currentlycontains 37 different comwundsselected on the basis of anticipated needs of EPA’s Office of Air Quality Planning and Standards. Table 1 lists the compounds in the repository, their concentration ranges, the number of cylinders of each compound, and the cylinder construction material. The audit materialsare in two concentration ranges. The low concentration range of 5 to 20 ppm covers the range of possible future emission standards. The high concentration
0 1983 American Chemical Society
range, 50-700 ppm, covers actual source emission levels found prior to the installation of control devices. The balance gas for all gas mixtures is nitrogen. I n some cases, a compound that serves as an internal standard is also added to the cas mixture. Other compounds will bi added to the repository as needed. The gaseous compounds in Table 1 are purchased in compressed gas cylinders from commercial suppliers. These cylinders, along with an appropriate delivery system, are used directly without dilution in the performance audits. The compressed gas cylinder is especially suitable as an audit material for a number of reasons including simplicity, portability, and Envimn. Sci. Technol.. Vol. 17. No. 6. 1983
2571
low cost; blind mode of operation (that is, the user cannot readily awrtain the relative or absolute concentration levels); and reliability and ruggedness for interstate shipping. Experimental procedures RTI analyzes the contents of the cylinders when they are first received, in order to check the manufacturer's reported values. If the RTI-measured values are not within 10%of the supplier's values, they are measured by a third party, usually the EPA. The cylinder contents are analyzed several more times over the course of a year to estimate stability. All measurements are made by gas chromatography (GC). Cylinders in the repository are sent to laboratories and field sites in relRLYT
sponse to audit requests received from the EPA. RTI reanalyzes the contents whenever the cylinders are supplied for use in these audits. These latter analyses provide values for comparison with the audit results and alsoallow a better estimate of the stability of the compounds in the cylinders. Instrumentation.Analyses are performed with a Perkin-Elmer Model 3920 B gas chromatograph with a flame ionization detector, a PerkinElmer Sigma 4 gas chromatograph with flame ionization and electron capture detectors, and a Tracor Model 560 gas chromatograph with a flame photometric detector. The Tracor instrument has been used principally for measurement of sulfur-containing species. A moderately polar column (Alltech Associates, 10% OV-IO1 on
Chromosorb WHP) is used for measurement of most of the hydrocarbons. Other columns used include Durapak n-octane on Porasil C (Waters Associates. Inc., nonpolar), 0.1% SP-1000 on Carbopack C (Supelco, Inc., moderately polar), and 0.4% Carbowax 1500 on Carbopack C (Supelco, Inc., polar). Column and detector temperatures used for each compound are described elsewhere by Jayanty et al. (2). Gaseous samples are injected onto the columns by means of manual gas sampling valves. These valves are equipped with interchangeable sample loops to allow the injection of fixed volumes of gas. Another system permits the injection of variable sample volumes. The operation of the system is based
1
Audit materials currently held in the repository
Benzene Ethylene
14 4
8-13 5-20
S AI
4
5-20
AI
4
-
4
17 4
6 Propylene Methanelethane
-
-
60-400 300-700 3000-20000 300-700 1000-6000 (methane) 200-700 (methane) 300-700
Propane
4
5-20
AI
4
Toluene Hydrogen sulfide Meta-xylene Methyl acetate Chloroform
2 4 2 2 2
5-20 5-20 5-20 5-20 5-20
S AI
2 2 2 2 2
300-700 300-700 300-700 300-700 300-700
2
5-20 3-10 20-80 5-20
S AI AI AI -
2
100-300
Carbonyl sulfide Methyl mercaptan Hexane 1.2-Dichloroethane Cvclohexane Methyl ethyl ketone Methanol 1.2-Dichloropropane Trichloroethylene 1.1-Dichloroethylene 1.2Dibromoethylene
4
2 4
-
-
S
S S
S AI AI AI AI AI
-
2 4 1
1000-3000 100-600 80-200
-
-
1 1 2 2 2 2
30-80
2
5-20 5-30 5-30 5-20
S LS
5-20 5-20 5-20 5-20 5-20
AI AI AI AI AI
5-20 5-20 5-20 5-20 5-20
AI AI AI S AI
-
1
75-200 75-200 75-200
Chlorobenzene 1 5-20 Carbon disulfide .AI = aluminum. S = steel. LS = low-pressure steel
AI AI
-
75-200
