£:
Report No. SR98-02-01
Additional Study of
Preconditioning Effects and
Other IM240 Testing Issues
~ [
prepared for
U.S. Environmental Protection Agency
February 2,1998
prepared by:
Sierra Research, Inc.
1801 J Street
Sacramento, California 95814
(916) 444-6666
-------
EPA-420-R-98-103
Additional Study of Preconditioning Effects
and Other IM240 Testing Issues
Prepared for:
U.S. Environmental Protection Agency
Regional and State Programs Division
Under Contract No. 68-C4-0056
Work Assignment No. 2-04
February 2, 1998
Prepared by
Sierra Research, Inc.
1801 J Street
Sacramento, CA 95814
(916) 444-6666
-------
DISCLAIMER
Although the information described in this report has been funded wholly or
in part by the United States Environmental Protection Agency under
Contract No. 68-C4-0056, it has not been subjected to the Agency's peer and
administrative review and is being released for information purposes only. It
therefore may not necessarily reflect the views of the Agency and no official
endorsement should be inferred.
-------
Additional Study of Preconditioning Effects
and Other IM240 Testing Issues
Table of Contents
1 Summary 1
2. Introduction .. 4
Background . ... 4
Project Scope 7
Organization of the Report ... .7
3 Description of EM240 Test Programs .9
Previous Test Programs .... ... 9
New Test Programs ..... 10
4. Vehicle Preconditioning Analysis 14
Analysis of Data from the Replicate IM240 Testing . 14
Analysis of Data from the Replicate Phase 2 Testing .17
Impact of Cutpoints on False Failures Due to Preconditioning 21
Modal Analysis of Replicate IM240 Testing 25
Conclusions and Recommendations from the Preconditioning Analysis 32
5 Development of IM240 Fast-Pass Cutpoints 34
Background 34
Previous Methodology 35
Alternative Methodologies .... 35
Selected Approach .41
Development of Fast-Pass Standards 45
6. Impact of Preconditioning and Cutpoints on Test Duration . . 47
IM240 Failure Rates Using the Current Fast-Pass Procedures 47
IM240 Failure Rates Using Full IM240 Test Scores 51
Test Cycle Duration for the Current Program ... 56
Impact of Tighter Cutpoints and Preconditioning on Test Times and Lane
Throughput With the Current Fast-Pass Procedures 58
Impact of Tighter Cutpoints and Preconditioning Test Times and Lane
Throughput With the Revised Fast-Pass Procedures 62
Conclusions from the Analysis of Test Duration .. 65
-iii-
-------
7 Assessment of IM240 Speed Variation Criteria ...
Examination of Current Criteria
Evaluation of Alternative Statistical Measures
Development of PKE-Based Variation Limits ..
8 Evaluation of NOx Correction Factors
Existing Correction Factor Equation
Review of the Original Data .... ....
Discussion of Results
Use of IM240 Data ...
Conclusions
9 References
Appendix A
Appendix B
Appendix C
Appendix D
-iv-
-------
1. SUMMARY
Under Work Assignment 2-04 of EPA contract #68-C4-0056, Sierra Research, Inc.
(Sierra) has performed several tasks related to improving the accuracy and efficiency of
the tests conducted under motor vehicle inspection and maintenance (I/M) programs As
described below, some of the work performed also resulted in the development of
proposed revisions to the test procedures used during the new vehicle certification
process
Revised IM240 Preconditioning Procedure - Additional analysis of IM240 preconditioning
requirements reinforces the preliminary conclusions of work performed during 1996,
which indicated that IM240 testing done in accordance with the current EPA guidance
results in the failure of many vehicles that would pass the test with further preconditioning
The test data collected during the performance of this Work Assignment also demonstrate
that the first 93 second-long "hill" of the IM240 driving cycle is relatively ineffective for
preconditioning. To ensure adequate preconditioning in the most efficient manner, the
most effective change to the IM240 test procedure would be as follows.
1 The first hill of the IM240 driving cycle should be deleted, and
2 An algorithm should be implemented that determines when the second hill should
be repeated based on emissions occurring during certain portions of the driving
trace
Detailed criteria for determining when to extend the duration of the test have been
developed on a pollutant-specific basis for the full IM240 test. Additional analysis is
required to develop an algorithm for the second hill only
Revised Fast-Pass Cutpoints - Under current EPA guidance, a vehicle may pass the
IM240 test without having the test run to completion if its emissions are sufficiently low,
however, all vehicles are required to run the first 30 seconds of the test The current
"fast-pass" cutpoints that apply between second 31 and second 239 are based on an
analysis of data collected during the tests of a large sample of vehicles in an 1M240 pilot
lane The cutpoint at each second is based on the emissions of vehicles that were marginal
failures at the end of the full test
Under this Work Assignment, a set of fast-pass cutpoints have been developed using a
more rigorous statistical approach Rather than basing the cutpoints on cumulative
emissions, consideration has been given to the rate of change in emissions that is occurring
by predicting full-cycle emissions from emissions during specific segments of the test.
This allows vehicles to be identified and passed that were not adequately preconditioned at
-------
the beginning of the test. This new approach, combined with more focus on eliminating
false passes, has reduced the frequency of false passes occurring under the current fast-
pass cutpoints by 50% while simultaneously reducing the average test time Additional
analysis is needed to develop fast-pass cutpoints for testing performed using a shortened
version of the IM240 (e g, with the first hill removed)
Speed-Variation/Excnrsion Criteria - Under the current EM240 test procedures, drivers are
required to meet certain criteria specifying the accuracy with which the target speed-time
trace is followed A speed-excursion criterion controls the deviation from the target trace
allowed at any point along the trace, a speed-variation criterion limits the cumulative
amount of deviation from the target trace over the entire test Under the current speed-
excursion criterion, the vehicle speed may vary from the target trace by as much as 5 mph
(The speed-excursion criterion requires that the vehicle be within 2 0 mph of the driving
trace, but deviations from the trace of ±1 second are also allowed. As a result, when the
trace specifies an acceleration rate of 3 mph/s, the allowable speed-variation increases to
5 0 mph) Although the allowable speed-excursion is significant itself, there is no limit to
the difference between the target acceleration rate and the actual acceleration rate A
limited laboratory testing program indicates that variations in the way a vehicle is driven
within the allowable speed range may affect emissions by more than 100%
The current limit on the total amount of speed-variation allowed over the entire IM240
cycle limits the frequency of the instantaneous speed-excursions that occur during the
course of the full test, however, the cumulative speed-variation criterion does not apply to
fast-pass or fast-fail testing. In addition, our analysis indicates that enormous variations in
emissions are still possible within the current speed-variation criterion
We are not recommending a change in the ±2 mph/±l second speed-excursion criteria that
currently apply Practical experience indicates that this range of excursion is necessary for
many vehicles However, we are recommending a fundamental change in the speed-
variation criterion. First, we are recommending that the cumulative variation allowed
during the entire test be changed from a simple statistical measure of speed-variation to
the variation in positive kinetic energy (PKE) change Because the power required to
drive a vehicle increases exponentially with speed, a speed-variation that occurs at high
speed requires more power than the same variation at low speed. Because of this, exhaust
emissions are better correlated with PKE variation than speed-variation Second, we are
recommending a modified PKE variation criterion be established for tests involving a fast-
pass or fast-fail Specific recommendations have been developed for the full IM240 test
Further analysis is required to develop speed-variation criteria for shortened versions of
the test (e.g, with the removal of the first hill).
