Comparison of Exhaust Emission Measurements
Hy the Federal Register Procedures
and by a Beckman 6800
Gas Chromatograph
by
Glenn D. Thompson
July 19, 1973
Environmental Protection Agency
Office of Air and Water Programs
Mobile Source Air Pollution Control
Emission Control Technology Division
Procedures Development Branch
Ann Arbor, Michigan
-------
Comparison of Exhaust Emission Measurements by the
Federal Register Procedures and by a Beckman 6800
Gas Chromatograph.
ABSTRACT
This report compares measurements of exhaust emission
components made using a Beckman Gas Chromatograph Model
6800 and the Standard Federal Register instrumentation.
The Beckman Gas Chromatograph Model 6800 as modified for
this comparison is a single instrument capable of measuring
total hydrocarbons, methane, CO and CC>2 •
In this comparison test, bag samples of CVS collected
exhaust from 1975 prototype vehicles were analyzed by both
the Federal Register instrumentation and the Beckman Gas
Chromatograph. Very good correlation between the two in-
struments was observed for the total hydrocarbon measure-
ments. The CO measurements showed some variations between
instruments, particularly in the high concentration region.
The CO2 measurements showed apparent random fluxuations
larger than either the total hydrocarbon or the CO random
fluxuations.
It is believed much of the observed measurement vari-
ations were a result of less frequent use and less rigorous
calibration of the Beckman instrument. If the Beckman
instrument was maintained in the manner of the Federal
Register certification instrumentation, comparable accu-
racy and repeatability could probably be achieved. There
are no EPA data, however, to experimentally verify this
at present.
-------
INTRODUCTION
In 1972 the ECTD Methane Exclusion Study was started
to determine the percentage of methane in 1975 prototype
automobile exhaust emissions. For this study total hydro-
carbons, methane, CO, and CO2 were measured from bag samples
of CVS diluted exhaust using a Beckman model 6800 gas chrom-
atography The same bag samples were also analyzed by the
Standard Federal Register procedures for total hydrocarbon,
CO, CO2/ and NO . It is the intent of this report to review
this data with ?he purpose of comparing the two instrument
systems.
DATA
From November 1972 to March 1973 a series of thirty
three, LA-4 cycles, three bag tests were run; however,
one test was voided. The bag samples from the remaining
32 tests were analyzed on the Federal Register instrument
train #16 and on the Beckman Gas Chromatograph* instrument
train #42. This yielded 96 measurement pairs for each
HC, CO, and CO-. In some instances an apparent error or
instrument malfunction had occurred. This was usually
characterized by a near zero reading for one of the
emissions on one of the instrument trains. It was
arbitrarily decided to delete any data where the two
values for any one emission component differed by more
than a factor of two. This resulted in a deletion of
13 data points and a remaining data field of 83 points
for each HC, CO and CO2.
In order to better visualize the data a scatter
plot of each measured component was printed. These plots
appear in Appendix I and II. It should be noted the program
prints an astrix each time one or more of the data points
fall in the computer assigned bins; therefore any astrix
may represent more than one data point. From these scatter
plots several points on the CO and CO2 plots were observed
to deviate from the obvious grouping of the remaining
points. It was decided to delete 7 more data points and
then calculate statistics on both the 83 point data field
and on the reduced 76 point data field. The reduced data
should give better estimation of the experimental parameters
since the calculations would not be affected by a few points
greatly deviating from the majority; however the statistic
tests of the parameters calculated from the 76 point data
field would be overally optimistic since all "bad" data had
*
Mention of company or product names is not to be considered
as an endorsement by the Environmental Protection Agency.
-------
-2-
been removed. The data and all calculations pertaining
to the 83 point data field are given in Appendix I, while
the reduced data and those corresponding calculations based
on the 76 point data field are given in Appendix II.
ANALYSIS
The data sets were analyzed using the Multiple
Regressions with Case Combinations program from the
Health Sciences Computing Facility of UCLA. This pro-
gram calculates a correlation matrix, a least squares
best fit straight line, and associated test statistics.