Perchloroethylene Vinvl chloride 1.3:Butadiene Acrylonitrile Methyl isobutyl ketone Cyclohexanone Methylene chloride Carbon tetrachloride Freon 113 Methyl chloroform Ethylene Oxide Propylene oxide Allyl chloride Acrolein
9
1
2 1
2 1 1
1 1 1
1 1 1
30-80 5-20 5-20 5-20 5-20
~~
S S
-
AI. S AI AI AI AI AI
S AI LS S S S
LS AI S
-
-
2 2 2 2
300-700 100-600 100-600 100-600
LS AI AI AI
2
300-700
LS
-
-
-
2
300-700
LS
1
50-100
AI
-
1
1
-
-
AI S AI
-
AI
on measurement of pressure differentials. The sample loop, the volume of which is known, is first evacuated. Then, after the vacuum source is closed off, the sample is introduced into the loop (the final pressure after sample introduction may vary from subambient to superambient). The pressure change with sample introduction is measured with a very accurate gauge, and the mass of sample introduced is then calculated according to the ideal gas law. Thus, calibration curves can be developed easily and quickly with gas from only one cylinder of known concentration. Other cylinders of higher concentrations can then be made to fall within the bounds of the calibration curve. Standardization and measurement. Several different forms of standards are used for calibration. The preferred standards are National Bureau of Standards (NBS) Standard Reference Materials (SRMs). Unfortunately, SRMs are available for only a few organic gases. The SRMs presently being used are methane and propane in nitrogen in compressed gas cylinders. If the compound to be quantified is not an available SRM but is of the same homologous series (such as alkanes or alkenes), a flame ionization detector is used, and calculations are based on relative response factors, which are related to carbon number. .
This procedure is described in the literature by Dietz (3). A second calibration method uses commercial permeation tubes. Such tubes are used for vinyl chloride and perchloroethylene. The tube is placed in a temperature-controlled chamber and dry "zero" (hydrocarbon-free) air is passed over the tube at a known flow rate. The resultant gaseous mixture is further diluted if necessary with additional dry hydrocarbon-free air in a glass dilution chamber. The final mixture is then passed into a gas sampling bulb for GC measurement or is drawn directly into the GC sample injection valve by a small pump downstream. The permeation tubes are not used until their permeation rates, which are determined periodically by measured weight loss, have been well established. A third method ofcalibration is the ''glass bulb" technique in which a known volume of the pure compound, either gas or liquid, is injected into an evacuated glass bulb or flaskof known volume. The flask is then returned to atmospheric pressure with a balance gas of choice. If a pure liquid is injected, total vaporization is assumed and the concentration is determined by applying the ideal gas law. Usually a minimum of IO /IL is injected to achieve acceptable volumetric accuracy. If 10 pL of the pure liquid results in too high a concentration. a solution
. .
.
.
of the analyte and some noninterfering solvent will be injected. Dilutions can be made by using additional bulbs or by pressurizing with a balance gas to a known pressure. Calibrations not performed with NBS SRM cylinder gases or commercial permeation tubes are performed by the glass bulb technique. With each of these standards, multipoint standard curves are prepared each time a sample is analyzed. Certaln quality control procedures are followed; for example, the permeation system and the glass bulbs are equilibrated with the sample gas before an aliquot is taken for GC measurement. Also, the glass bulbs used are cleaned thoroughly before each use by repeated evacuation. An NBS standard cylinder of methane is analyzed to verify the constancy of the detector response. Because of condensation effects at elevated concentrations and pressures, several compounds are contained in low-pressure steel cylinders. However, reduced delivery concentrations may still be encountered, especially if these cylinders are exposed to cold temperatures. RTI uses a procedure developed by Scott Environmental Technology, Inc., that will return the delivery concentration of a low-pressure cylinder to the true value. This procedure involves slow, mild heating (moderately warm to the touch) and
..
-P --
..
.
.