Because of the significant variation in exhaust emissions that is possible under the current
speed-excursion criteria, we are also recommending that a PKE-based speed-variation
criterion be added to the Federal Test Procedure (FTP). This would be effective in
eliminating what we believe to be a common practice of driving vehicles on the official test
more smoothly than the target trace for the purpose of minimizing emissions It would
-2-
-------
also be effective in eliminating the concern that poor drivers could cause
unrepresentatively high emissions to occur during compliance testing
NOx/Humiditv Correction Factor Revisions - Since the early days of the motor vehicle
emissions control program, it has been observed that NOx emissions from internal
combustion engines are dependent on the temperature and humidity Higher ambient
temperatures and/or lower humidity raises the peak combustion temperature, which, in
turn, increases NOx formation To account for variations in humidity, the FTP specifies a
humidity correction factor Because of the relatively narrow temperature range (68-86 °F)
over which the official certification test is performed, there is no correction specified for
temperature.
Current EPA I/M guidance requires that the humidity correction factor specified in the
FTP be applied during I/M testing However, the current guidance states that the
maximum ambient temperature used in calculating the correction factor should be 86°F,
the maximum temperature allowed during the FTP This limitation was imposed to avoid
going outside the temperature range of data set used to develop the humidity correction
factor in the first place However, field experience indicates that the 86°F cap on the
temperature used to calculate the humidity correction factor is resulting in significant
under-estimation of NOx emissions during tests that occur when the actual ambient
temperature exceeds 86°F For example, NOx emissions failure rates are lower during the
summer in Phoenix, Arizona
Under this Work Assignment, Sierra was required to review the derivation of the
currently-specified humidity correction factor and recommend an approach for developing
an improved correction factor. Because of the potential significance of the problems we
identified with the current correction factor, the level of effort invested in this portion of
the Work Assignment went beyond our original plans. Our analysis shows that the current
practice of doing NOx corrections based only on humidity differences should be changed
Over the wide range of conditions experienced in I/M testing, both temperature and
humidity need to be accounted for As an interim solution, we derived a new temperature
and humidity correction factor that better matches the original data used to develop the
current correction factor Immediate implementation of this new correction factor would
significantly reduce the false passes now occurring during high temperature testing. As
specified in the Work Plan, we have also developed a proposed approach for developing a
more representative correction factor using the large amount of data available from
programs using IM240 testing
As in the case of speed-excursions/variations, we recommend that similar changes be made
to the correction factor currently contained in the FTP Although a less extreme range of
temperature and humidity occur during certification, surveillance, and in-use compliance
testing, our analysis indicates that the current NOx correction factor does not represent
the best fit to the data from which it was developed More importantly, the current
correction factor is based on the tests of vehicles with absolutely no NOx controls. There
is serious question as to whether the relationship between NOx, temperature, and humidity
for those vehicles is representative of present generation vehicles
-3-
-------
2. INTRODUCTION
Background
Under the Clean Air Act Amendments of 1990, metropolitan areas with the most serious
air quality problems are required to implement so-called "enhanced" I/M programs. One
element of an enhanced program is a more effective test procedure than the simple idle
tests used in "basic" I/M programs Two different test procedures for exhaust emissions
testing in enhanced programs have been approved by EPA the "IM240" test, and the
"Acceleration Simulation Mode" (ASM) test. Both of these procedures have been shown
to be capable of separating vehicles with excessive exhaust emissions from other vehicles,
however, the accuracy of the test depends on whether tested vehicles have been
adequately preconditioned and whether the speed-time profile associated with each test
procedure is closely followed With either procedure, the efficiency of the testing process
depends on how quickly accurate decisions can be made as to whether a vehicle should
pass or fail
Inadequate preconditioning of vehicles prior to testing is a potential cause of inaccurate or
inconsistent test results because exhaust emission levels depend on how thoroughly a
vehicle has been warmed up Before the vehicle is thoroughly warmed up, high emissions
can be caused by air-fuel ratio enrichment or an inactive catalytic converter In addition,
increased emissions due to purging of loaded evaporative emissions canisters may also be
an issue associated with inadequate preconditioning prior to I/M testing
Inadequate vehicle preconditioning has previously been identified as a cause of false
failures in I/M programs Under the current EPA "high-tech" test guidance1*, IM240
preconditioning procedures are woven into the "two-ways-to-pass" standards. Vehicles
that exceed the emissions standards established for the entire 239-second test are passed
or failed based on emissions occurring during the last 146 seconds of the test (Phase 2)
The separate set of standards that apply to Phase 2 are slightly more stringent For
vehicles that initially demonstrate high emissions, the first 93 seconds (Phase 1) of the test
are used to precondition the vehicle for the second phase of the test In addition, EPA
calls for a "second-chance" test whenever a vehicle fails the initial test by less than 50% of
the standard and was in a queue for more than 20 minutes before being tested
Superscripts denote references listed in Section 9 of this report
-4-
-------
Considerable data have already been collected regarding the preconditioning requirements
for IM240 testing. During 1996, Sierra conducted an evaluation of this issue using data
obtained from a sample of vehicles recruited from IM240 lanes in Phoenix, Arizona, and a
laboratory test program at Sierra's facilities in Sacramento 2 Preliminary conclusions from
the evaluation are summarized below
1 Most vehicles that have been waiting in a queue for 15-30 minutes prior to testing
are not thoroughly warmed up by running Phase 1 of the IM240 test
2 Phase 2 of the IM240 test procedure is more effective for preconditioning than
Phase 1
3 Using the current IM240 test procedures, it is estimated that 25% of the vehicles
failing the final IM240 standards would pass with further preconditioning.
4 Vehicles that would benefit from further preconditioning can be identified through
modal analysis of the emissions recorded during the IM240 test
Based on the above conclusions, two options (A and B) were recommended for changing
the current preconditioning procedures-
Option A
1 Retain existing IM240 test procedure and two-ways-to-pass standards.
2 Repeat entire IM240 if.
• Phase 2 emissions failure is marginal, or
• emissions near end of Phase 2 are relatively low; or
• emissions during Phase 2 are significantly lower than during Phase 1.