A table of the correlation coefficients comparing
the two different instrument measurements of the same
parameter is given below:
83 Point Data 76 Point Data
Field Field
HC .y975 .99B1
CO .9909 .9973
C02 .9650 .984 8
As anticipated there is a very good correlation
between the two instrument systems. The best corre-
lations occur for total hydrocarbon, and the worst for
c02. 'l'he complete correlation coefficient matrix for
the 83 point data''field is given in Appendix I. The
corresponding correlation coefficient matrix for the
76 point data field is presented in Appendix II.
The method of least squares was used to calculate
a regression line of the form:
XFR = B1 xgas + Bo
Where:
XpR = Represents the emission component measured
by the standard Federal Register analysis.
X_ = Represents the same emission component
measured by the Beckman Gas Chromatograph.
B]_ = The slope of the regression line.
B0 = The ordinate intercept.
-------
-3-
The following values for slope and intercept were
obtained.
83 Point Data
Field
76 Point Data
Field
Slope
Intercept
Slope
Intercept
HC
CO
co2
.99969
.91417
.94201
-0.7838
8.6167
0.1733
.99737
.90284
.94191
-0.91211
5.89731
0.15835
The intercepts are easier to visualize when expressed
as a percentage of the maximum ordinant value:
83 point data field 76 point data field
intercept as a per- intercept as a per-
centage of max ordi- centage of max ordi-
nate value. nate value.
HC -0.46 % -0.54 %
CO 1.0 % 0.71 %
CO2 6.2% 5.7%
Since the data represents measurements of the same
parameters by different instrument systems, the theo-
retically expected curve would have intercept of zero
and slope of one. It is desirable to test if the cal-
culated experiment values are significantly different
from these theoretical values. The hypothesis inter-
cept = 0 and slope = 1 may be tested using the student
"t" statistic.-'- A table of the calculated t values and
a summary of the above tests for the 83 point data field
and the 76 point data field are given in Appendices I
and II respectively.
It can be stated with 95% confidence that all
experimental intercepts are non-zero and that the CO
and CO2 slopes differ from one. It should be noted
from the calculated t values that greater confidence
in these predictions can be made with the reduced
data field.
DISCUSSIONS
The agreement between the two instruments is best
for the total hydrocarbon measurements. This is to be
expected since both systems use a similar FID approach
to measure total hydrocarbons. The intercept of the
HCfr vs. HCqas fitted curve is only 0.5% of the total
HC range. This seems within normal experimental accu-
-------
-4-
racy, however since the two measurements of total hydro-
caroons agree very well, this intercept is statistically
significant. The slight non-zero intercept probably
represents a zero point mis-calibration of one or both
instruments.
The line fitted to the CO data had a slope appro-
ximately 9% lower than expected. There are several
possible contributing factors. First, the calibration
concentration used for the gas chromatograph was only
8% of full instrument scale, since the readable accu-
racy of the pen is about + 0.2% of full scale, a cali-
bration error of + 2.5% is very easy. This may be a
contributing factor, but should not be responsible
for the observed 9%. A more probable explanation is
apparent from the scatter plots where a non-linearity
of the data can be noted. Specifically the Beckman
gas chromatograph gave higher readings than the stan-
dard Federal Register instruments and this effect was
more pronounced in the high concentration range. The
least squares fitted straight line is quite sensitive
to extreme points hence the few non-linear appearing
points in the high concentration region would cause the
fitted line to have a reduced slope and a positive
intercept.
Several observations are consistent with the non-
linear data hypothesis. First the percentage difference
between the means of the measurements from the two instru-
ments is only 3.7% vs. the 9.0% variation in the slope.
Second the intercept of the CO fitted line is about 1%
of the maximum value. Also the calibration concentration
is only about 1/4 of the maximum measured concentration
hence non-linear deviations m the high concentration
regions are very plausible.
As evident from the scatter plots, the CO2 data
showed the greatest random variation between instruments.
Consequently the correlation coefficients were lower for
CO2 than either HC or CO. These variations may represent
instrument fluxuations or drift. The CO2 reading of a
Beckman gas chromatograph monitoring a calibration gas
was observed to drift downward about 4% in one hour. At
the time this was attributed in insufficient instrument
warm-up time which might be a problem of the test data.