,--
TABLE 2
Sample cylinder analysis data: benzene NO. Cyllndn SolmtNcilon
1A AI
B AI
1c
10 AI
1P
AI
10 S
1R S
Manufacturer concentration
65.4
324
200
117
8.04
9.85
9.89
hlinc+,
Date Day (ppm)
RTI concentration
7/27/77 (79.0) 136 (74.0) 156 (78.0) 167 (80.0) 630 (77.9) a
% changelmonth
Two std.dev. of
7/27/77 (374) 136 (337) 156 (350) 167 (355) 402 (331) 433 (343) 969 (358) 1274 (348) 1491 (324)
-0.01 0.42
7/27/77 (241) 247 (216) 252 (215) 381 (218)
-0.10
0.18
s
7/27/77 (138) 29 (144) 157 (134) 252 (129) 290 (127) 414(127) 1247 (132)
-0.87
0.66
4/21/78 (8.37) 4 (8.33) 25 (8.20) 26 (8.34) 56 (8.19) 134(7.81) 434 (8.21) 766 (7.93) 1222 (7.68)
-0.12 0.27
-0.16 0.10
4/21/78 (9.99) 5 (9.88) 25 (10.1) 332 (9.71)
-0.25 0.26
4/21/78 (10.0) 4 (10.1) 13 (9.73) 332 (9.77) 1018(9.46) 1270 (9.64)
-0.10
0.08
Yo channdmnnthb AI = alumlnm. S = ateel
'E W Y
a Criteria facalcuiaticn are mat 81 east t a r measwerents h ~ v been e w a r n e dand mat at b s t a oneyear span BXISIS between uw first end last analysis. Analyiical condnions: Flame kmization detecta. 10% ov-101 on chmmosab W column a100 'C. Calibration: Reagentgade benzene liquid is s 3 d 89 a standard. Glass-bdb dilution technique is used fameking UW series of standards tor calibre
II i
IiO".
Envkan. Scl. Tsohnol.. Vol. 17. No. 0. 1983
BOA
vigorous horizontal rolling of the audit gas cylinder before sampling. Audit gas reanalysis and data treatment. Analysis results collected over time are tabulated and reported in an annual project status report (2). The data included in a typical page of this report are shown in Table 2. This table shows cylinder numbers as assigned by RTI, cylinder construction material, and concentration reported by the manufacturer upon delivery of thecylinder to RTLThedateshown is the date the cylinder contents were first analyzed by RTI. The “day” values are the number of days after the first analysis by RTI, and the values in parentheses are the ppm concentrations determined on each particular analysis day. The usual analysis frequency for all audit materials is: immediately when received from the gas manufacturer, four weeks after the first RTI analysis, eight weeks after the first RTI analysis, one year after the first RTI analysis, and immediately before any request that would result in the use of the cylinder gas in a performance audit. The stability of the contents of each cylinder is estimated as percent change in concentration per month. Such an estimate is made only when a minimum of threeanalyses have been per-
formed on the contents of a particular cylinder. A linear regression analysis of the data is performed, with x as days and y as ppm, to determine this value. Then, percent change per month is calculated as:
70change/month =
X I00 X 30 d b where b = intercept and m = slope. The standard deviation of the percent change per month is calculated when a minimum of four analyses have been performed. This calculation involves the use of conventional statistical methods. Standard deviation values are doubled and presented at the bottom of Table 2. It can be argued that a linear model may not be appropriate for all compounds. That is, some compounds may change rapidly soon after they are prepared, and then change only slowly, or not at all after that. The analysis frequency, however, is determined primarily by the request for the audit material for performance audit. Problem compounds. Several of the compounds prepared as gases in cylinders were found to be unsuitable for use as audit materials. Among these were aniline, cyclohexanone, paradichlorobenzene, and ethylamine, which all have relatively high boiling points. Unacceptable variability in results of analyses of these compounds was noted, which apparently was attrib-
utable to losses to the regulators, sample transfer lines, valves, and soon. These problems could be reduced by the use of heated systems, but such systems are impractical for audits. Further work is in progress to determine the actual reasons for this analytical variability. One additional compound, formaldehyde, was ordered, but the specialty gas manufacturer indicated that cylinders of this compound in the gaseous state could not be prepared because of its instability at room temperature. Performanceaudits As stated, RTI supplies repository cylinders for audits upon request from the EPA. These requests are sometimes directed through the EPA from state and local government agencies or from their contractors. However, the contractor must be performing source emission tests on behalf of the EPA or one of the other agencies to qualify for the performance audit. When a request is received, the contents of the cylinders to be shipped are analyzed, the cylinder pressures are measured, and the cylinders are shipped by truck to the audit site. Cylinder gas regulators also are provided when requested. General instructions for performance of the audit are included with the cylinders. The audit results are reported to the EPA or the requesting agency audit co-