Option B
1 Eliminate Phase 1 and make initial pass/fail decision based on running only
Phase 2
2. Give second-chance test (another Phase 2) for all vehicles that initially fail
3 Give third-chance test if emissions during second-chance test are significantly
lower than emissions during initial test
In addition to the false failures caused by inadequate preconditioning, inadequate control
over vehicle operation during the IM240 test procedure can contribute to inaccurate
results The ability of a driver to follow the IM240 speed-time trace has a significant
-5-
-------
effect on the emissions recorded during the test To limit this variation in test results,
tolerances are applied to driver performance However, measured test-to-test variability
indicates that the tolerances specified by EPA are not adequate These tolerances include
speed excursion limits within ±2 mph of the target speed within 1 second of any given test
time, and a standard error (speed variation tolerance) of ±2 0 mph over the entire drive
cycle for all full IM240 tests Within the specified tolerances, it is possible for less skilled
drivers to introduce more severe throttle variation by overshooting the target trace while
staying within the specified tolerances The variation in emissions can exceed 100%
compared to a drive that more precisely conforms with the target trace
More stringent speed excursion tolerances on the IM240 driving cycle are not considered
practical, but more effective speed variation tolerances can be employed, especially for
"fast pass" and "fast fail" testing. Currently, there is no speed variation tolerance applied
to anything other than the full IM240 test. Developing speed variation criteria for variable
driving distance is clearly needed to identify when unacceptable speed variability occurs
during shortened versions of the IM240 cycle
There are also alternative speed variation criteria that may minimize the variation in
emissions while still being feasible for use by minimally trained drivers with a reasonable
aptitude for dynamometer driving One alternative criterion is the total Positive Kinetic
Energy (PKE) change per mile traveled during the IM240 cycle This criterion proved to
be well correlated with throttle variation during previous work performed by Sierra for the
EPA Certification Division It may also be possible to adjust actual emissions results
based on PKE results or another vehicle power-related parameter to explicitly account for
the effect of driver variation on vehicle emissions.
Two other IM240 issues that need to be investigated are the fast-pass phase-in standards
and the humidity correction factor formula for NOx emissions contained in EPA's high-
tech test guidance The fast-pass standards were developed in 1993 based on emissions
data from roughly 3,700 IM240 tests The selected standards represent the tenth-lowest
cumulative emissions at each second of the test for all vehicles in the dataset that failed the
1M240 using the two-ways-to-pass criteria. These standards need to be reevaluated using
additional IM240 test data that have been collected to date to improve the robustness of
the analysis, with the objective being to reduce average test times and false passes It also
appears that a re-examination of the analytical approach used to set the current fast-pass
standards would be of benefit, to determine if an alternate approach could be developed to
further minimize both false-pass rates and average test times
Changes in atmospheric humidity can significantly affect vehicle NOx emissions
Increased humidity leads to a decrease in the maximum flame temperature in the engine
cylinder, which in turn results in a reduction in NOx emissions As part of the Federal
Test Procedure (FTP), all NOx emission results are corrected to standard humidity using a
correction formula based on Ferrel's Equation The same formula is incorporated into the
high-tech test guidance to correct IM240 test results to standard conditions
-6-
-------
Because the nonlinearity of the correction formula results in anomalous results at higher
temperatures, the test guidance specifies that a temperature of 86°F should be used in the
formula whenever the ambient temperature is 86°F or above This temperature cap has,
however, resulted in the undercorrecting of test results at high temperatures, such as those
experienced in Arizona during the summer months For example, the correction factor for
ambient conditions of 85 °F and 85% relative humidity is roughly the same as for 95 °F and
65% humidity In the latter case, however, the test guidance caps the temperature at
86°F The resulting use of 86°F and 65% humidity produces an incorrect reduction in the
correction factor, which in turn improperly lowers reported NOx emission readings This
computational error appears to be reflected in the Arizona NOx emissions data, which
show a decreased summertime fail rate and lower average emissions The derivation of
this formula therefore needs to be reevaluated and improved for temperature above 86°F
Project Scope
Under Work Assignment 2-04 of contract #68-C4-0056, EPA directed Sierra to perform
the following tasks
1 Analyze and evaluate procedures to address IM240 false failures due to insufficient
preconditioning,
2 Develop optimized speed variation criteria to minimize emission variations due to
driver variability;
3 Examine the feasibility of adjusting emission results due to speed variations using a
vehicle power-related parameter such as PKE,
4 Reevaluate the fast-pass phase-in cutpoints using existing Arizona data in order to
minimize test time, false failures and false passes, and
5. Evaluate the sensitivity and accuracy of the NOx correction factor formula.
To accomplish the above tasks, Sierra undertook and completed several analysis efforts,
the results of which are documented in this report. A separate project element involved
the design and implementation of an emissions test program at IM240 inspection lanes
operated by Gordon-Darby in Phoenix, Arizona. The test program was designed to
collect additional emissions data for analysis related to the evaluation of alternate IM240
preconditioning procedures
Organisation of the Report
Immediately following this introduction, Section 3 summarizes the previous
preconditioning test program conducted for EPA in 1996, and provides details regarding
-7-
-------
the 1997 Arizona test program and the data collected in that program. Section 4 presents
the results of the vehicle preconditioning analysis, while Section 5 discusses the
development of revised IM240 fast-pass cutpoints Section 6 details the impact of
changes in preconditioning procedures and cutpoints on projected IM240 test times.
Section 7 summarizes the results of the assessment of IM240 speed variation criteria, and
Section 8 presents an evaluation of the current NOx correction factor formula. Section 9
lists the references that were cited in the report.
it it ii
Trfrtf
-8-
-------
3. DESCRIPTION OF IM240 TEST PROGRAMS
Previous Test Programs
The previous 1996 preconditioning study made use of IM240 data that were collected in a
number of different test programs conducted both in Gordon-Darby's I/M lanes in Arizona
and in Sierra's emissions laboratory The primary elements of that testing are summarized
below.
• Replicate IM240 Testing - To better determine whether the full IM240 provides
adequate preconditioning, Gordon-Darby ran back-to-back 1M240 tests on over
300 vehicles. The vehicles selected for testing were those that passed the start-up
emissions standards but failed the final emissions standards, i.e., they would be
considered marginal failures under the final emissions standards The selection
criteria for this series of tests resulted in vehicles waiting in a queue for 15 to 20
minutes prior to IM240 testing The purpose of this testing was to determine (1) if
a single IM240 test adequately preconditioned a vehicle that had been waiting in a
queue, and (2) if the information gained in the initial 1M240 test could be used to
predict whether the vehicle was adequately preconditioned
To supplement Gordon-Darby's testing, Sierra also conducted replicate 1M240
tests on a smaller sample of vehicles under more controlled laboratory conditions.