-------
-5-
The fitted straight line had a 6% lower slope than
expected, and the intercept was high, approximately 6%
of the maximum measured concentration. The means of
the two measurements showed the same discrepancies, the
mean of gas chromatograph data was 6% higher than the
mean of the Federal Register instrument data. This is
significant and the 6% error is consistent, however it
is difficult to draw any conclusion about the source of
this error because of the scatter of the data points.
It would be very difficult to get an accurate instru-
ment calibration if drift or non-reproducability is a
problem.
CONCLUSIONS
The total hydrocarbon measurement agreed very well
and no improvements appear easily available.
The CO measurement problem occurs in the high con-
centration region. More data points in the region would
be necessary to ascertain if the apparent non-linearity
is a systematic effect. Additional gas chromatograph
calibration points in the high concentration range would
be most useful, however the present calibration point
near the data mean should not be eliminated. It was
noted the maximum concentration measured by the gas
chromatograph was about 40% of the instrument recorder
scale. If no higher concentrations are routinely en-
countered, changing the recorder scale could improve
accuracy.
Removing the random fluxuations would be a major
improvement in the CO2 data. The Federal Register pro-
cedure uses NDIR to measure CO and CO2, while the Beck-
man gas chromatograph uses a methanator and FID to
measure both CO and C02• It would therefore seem pos-
sible for the CO2 data to have as low random fluxuations
as the CO data. If this were the case, better calibra-
tion could result.
SUMMARY
The Beckman Gas Chromatograph Model 6800 when modi-
fied as supplied to the MVEL is capable of measuring total
hydrocarbons, methane, CO, and O^. The Beckman 6 800 is
designed as an ambient air quality instrument, hence no
sensitivity problems were expected or observed. Also no
problems of exceeding the operating range were encountered.
-------
-6-
When comparing bag sample measurements made on the
Federal Register instruments and on a Beckman model 6800
gas chromatograph significant differences were found for
CO and CC^. It is felt these variations represent flux-
uation of insufficient calibration of one of the instru-
ments, probably the gas chromatograph. Investigation of
more thorough calibration and adjustment of the gas
chromatograph is recommended if better agreement is
desired.
-------
REFERENCE
^"Ostle, B Statistics in Research, The Iowa State
University Press Ames, Iowa 1963.
-------
APPENDIX I
(83 Point Data Field)
-------
Table .1
(83 Data Points)
Federal
Rerjistral
XnstruiTiOdtat J on
HC
CO
COz
No.
PPM
, PPM
%
l
97.GL
5 5 5 . 5 8
iTTT
2
4.45
12.33
1.13
J
1C.31
62.21
1.48
4
43.55
iee.es
1.65
5
IC.2C
61.43
1.41
6
117.t9
436.39
1.7C
7
12.1?
4 0.13
1.13
a
54.24
9 4.76
1 .41
9
53.95
246.25
1 .69
10
?.e<
1.64
1.13
11
e.C4
ec.4i
1.46
12
9t.5c
445.14
1 .11
13
4.25
11.11
1 .14
14
e.54
55.5?
1 .49
IS
76.65
546.06
1.63
16
1-4.61
47.5S
1.27
I?
92,04
94.C5
1.43
18
US. 3 5
*33.4 C
1.6C
I?
2.93
3C.58
C.97
20
1C.C7
79.69
1.46
21
117.11
404.14
1.55
22
3,!t
0.23
1.14
23
6.25
56.24
1.45
24
93.CI
570.90
1.63
25
U%t4
S3. 19
1.30
26
44.14
134.37
1.55
27
64.1 L
235.62
1.63
2a
2.92
3.60
1.C4
29
7.55
51.50
1.46
3C
113.7)
666 .32
1.63
31
3.98
13. CZ
l.U
32
9.39
57. C7
1.39
33
15.79
312. C6
2.76
34
23.25
215. 11
2.29
35
56.66
613.C6
t. 5 1
36
5c.se
284.63
1.99
37
.l.U
1.77
U19
38
1C.95
63.55
1.62
39
32.45
479.69
2.66
40
5.13
22.33
1.4C
41
19.74
43.52
1.99
42
19.27
295.26
2.10
43
5.62
3. 31
1.76
44
60.34
546.77
i.es
45
13. t 7
61.33
1 .49
46
21.26
34C.4S
2.16
47
2. fc 3
1C.71
C.94
46
5.80
4.17
1.75
49
1 10.62
315.C2
2.14
50
3.25
9.17
1.50
51
11 .SO
53.90
2.03
52
67.5J
195.!7
2.27
53
2.65
2.92
1.53
54
2) .3C
57.Ct
l.ie
55
95.63
222.TO
1.7L
56
n.*i
66.2-4
1.43
57
36.