Evaluation procedure In the stability evaluation process, linear regression models are fit to the analysis data collected over time. These models allow data to be pooled so that, for example, ail benzene results can be tested collectively. Stability tests are under way on all 37 compounds in the repository. For eight cwnpwnds. sufficientdata are available to permit a meaningful statistical evaluation for stability. However, the other 29 compounds have not been analyzed often enough to permit meaningful statistical tests. The first linear regression model used to evaluate the compounds is one that fits a common slope for ail cylinders and a distinct intercept for each cylinder of a particular compound. The model is: k$’ = a!‘)
+ bcC)X+ e!’:
(1)
where fip’ is the concentration (in ppm) for compound C for cylinder i at day x; a? is the intercept for corn pound C for cylinder i. with concentration at time zero: bee) is the slope for
260l
Envim. Sci. Tednol.. Vol. 17. No. 6. 1983
compound C.Le.. change in concentration (A ppm) with chacge in time (A d); X i s time in days; and e:’ is a statistical error term. These error terms are assumed to be independent, identically distributed, random variables for all cylinders wim distribution NO, a2). The second model allows a unique slope and intercept for each cylinder i; that is: r(C’ = a\c)+ b!C’X + ,“ (2) where b\’) is the slope for compound C for cylinder i. The third model allows unique
slopes for cylinders within concentration levels designated as high, medium, or low, or just high and low. This model is: v”, = + b‘CIX + elCI 1(1* I l0lX (3) where ;9 ; is the concentration (in ppm) at day x for compound C for the cylinder i of concentration level i: b\” is the slope for compound C for concentration level i; and e!$ is the error term.
In addition to the above, a model that allows for different slopes for aluminum vs. steel cylinders was investigated for benzene. The model is:
fi& = a\’’ + blf’X + e\& (4) where fi& is the concentration (in ppm) for compound C for cylinder i of
concentration level ’ and cylinder mteriai k at day x; bid is the slope for compound C for concentration level i of cylinder material k: and e$&. is the error term. The purpose of fitting the collected analytical data to these modeis is twofold. The most important objective is to determine whether or not the calculated slopes. that is, changes in concentration (ppm) with time (d). are significantly different from zero, using as a statistic the ratio of the slope to its standard error. The second purpose is to determine whether there are differences in stability between concentration ranges or between types of cylinders.
ordinator. Todate, 86 performance audits have been completed. Representative audit results are presented in Table 3. As noted previously, high- and low-level concentrations were included in most audits. The average audit result bias, that is, (client-measured concentration - RTl-measured concentration)/ (RTI-measuredconcentration), for the 86 audits was -0.3% with a standard deviation o f f 10.5%.This low average is somewhat fortuitous in light of the value of the standard deviation. Audit results ranged from -24.8% for a hydrogen sulfide measurement to +32.5% for a toluene measurement. The principal results of these audits have been to assure all concerned that analyses are being performed properly, to uncover problems that could be corrected, and to document the accuracy of the source emission test analysis. Stability study The data collected over time from the determinations of cylinder content concentrations are being used to estimate stabilities of the repository gases. Cylinder gasstabilitydata are important for several reasons. First, audit materials used by the EPA must be stable, or they are unacceptable as audit standards. Second, if organic gases in cylinders are stable, other investigators may more readily use cylinder gases as calibration standards or quality control check samples. Finally, if organic gases in cylinders are stable, future EPA regulations may require performance audits as a means of assessing the accuracy of analysis data. Fitting models to data. The fit of the various models to the data for eight compounds was tested. Only benzene was tested with all four models. Though both high and low levels of hydrogen sulfide are included in the repository, only low levels were included in this stability test because of the small number of measurements made of the high levels. The mean square errors resulting from the fit of the data to each of the mathematical models were determined. Statistical tests based on analysis of variance procedures were used to assess the differences between each more complex model and a simpler one ( 4 ) . Results of these tests are listed in Table 4. As noted, probability @) values