Sierra tested nine different employee-owned 1989 and later model passenger cars
and light trucks that were selected to span a wide range of manufacturers, engine
sizes, and catalyst locations In Sierra's testing, vehicles were given three back-to-
back tests following a more precisely controlled hot soak
• Replicate Phase 2 Testing - The results of the replicate IM240 testing indicated
that a full IM240 cycle appeared to be adequate for preconditioning purposes
Based on these results, another sample of vehicles was recruited and subjected to
further testing After the vehicles were tested using the full IM240 cycle, they
were soaked for a minimum of 30 minutes and then tested using two Phase 2
portions of the IM240 cycle run back-to-back The purpose of this testing was to
determine whether the Phase 2 portion of the IM240 test would be adequate to
fully precondition a vehicle that had been soaked while waiting in an I/M lane
queue (A positive finding would indicate that reversing the order of the phases in
the IM240 test might be a relatively easy way to provide adequate preconditioning
during the test) Vehicles for this test program were again selected based on
-9-
-------
whether they would have failed the final EPA-recommended emission standards
during the initial IM240
To supplement Gordon-Darby's testing, Sierra also conducted replicate Phase 2
tests on the same sample of vehicles used for replicate IM240 testing in the
laboratory The test sequence involved warming-up each vehicle with an IM240,
hot soaking the vehicle for 30 minutes, and then running three Phase 2s in a row
Based on the results of these test programs, Sierra proposed two options to ensure
adequate preconditioning for the EM240 test (a) analyze the modal emissions of the initial
test to determine if another IM240 is necessary, or (b) eliminate Phase 1 of the test
entirely and give a second-chance test to all vehicles that fail the initial Phase 2; a third
Phase 2 would be given if emissions during the second test are significantly lower than the
first
During the summer of 1997, Gordon-Darby collected additional data to enhance the
database needed to evaluate and refine the preconditioning procedures proposed in the
1996 study. The new test programs are described below
New Test Programs
Replicate IM240 Testing - Similar to the replicate IM240 testing conducted for the
previous study, this program tested failing vehicles over back-to-back IM240s Vehicles
selected for this program were chosen at random, and they had failed the EPA IM240
start-up cutpoints. This is an important distinction between this test program and the
previous replicate IM240 testing, which had been based on vehicles failing the final
cutpoints but passing the start-up cutpoints (i e , the sample used for the 1996
preconditioning study had been "marginal" failures under the final cutpoints). The test
sequence used in this program is shown in Figure 1
A total of 336 vehicles were included in the replicate IM240 testing for this study; the
model year distribution of those vehicles is given in Figure 2 As seen in the figure, the
sample peaks in the 1984 to 1988 model year range, which is consistent with the failures
observed in the state program. Although the failure rate is higher for the older model
years, there are fewer of them in the fleet due to normal attrition. Thus, a maximum
number of failures is observed in model years that have a moderately high failure rate
coupled with a large total number of vehicles In addition, the large increase in the
number of failures in the 1984 model year is partially a result of a change in HC cutpoints
for light-duty trucks from 7 5 g/mi to 3 2 g/mi
Replicate Phase 2 Testing - The replicate Phase 2 testing performed for this study made
use of a slightly revised test protocol compared to the 1996 study The test sequence used
-10-
-------
Figure 1
Test Sequence Used to Investigate
Replicate IM240 Testing in Arizona Test Lanes
IM240
-> +-
IM240
Time
Figure 2
Model Year Distribution in the
Replicate 1M240 Sample
81 82 83 84 85 86 87 88 89 90 91 92
Model Year
-U-
93 94
-------
for this subfleet is illustrated in Figure 3, which shows that the vehicles tested in this
program were subjected to the following procedure
• 30-minute idle,
• two full IM240 tests,
30-minute idle, and
• three Phase 2IM240 tests
(The Phase 2 portions of the initial lM240s are herein referred to as Phase 2A and Phase
2B; the three Phase 2 tests following the last idle period are referred to as Phase 2C, Phase
2D, and Phase 2E)
The vehicles in this program were selected at random, and both passing and failing
vehicles were included in the sample Vehicles were given their official state I/M
inspection prior to starting the initial 30-minute idle period Motorists participating in the
program received a monetary incentive of $100
A total of 101 vehicles participated in this part of the study, and the model year
distribution is illustrated in Figure 4 As observed in that figure, there are more newer
vehicles in this sample than in the replicate IM240 sample, which is consistent with the test
protocol that called for a completely random sample to be procured.
Figure 3
Test Sequence Used to Investigate
Replicate Phase 2 Tests in Arizona Test Lanes
60 —
10
30-Mmute Phase Phase Phase Phase 30-Minute Phase Phase Phase
Idle 1A 2A 18 2B Idle 2C 2D 2E
Time
-12-
-------
Figure 4
Model Year Distribution in the
30-Minute Idle/Replicate Phase 2 Sample
81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97
Model Year
¦13-
-------
4. VEHICLE PRECONDITIONING ANALYSIS
This section of the report presents the analysis of the replicate IM240 data and replicate
Phase 2 data collected by Gordon-Darby in its Arizona I/M lanes For ease of comparison
to the results from the 1996 preconditioning study, many of the tables and figures in this
section follow the same format as the previous report When comparing results between
the two studies, however, it must be kept in mind that different vehicle selection criteria
were used For example, the replicate IM240 database for the 1996 study was based on
vehicles that had failed the final IM240 cutpoints but passed the start-up cutpoints (i e.,
they were "marginal" failures under the final cutpoints), while the replicate IM240 tests
conducted for this study were on vehicles that had failed the start-up standards only
Similarly, the replicate Phase 2 testing performed for the 1996 study included only failing
vehicles (based on the final cutpoints), whereas the replicate Phase 2 testing for this study
was based on a random sample of vehicles (both passing and failing). In addition, the
replicate Phase 2 testing in this study used a slightly different, more controlled test
protocol than in the 1996 preconditioning study
Analysis of Data from the Replicate IM240 Testing
Table 1 summarizes Gordon-Darby's replicate IM240 results from the 1997 test program.