2£2.< 7
2. 34
56
5.5?
3.£5
1.59
59
43.36
4C»6 2
?.C4
60
58.51
29).CI
1 .64
41
54.65
21C. U
1.41
12
36.53
5C0.61
2. 39
6 3
i ."5
126.4]
l.tC
64
6.20
*;i .53
2.C*
65
3L.55
3C7.2?
1.26
66
1.52
41,76
C.E0
67
5.51
111.74
1.32
K.n
I2t .22
1.^9
69
5. 24
5 . 46
1.20
70
33.93
31. 38
1.67
IV
26. 37
2*9.32
1.16
72
14.45
2.32
0.(2
73
52.72
71.75
1 .CI
74
44! .4 ?
46 2.C7
. ? . 3C
75
4. 59
KC.tS
1.62
76
30.94
113. CC
I.1C
77
37.65
49.76
i.et
/e
2f .3«
411 .38
1.21
7*
2.16
21.41
0.7d
CO
4.32
46.47
I.C9
ei
7. 5 J
2ie.5)
2.51
82
3.97
e.2t
I .f 6
G 3
17.33
102.ee
2..M
B^ehma n
GdS
Chromatograph
HC
cu
C02
PPM
PPM
I
1CT7T5
6 37TT;
ITTT
3.SI
10.CO
1.03
1 1.12
60.CC
1.39
43.46
i£2.5C
1.56
1 1.50
72.50
1.39
117.14
497.€4
1 .64
14 . t 6
42.50
1.L1
5£.c9
ICC .CC
1.33
54.52
22 7 .5C
1.61
3.€9
2.72
1 .03
9.04
77.5C
1.37
9 3.45
495.CC
l.tl
4.55
7.50
1 .06
9.72
57.50
1 .39
£1.44
530.CC
1-70
L6.34
52.5C
i. ie
9^ .40
1C0.CC
I. 34
172.07
?C6.4S
1.61
2.07
I 7 . 5C
I.09
19.90
90.ee
1.39
119.50
4 50.31
1.51
3.95
2.5C
1.10
7 .04
55.CO
1.33
96.2 2
657 .50
1.75
U.52
itc.cc
1 .25
44 .90
140.CO
1.45
65.26
265.CO
1 .56
2.29
2.50
C.U3
€.77
55.CO
1.44
3 10.67
737.50
1.57
2.37
12.50
1.08
1C.29
60.CO
1.16
17.52
375.CO
2.80
35.29
215.00
2.25
5 6.01
e9C.CC
1.43
46.82
32C.61
2.CO
C.86
Q.<33
1.1*
10.68
56.ee
1 .36
37.05
478 .02
2.67
5.32
17.77
1.37
22.15
45.30
1.93
21.49
307.50
2.24
6.93
2.50
1.88
63 .66
ai6.47
1 .56
12.49
6B. C9
1.21
22.15
335.00
2.12
2.67
7.50
0.68
4.51
5.CO
1.26
102.99
262.91
1.96
4.1Z
5.CO
1.38
34. 56
47.50
1.87
72.57
160.CO
2.16
3.69
2.50
1.39
26.65
47.5C
1.65
9C .94
247.50
U57
15. 70
45.55
1 .15
36.32
33 7 .*0
2 .64
5.91
2.50
1 .36
4 3.20
35.00
1.67
55.96
312.50
1.53
56.51
2CC.C0
1.24
3C.5L
¦493. as
2.20
t ¦ 2 B
k 5 3.53
1.56
77.3?