As shown in the table, 19% of the 1981 and later model passenger cars and light trucks
failing the initial IM240 test (based on the start-up cutpoints) passed when immediately
retested; 6% passed the final cutpoints on the retest after failing the start-up cutpoints on
the initial test. In terms of specific pollutants, the impact on NOx was most pronounced in
that 27% of the NOx failures passed the start-up NOx standard on the second test, and
9% passed the final NOx standard on the retest The results for HC and CO were similar,
with 23% and 20% of the fadures going on to pass an immediate retest, and 5% passing
the final standards on the retest
Based on the testing conducted for this study, the fraction of vehicles passing an
immediate retest under the start-up standards is lower than that observed in the database
constructed for the 1996 preconditioning study (19% versus 47%). However, the 1996
study results were based on "marginal" failures (i.e , vehicles passing the start-up
standards but failing the final standards) Since the likelihood that a vehicle will pass a
retest increases as emission levels get closer to the cutpoint, this is not an unexpected
result In fact, in the 1996 study, Sierra estimated that approximately 25% of the final
cutpoint failures would pass a retest This is reasonably consistent with the 19% retest
-14-
-------
Table 1
Replicate IM240 Test Results in I/M Lanes
All 1981 and Later Passenger Cars and Light Trucks Combined
(336 Vehicle Sample)
Pollutant
# Failing
1st IM240
(Start-up Stds)
# Passing
2nd IM240
(Start-up Stds)
% Passing
2nd IM240
(Start-up Stds)
# Passing
2nd IM240
(Final Stds)
% Passing
2nd IM240
(Final Stds)
Any
336
65
19%
20
6%
HC
185
42
23%
9
5%
CO
143
28
20%
7
5%
NOx
146
40
27%
13
9%
passing rate in Table 1 (A detailed analysis of false failure rates as a function of cutpoints
is presented later in this section of the report)
Differences Between Passenger Cars and Light Trucks - The replicate IM240 tests in the
Arizona I/M lanes were analyzed to determine whether there were any differences between
passenger cars and light-duty trucks That evaluation, summarized in Table 2, shows that
passenger cars appear to be more sensitive to preconditioning effects than light-duty
trucks. This could partially be due to the difference in cutpoint stringency between cars
and light trucks under the start-up standards (light-duty trucks have much less stringent
IM240 cutpoints than passenger cars under the start-up standards) It is interesting to
note, however, that there is a high percentage (34%) of LDT NOx failures that pass on a
retest, with nearly 20% passing the final cutpoints after failing the start-up cutpoints
The results presented in Table 2 are slightly inconsistent compared to the 1996
preconditioning study, which showed that passenger cars and light trucks had nearly
identical retest pass rates Again, however, this could be a result of differences in vehicle
selection criteria between the 1996 and 1997 replicate IM240 test programs
Model Year Range Differences - Table 3 shows how the results for prp-1986 model year
vehicles compare to 1986 and later model year vehicles Consistent with the results of the
1996 study, the newer vehicles in the 1997 database appear to be more sensitive to
preconditioning than the older vehicles For example, 26% of the 1986 and newer model
year vehicles failing the initial IM240 test for HC, CO, or NOx passed the second test.
The percentage of failing vehicles passing the second test drops to 12% for the pre-1986
model year vehicles. In addition, the vast majority of vehicles that failed the start-up
standards and went on to pass the final standards are from the 1986 and later model year
group. This is likely related to the fact that vehicles in the newer model year group have
emissions that are closer to the standard than the older vehicles.
-15-
-------
Table 2
Replicate IM240 Test Results in I/M Lanes
All 1981 and Later Passenger Cars vs. Light Trucks
(336 Vehicle Sample)
Vehicle
Type
Pollutant
ft Failing
1st IM240
(Start-up Stds)
it Passing
2nd IM240
(Start-up Stds)
% Passing
2nd IM240
(Start-up Stds)
U Passing
2nd IM240
(Final Stds)
% Passing
2nd IM240
(Final Stds)
Passenger
Car
Any
253
52
21%
13
5%
HC
134
35
26%
7
5%
CO
119
25
21%
7
6%
NOx
114
29
25%
7
6%
Light-
Duty
Truck
Any
83
13
16%
7
8%
HC
51
7
14%
2
4%
CO
24
3
13%
0
-
NOx
32
11
34%
6
19%
Table 3
Replicate IM240 Test Results in I/M Lanes
Pre-1986 vs. 1986 and Later Model Year
(336 Vehicle Sample)
Model
Year
Pollutant
U Failing
1st IM240
(Start-up Stds)
if Passing
2nd IM240
(Start-up Stds)
% Passing
2nd IM240
(Start-up Stds)
if Passing
2nd IM240
(Final Stds)
% Passing
2nd IM240
(Final Stds)
Any
153
18
12%
0
-
Pre-
HC
98
13
13%
0
-
1986
CO
77
11
14%
2
3%
NOx
55
11
20%
0
-
Any
183
47
26%
20
11%
1986
and
later
HC
87
29
33%
9
10%
CO
66
17
26%
5
8%
NOx
91
29
32%
13
14%
-16-
-------
Analysis of Data from the Replicate Phase 2 Testing
As described in Section 3 of this report, vehicles in the replicate Phase 2 sample were
subjected to the following test sequence protocol-
• 30-minute idle,
• two foil IM240 tests (the Phase 2 portions of these tests will be referred to as
Phase 2A and Phase 2B);
• 30-minute idle; and
• three Phase 2 IM240 tests (referred to as Phase 2C, 2D, and 2E)
Vehicles tested in this portion of the study were selected at random and include both
passing and failing vehicles A total of 101 vehicles were included in this program, and the
results of their replicate IM240 tests are given in Table 4 by vehicle type
Table 4
Number of IM240 Failures for Vehicles Included in the Multiple Phase 2 Testing
Passenger Cars vs. Light-Duty Trucks
(101 Vehicle Sample)
Vehicle
Type
Sample
Size
Pollutant
Start-Up Standard
Failures
Final Standard
Failures
1st Test
2nd Test
1st Test
2nd Test
Passenger
Cars
67
Any
5
4
15
12
HC
2
1
10
5
CO
2
2
3
4a
NOx
3
2
9
6
Light-
Duty
Trucks
34
Any
2
0
9
5
HC
1
0
7
3
CO
0
0
2
2
NOx
1
0
3
2
a The increase in CO failures on the second test is a result of two vehicles that failed the second test after passing
the first One failing vehicle passed the second test, and two vehicles failed both
-17-
-------
Mean Phase 2 Scores - One of the primary motivations for this series of tests was to
determine the relative effectiveness of the full IM240 versus Phase 2 of the 1M240 at
preconditioning vehicles that had been waiting in a queue. A simple evaluation that can be
performed is to review the mean Phase 2 scores during the replicate IM240s (i e, Phase
2A and 2B) to the mean Phase 2 scores during the replicate Phase 2 testing (i.e., Phase
2C, 2D, and 2E). Similar mean emission rates for Phase 2B and Phase 2D would indicate
that a single Phase 2 (i.e, Phase 2C in this case) is as effective as a full IM240 plus a
Phase 1 (i.e, the total amount of testing leading up to Phase 2B) at preconditioning a
vehicle,
Comparisons of means were made using a t-test for the difference between two means.3
The null hypothesis in this test is that the two means are identical. From the usual test
statistic, the probability that the null hypothesis was correct was found These
probabilities are reported below as "p-values" to indicate the statistical strength of the
conclusions presented
The mean Phase 2 emission rates of the 101-vehicle sample in the replicate Phase 2
database are summarized in Table 5 Based on the results presented in that table, the
following observations are made
• The results of Phase 2A (which follows the first Phase 1 of the IM240 test) and of
Phase 2C (which immediately follows a 30-minute idle period) are significantly
different for HC (p-value = 0.014), from this difference, it is clear that Phase 1 of
the IM240 provides some degree of preconditioning. However, the HC results in
the Phase 2B test are significantly different from the results in the Phase 2A test
(p-value = 0 097), this shows that a single Phase 1 is inadequate to fully
precondition a vehicle for the sample used here
• The mean emissions from Phase 2B and Phase 2D are nearly identical (The p-
values that the 2B and 2D means are the same are 0 994, 0 963, and 0 988 for HC,
CO, and NOx, respectively) This leads to the conclusion that a single Phase 2
(i e, Phase 2C in this series of tests), is as effective as a full IM240 plus a Phase 1
(i.e, Phase 1A, Phase 2A, and Phase IB) at preconditioning a vehicle that has been
idling in a queue for 30 minutes. This represents a significant difference in overall
preconditioning time - 146 seconds versus 334 seconds
• Based on the results of Phase 2C, 2D, and 2E, only a small degree of additional
preconditioning occurs by running a second Phase 2 prior to the final test This
additional preconditioning is not statistically significant for the means, the p-values
that the 2D and 2E means are the same are 0 712, 0 853, and 0 851 for HC, CO
and NOx, respectively This is illustrated in Figure 5, which shows the relative
HC, CO, and NOx emissions for the three Phase 2 tests following a 30-minute idle
However, as discussed below, some vehicles may need a second Phase 2
preconditioning cycle if there has been a significant engjne-off or idle period prior
to testing.
-18-
-------
Table 5
Mean Emissions Results from Replicate Phase 2 Testing of 101 Vehicles (g/mi)
Phase
HC
CO
NOx
~ jv-iviinuie luie ~~~~~~ —
2A
051
7 72
1.34
2B
0 36
7.35
1 20
jv-ivuiiuie iuic — - — ~
2C
0 75
9.16
161
2D
0 36
7.43
1 20
2E
0 34
7 11
1.17
Figure 5
Replicate Phase 2-Only Emissions
Following a 30-Mlnute Idle
(101 Vehicle Sample)
1.2-,
i 1— .
Phase 2 #1 Phase 2 #2 Phase 2 #3
<2C) (20) (2E1
¦ HC
~ CO
M NOx
Comparison of IM240 and Phase 2 Results for IM240 Failing Vehicles - Of considerable
interest in this program were the vehicles that failed the first IM240 test following the 30-
minute idle period and went on to pass the second test In addition to those vehicles, there
were 2 vehicles that passed the final CO cutpoints on the first IM240 test, but failed on the
second A summary of the vehicles in the replicate Phase 2 sample that had inconsistent
pass/fail results between the first IM240 and the second IM240 (based on the final
cutpoints) is shown in Table 6
-19-
-------
Table 6
Replicate Phase 2 Test Results for Vehicles with Inconsistent Pass/Fail Scores
Between the First IM240 and the Second IM240 Based on EPA's Final IM240 Standards
Hydrocarbon Emission Results
Test
Vehicle
Model
1st IM240
2nd IM240
Composite IM240
q/mi)
Phase 2 Scores (q/mi)
VIN
Date
Class
Year
P/F
P/F
Cutpoint
1st Test
2nd Test
Cutpoint
P2A
P2B
P2C
P2D
P2E
1FABP4034GG124649
970821
LDV
1986
F
P
08
1 49
0 61
0 50
0 82
0 61
1 80
0 55
0 49
1FABP62F4JH136083
970812
LDV
1988
F
P
08
0 87
0 43
0 50
0 66
0 44
0 85
0 48
0 37
1G1JF31W4K7162806
970819
LDV
1989
F
P
08
1 23
0 50
0 50
054
0 38
1 29
0 51
0 38
1G6DW51Y3KR704759
970813
LDV
1939
F
P
06
1 28
0 42
0 50
0 86
0 46
1 35
0 48
0 47
1GNCS13Z8M2316674
970813
LDT
1991
F
P
1 6
1 63
0 16
1 00
1 06
0 13
1 23
0 13
0 06
1GNER16K2MF110830
970814
LDT
1991
F
P
1 6
2 65
1 48
1 00
2 63
1 35
3 12
1 83
1 21
2P4FH41G6FR354315
970812
LDT
1985
F
P
1 6
1 67
1 49
1 00
1 84
1 56
2 14
1 70
1 87
3GCCW80H1FS916333
970819
LDT
1985
F
P
1 6
1 67
1 32
1 00
1 70
1 33
1 91
1 32
1 34
JHMSZ5327CC134515
970815
LDV
1982
F
P
08
0 88
0 69
0 50
0 72
0 63
0 88
0 66
0 64
Carbon Monoxide Emission Results
Test
Vehicle
Model
1st IM240
2nd IM240
Com
sosite IM240
(q/mi)
Phase 2 Scores (q/mi)
VIN
Date
Class
Year
P/F
P/F
Cutpoint
1st Test
2nd Test
Cutpoint
P2A
P2B
P2C
P2D
P2E
1G6DW51Y3KR704759
970813
LDV
1989
F
P
15 0
16 87
14 06
12 0
17 40
17 16
18 56
14 99
18 20
JT2MX73E3F0011461
970813
LDV
19B5
P
F
15 0
12 02
15 21
12 0
10 39
14 79
10 56
10 14
17 05
1G1BU51H8HX219475
970820
LDV
1987
P
F
15 0
27 42
16 02
12 0
3 62
14 91
30 91
19 04
1384
NOx Emission Results
Test
Vehicle
Model
1st IM240
2nd IM240
Comi
sosite IM240
(q/mi)
Phase 2 Scores (g/mi)
VIN
Date
Class
Year
P/F
P/F
Cutpoint
1st Test
2nd Test
Cutpoint
P2A
P2B
P2C
P2D
P2E
4T1SV21EXMU417927
970B12
LDV
1991
F
P
20
2 05
1 46
2 0
2 22
1 68
2 41
1 52
1 64
JT2MX73E3F0011461
970813
LDV
1985
F
P
20
4 73
1 84
2 0
3 88
2 01
5 67
2 00
2 47
1FABP4034GG124849
970621
LDV
1986
F
P
20
2 41
1 40
2 0
2 36
1 58
2 03
1 50
1 41
1GNER16K2MF110830
970814
LDT
1991
F
P
2 5
3 89
1 49
2 5
4 21
1 68
5 16
2 35
1 61
-------
For vehicles failing the first IM240 and passing the second, the results presented in
Table 6 show a significant decrease in Phase 2D emissions relative to Phase 2C (recall that
Phase 2C follows the 30-minute idle period) In some cases, emissions continue to
decrease through Phase 2E, which may be needed if a vehicle has been soaked or idled for
an extensive length of time Although the Phase 2D scores are not always below the
Phase 2 cutpoints established by EPA for the IM240 "two-ways-to-pass" algorithm, in
general they are relatively close In some cases, even the third replicate Phase 2 (i e, 2E)
has higher emissions than the Phase 2 cutpoint However, the Phase 2 cutpoints
developed by EPA can be thought of as "secondary" standards in that the pass/fail
decision is also based on the full IM240 score If an I/M program were to adopt the Phase
2 test as the official I/M test, a new set of cutpoints would have to be developed based
solely on Phase 2 testing
Two vehicles listed in Table 6 failed the second IM240 for CO after passing the first On
one of the vehicles, this appears simply to be a result of test-to-test variability. That
vehicle had composite and Phase 2 scores of 12 0/10 4 g/mi on the first IM240 and
15 2/14 8 g/mi on the second (The composite and Phase 2 CO cutpoints for this vehicle
are 15.0 g/mi and 12 0 g/mi, respectively) Thus, although the vehicle failed the second
test, it was very close to the cutpoint, and the results on the first IM240 were not
significantly below the cutpoint. The second vehicle that passed the first IM240 and failed
the second had relatively high emissions for the composite IM240 on the first test (27 4
g/mi), but its Phase 2 score was very low (3.6 g/mi) and it passed based on the "two-
ways-to-pass" criterion The vehicle failed the second IM240, but its composite score was
fairly close to the cutpoint (i e., 16.0 g/mi versus 15 0 g/mi) The Phase 2B, 2C, 2D, and
2E CO emissions for this vehicle were much higher than the Phase 2A results. It is
unclear what caused this apparent anomaly
Based on the results of this test program, it is clear that Phase 2 of the IM240 is a much
more effective preconditioning cycle than Phase 1 of the IM240 However, from a
practical perspective, it is unclear that I/M program managers would be receptive to
eliminating the current IM240 in favor of Phase 2. One of the primary concerns related to
the current version of the IM240 test is the high-speed operation conducted in Phase 2.
(This concern is related to the public's perception that the noisier operation during this
part of the test is somehow detrimental to the vehicle.) By maintaining Phase 1, and