370.eo
2.34
31.32
330.25
1.19
1.34
¦4 2.66
0.74
6.45
315.22
1.C6
22.92
240.38
1.26
5.66
5.23
1.15
34.70
30.3 2
1.66
29. CO
198 . f 5
1.1C
14.9)
2.63
0.59
53.55
77.70
0.97
49.12
5?O.C9
2.41
4.It,
ICO.93
1.59
31. n
125.22
1.04
2e.
52.7C
I.CC
3C .fi9
270.24
1.19
2.C2
25. 1 5
0.76
5.CB
42.71
1 .07
I .32
265.<56
2.59
3.9^
15.71
1 .61
U.GJ
103.
2.35
-------
Table 2
Scatter Plot of 83 Point HC Data Field
-l.
87 m
Beckman Gas Chroraatograph Measurements
of Total Hydrocarbons (PPM)
17?-
-------
Table 3
Scatter Plot of fcJ3 Point CO Data Field
«35.
£ H
(D CU
CJ M
W
C
tD
fD
3 H.
fD W
a Ct-
rl- <0
« n
O M
Ct» JJ
w
O rt
O H
~d CD
~O D
S rt
£U
rt
p-
O
D
* •
• • • *
• «
a
* *
-1.
• 4«
*«*«»
«*•*
• 4
*
-1.
•<55.
911.
Beckman Gas Chromatograph
Measurements of CO (PPm)
-------
Table 4
Scatter Plot of 83 Point C02 Data Field
3 1*"
~ •
4
~ * 4
444
* *44 *
4 4 4
4 4 4
4 4 4
4 4 444 4
4 4 4
4 4
4
444
44
44
.4
1.8
Beckman Gas Chromatograph
Measurements of CO2 (%)
-------
Table 5
Correlation Coefficient Matrix
83 Point Data Field
u c
4-> Q)
qj c e
X C £
Q) H
Cm
A W
in
0.-P
<0
nJ C
0
M
-------
Table 6
Calculated t Values
For 83 Point Data Field
Hypothesis Hypothesis
Intercept = 0 Slope = 1
HC 2.035 0.039
CO 2.174 6.217
C0o 3.868 2.040
Summary of Testing
At 95% Confidence Level
HC
CO
CO'
Hypothesis
Intercept = 0
reject
reject
reject
Hypothesis
Slope = 1
accept
reject
reject
The t statistic for 95% confidence level
for a sample of 83 is 1.989.
-------
APPENDIX II
(76 Point Data Field)
-------
Taolc 1
(76 Data Points)
Federal Pegj^tcr RecKn^n r:as
Instrurcntation Chromatograph
HC
CO
C02
HC
CO
C02
No.
PPM
PPM
%
PPM
PPM
%
^———
—
——
1
97,CI
559-50
1.71
101.25
637,e4
L.62
2
4.45
12.33
1 .13
3.81
to.CO
t.ue
3
IC.31
62.21
1 »4d
11.€2
60.CO
1.39
;
43. 55
168.C9
W trj
AQvAt,
152.53
I.it
5
ic.?e
<1.43
1.41
11 . ? J
72.50
] .39
6
iw.64;
436.39
1.7C
117. 14
4*17.e4
1.64
7
12.17
AC. 13
1 .13
14. 16
42.50
1.11
£
54.26
54. 74
1.41
56. 89
1C3.C0
1 .33
9
5 3.95
246.25
1.69
54.52
2 2 7.53
1.61
10
3.64
1.64
1.13
3 .€9
2.72
1.08
11
e.C4
8C.4 L
I .4£
9.04
77.50
1.37
12
94.5C
445.14
1 .71
93.45
495.CO
1 .61
13
4.25
11.11
1 .14
4.55
7.50
1.08
14
6.94
55.52
1 .49
9.72
57.53
I. 39
15
14.6 1
47.58
1.27
16. 14
52.50
1. IS
16
9 2 .C 4
94.C5
1 .43
96.40
ICC.CO
1.34
17
ItS.75
631.40
1.60
172.C7
90S .43
1 .61
16
1C.C7
79.49
1 .46
19.90
90. CO
1 .39
19
]17.j 1
4C4.14
1.55
119 .5 i>
450.31
1.51
20
3.3®
C. 33
1. U
3.95
2 .50
1.30
21