making use of fast-pass cutpoints, many vehicles pass before Phase 2 is reached,
particularly the high-speed portion of that phase
Impact of Cutpoints on False Failures Due to Preconditioning
The scope of work for this work assignment also called for an assessment of how much
the current cutpoints could be tightened below the phase-in cutpoints without resulting in
an excessive failure rate. In addition, the false failure rate (at various cutpoints) due to
preconditioning was to be determined as part of this task
-21-
-------
Using the 2% random IM240 sample from the Arizona program, failure rates for 1981 and
later model year light-duty vehicles (i.e., passenger cars) were generated under four sets of
composite 1M240 cutpoints-*
• 0 8 g/mi HC, 15 0 g/mi CO, 2 0 g/mi NOx,
• 1 2 g/mi HC, 20 0 g/mi CO, 2 3 g/mi NOx;
• 1 6 g/mi HC, 25 0 g/mi CO, 2.6 g/mi NOx, and
• 2 0 g/mi HC, 30 0 g/mi CO, 3.0 g/mi NOx.
The first set of cutpoints represents the final IM240 standards for 1983 to 1995 model
year light-duty vehicles, and the last set of cutpoints reflects the start-up IM240 standards
for 1983 to 1990 model year light-duty vehicles The two sets of standards between the
final and start-up standards were chosen to reflect varying levels of stringency. For this
analysis, only light-duty vehicles were considered (i e, light-duty trucks were removed
from the analysis), and the same standards were applied to all 1981 and later model year
vehicles This approach avoided the complications of applying different IM240 standards
to vehicles that were certified to essentially the same numerical emission standards The
failure rates under the current preconditioning approach are given in Table 7 for the fleet
as a whole and for the following model year groups
• 1981 to 1984,
• 1985 to 1990, and
1991 and later
As observed in Table 7, going from the 2 0/30/3 0 HC/CO/NOx cutpoints to the
0 8/15/2.0 HC/CO/NOx cutpoints doubles the fleet-wide failure rate, and the difference (in
terms of relative differences) is most pronounced for newer model year cars. Table 7 also
shows that the impact of tighter cutpoints on failure rates is not linear For example, there
is a more pronounced effect in going from 1 2 g/mi HC to 0 8 g/mi HC than any of the
other HC cutpoint increments This indicates that it might be worthwhile to consider
limiting the HC cutpoints to 1 2 g/mi for older vehicles (i.e., pre-1991) as a trade-off
between cutpoint stringency (and resulting failure rates) and emission benefits
Phase 2 cutpoints were also developed for each set of composite cutpoints, and the "two-ways-to-pass"
algorithm was employed in determining pass/fail rates
-22-
-------
Table 7
HM240 Failure Rates as a Function of Cutpoint for 1981 and Later
Passenger Cars Based on the 2% Random Sample Database
(17,000 Vehicle Sample)
Pollutant
Cutpoint
(g/mi)
All
Vehicles
1981 - 1984
Model Year
1985 - 1990
Model Year
1991+
Model Year
HC/CO/NOx
0.8/15/2 0
32%
80%
41%
6%
1 2/20/2 3
24%
69%
29%
3%
1 6/25/2 6
19%
60%
22%
2%
2 0/30/3 0
15%
51%
16%
1%
HC
08
23%
63%
29%
4%
1 2
15%
48%
18%
2%
1.6
11%
37%
12%
1%
2.0
8%
30%
8%
<1%
CO
15
15%
47%
17%
2%
20
11%
37%
12%
1%
25
8%
31%
9%
1%
30
7%
26%
7%
<1%
NOx
20
17%
F 47%
21%
2%
2.3
13%
39%
15%
1%
2.6
10%
32%
12%
1%
3 0
7%
25%
8%
1%
The false failure rates as a result of inadequate preconditioning were also estimated for the
cutpoints analyzed above. This was done by first determining the percentage of vehicles in
the replicate IM240 databases that passed an immediate retest based on these different
cutpoint scenarios.* (This is similar to the results presented in Tables 1 to 3 ) The results
*
The 1996 and 1997 replicate IM240 databases were combined for this analysis Overall, there were 328
vehicles (that passed the start-up IM240 cutpoints but failed the final IM240 cutpoints) m the 1996 database
and 336 vehicles (that failed the start-up standards only) in the 1997 database The failure rate in the current
program is expected to double when the final standards are implemented This implies that roughly half of the
failures would fail the start-up standards, while the other half would pass the start-up standards and fail the final
standards Thus, simply combining the 1996 and 1997 databases (which contain approximately the same
number of vehicles) should give a good representation of failing vehicles tn the fleet under the final standards
-23-
-------
from this analysis were then applied to the overall failure rates in Table 7 to arrive at a
false failure rate as a fraction of the fleet, which is given in Table 8
Table 8
Replicate IM240 Results and Estimated False Failure Rate
as a Function of EM240 Cutpoints
(664 Vehicle Sample)
Pollutant
Cutpoint
(g/mi)
# Failing
1st IM240
# Passing
2nd IM240
% Passing
2nd IM240
False Failures
as a Fraction
of the Fleet4
HC/CO/NOx
0 8/15/2 0
487
120
25%
8%
1 2/20/2 3
391
106
27%
6%
1 6/25/2 6
319
82
26%
5%
2 0/30/3 0
246
52
21%
3%
HC
0.8
375
140
37%
9%
1 2
267
95
36%
5%
1 6
184
64
35%
4%
2.0
126
32
25%
2%
CO
15
255
63
25%
4%
20
188
47
25%
3%
25
154
36
29%
2%
30
130
30
23%
2%
NOx
20
228
49
21%
4%
23
175
42
24%
3%
26
146
37
25%
3%
30
99
22
22%
2%
a False failures as a fraction of the fleet were derived by multiplying the fleet failure rate under each set of
cutpoints by the fraction of failing vehicles passing an immediate retest
-24-
-------
Of note in Table 8 is that the percentage of failing vehicles that passed an immediate retest
is relatively constant across the cutpoint scenarios analyzed here, with the more stringent
cutpoints generally having a larger fraction of vehicles passing the second IM240
However, because of the difference in fleet failure rates among the cutpoint scenarios, the
false failures as a fraction of the fleet are much lower for the less stringent cutpoints.