<.25
56.24
1.45
7.04
55-00
1.33
22
53.CI
538.93
1 .E3
96 .22
657.50
1.75
23
16.64
93.19
1.30
11 > 52
JCC.CO
1.26
24
44.14
134.37
1.55
44 .90
140.CO
3 .45
25
tA.U
235.£2
1.63
65.26
265.tO
1 .56
26
2.52
3.60
1 .C4
3 .29
2.50
0.ti3
27
7.55
51.50
1 .46
8.77
55.00
1.44
26
113.71
666.32
1.63
110.67
737.50
1.57
29
3. 50
13. C2
1.11
2.37
12.50
1.08
3C
5.35
57.C7
1.39
1C.29
60 .00
1. 16
31
13.75
312.C6
2.76
17.52
375.00
2.40
32
33.25
215.11
2.29
35.29
215.CO
2.25
33
5t.6t
813.C6
1.51
58.01
890.CO
1.43
34
5c.ee
284.8 3
1 .99
46.62
320.61
2.00
35
l.H
1.77
1.19
c.et
0.93
1.14
36
1C.95
63.55
1.62
IC.8U
56.68
1.36
37
32.45
479.69
2.66
37.05
470.02
2.67
38
5*13
22.33
1 .40
5.32
17.77
1.37
39
IS.74
43.52
1.99
22.15
Vj.38
1.93
40
15.27
295.26
2.10
21.49
307.50
2.24
41
5.62
3.31
1.76
4.93
2.50
i .ee
42
ec.34
548.77
i.e*
6 3.66
6J6.47
1.53
43
12.17
81.33
1 .49
12.49
68.09
1.21
44
21.28
340.48
2. If
22.15
335.CO
2.12
45
2.63
1C.71
0.94
2.67
7.50
C.63
46
3.25
9.17
1.5C
4.12
5.00
1.3a
47
11. 9C
51.90
2.C3
14.58
47.50
1.47
4e
67.51
195.17
2.27
72.57
160.00
2.16
49
2.65
2.92
1.53
3.69
2.50
1.J9
5C
21.3C
57.C 1
1.93
26.65
47. >0
1.65
51
55.67
222. 7C
1.71
90.94
24 T.50
1.57
52
n.ei
6?.24
1.43
15.76
45.55
1.15
53
36.C4
262.£7
2.74
36.32
317.50
2.64
54
5.57
3.25
1.59
5.91
2. 50
1.36
55
5 E»51
293.01
1.64
55.96
312.50
1.53
56
5 4.6$
210,11
1.41
56.53
200.00
L . 24
57
3C.53
5CC.43
2.3S
36.51
4
-------
Table 2
Scatter Plot of 76 Point HC Data Field
4
44*
4
*
4* *
* *
44#
4 *
44
44
44*
*4 4
4444
4 4#
0.
-I.
e?.
Beckman Gas Chromatograph
Measurements of Total Hydrocarbons (PPM)
-------
Table 3
Scatter Plot of 76 Point CO Data Field
839. ~-
• 4
4
* *
4 4
4 • * *
-1.
• 4*
Mil
444
~ *4 4
I* *
-1.
*455*
911.
Beckman Gas Chromatograph
Measurements of CO (PPM)
-------
Table 4
Scatter Plot of 76 Point C02 Data Field
3.1:
ro
a
fD
K
H
0)
CU
0)
t—1
w
c:
w
t-f
n>
(5
-------
Table 5
Correlation Coefficient Matrix
76 Point Data Field
n
0
01-r) B
•rl JJ JJ
id c
+j r}
xi v) a)
0) C2
Cm H
W
W Oi-U
fl3 (0 C!
a M (U
CP 6
COO)
nj 4J ^
E rt) 3
3 e w
O O ni
0)
-------
Table 6
Calculated t Values
For 76 Point Data Field
Hypothesis Hypothesis
intercept = 0 Slope = 1
HC 2.647 U.371
CO 2.630 12.536
C02 5.175 3.U10
Summary of Testing
At 95% Confidence Level
Hypothesis Hypothesis
Intercept = 0 Slope = 1
HC reject accept
CO reject reject
CO2 reject reject
The t statistic for 9b% confidence level
for a sample of 76 is 1.992.
------- |