Modal Analysis of Replicate IM240 Testing
The emission results from the 336 failing vehicles tested in the replicate IM240 phase of
this study were also analyzed on a modal, or second-by-second, basis In addition, the
IM240 data from the replicate Phase 2 testing were evaluated on a modal basis. This
evaluation was motivated by the results of the 1996 preconditioning study that indicated it
was possible to successfully predict whether a vehicle was adequately preconditioned by
analyzing its second-by-second IM240 emission results
Mean Concentration Estimates - One of the parameters investigated in the 1996 study was
IM240 second-by-second emissions concentrations Although the IM240 is a mass-based
emissions test, it is possible to estimate tailpipe concentration from measurements of dilute
exhaust by calculating the dilution factor using equations contained in the high-tech
guidance. This technique provides an acceptable approximation for cars running near
stoichiometric * By comparing tailpipe concentrations near the end of Phase 1 to
concentrations near the end of Phase 2, it is possible to determine if a car was adequately
warmed up at the start of the test
The mean second-by-second HC concentration (in ppm C) for the 336 vehicles in the
replicate IM240 sample is illustrated in Figure 6 for the first 1M240 (which followed a 30-
minute idle period) and for the immediate retest As observed in the figure, the
concentration profile for the first test is higher than for the second test up to about second
200, where the two profiles converge (indicating that the vehicles, on average, are fully
warmed up at that point) The difference in HC concentration profiles is even more
pronounced for vehicles that failed the first and went on to pass the second test (based on
the start-up IM240 standards). This is illustrated in Figure 7. Although the concentration
profiles do not completely converge for these vehicles, it is clear that the vehicles are
warming up as the first test progresses This is also seen in Figure 8, which shows the
ratio of the first test HC concentration to the second test HC concentration for vehicles
failing the first and passing the second test (fail/pass) and for vehicles failing both tests
(fail/fail). As might be expected, the slope of the ratio with test duration is steeper for the
fail/pass sample than it is for the fail/fail sample
*
While the equation is considered acceptable for the purpose used in this analysis, the assumptions on which it
is based introduce some error Sources of error include the assumption that emissions are measured on a "wet"
basis (1 e, without the removal of water from the sample), the inadequate accounting for the effects of
pollutants in the background (dilution) air, and the fact that it becomes less accurate as the air-fuel ratio of the
vehicle under test deviates from stoichiometric Required adjustments to the equation to address these issues,
while feasible, are beyond the scope of this analysis and are not necessary at this time
-25-
-------
Figure 6
Mean Second-by-Second IM240 HC Concentration for
All Vehicles in the Replicate IM240 Sample
50
100 150
Time (sec)
200
250
¦1st IM240
¦2nd IM240
-Speed
Figure 7
Mean Second-by-Second IM240 HC Concentration for
Vehicles in the Replicate IM240 Sample
Failing the First Test and Passing the Second
Time (sec)
1st IM240
2nd IM240
~—--Speed
-26-
-------
Figure 8
Ratio of First Test/Second Test HC Concentations
For Vehicles in the Replicate IM240 Sample
by Pass/Fail Status
3.5
.v
• s." ;\
* * 1
r,j v *
60
-- 50
40 _
£
a
E
30 ~
«
a
a.
4- 20 w
-¦ 10
50
100 150
Time Sec)
200
L 0
250
Fail/Pass
Fail/Fail
- ——> Speed
Identifying Inadequate Preconditioning - The results of the 1996 preconditioning study
showed that emissions occurring during specific portions of the IM240 test were
significant in indicating whether the vehicle was adequately preconditioned. The following
four sections were of particular interest
• seconds 40 to 45 (for NOx concentration),
• seconds 75 to 80 (for HC and CO concentration),
• seconds 175 to 199 (for HC, CO, and NOx mass), and
seconds 209 to 214 (for HC, CO, and NOx concentration)
The designations of these particular segments of the IM240 speed-time trace are
consistent with EPA's convention of initiating data collection at second "0" and making
the last reading at second "239 " Each of the four sections of the IM240 trace is
illustrated in Figure 9 Other parameters determined to be important in whether a vehicle
passed a second test after failing the first were (1) average emissions over Phase 2 (i e.,
seconds 94 to 239), and (2) the difference between Phase 1 (i.e, seconds 0 to 93) and
Phase 2 emissions A detailed description of the retest algorithm developed in the 1996
preconditioning study can be found in the final report on the study,2 a brief summary is
provided below
-27-
-------
Figure 9
Sub-Sections of IM240 Trace
Used During Modal Analysis
For HC and CO emissions, the 1996 analysis indicated that one of the primary
determinants of whether a vehicle would pass a second IM240 after failing the first test
was the ratio of the average tailpipe concentration near the end of the first phase of the
test (seconds 75 to 80) to the tailpipe concentration near the end of the second phase
(seconds 209 to 214) These two segments of the trace were selected because they are the
last portions of each phase before the beginning of a deceleration. A high Phase 1/Phase 2
ratio indicates that the vehicle was not completely warmed-up in the first phase of the test.
Because it was found that many vehicles had very low NOx concentrations during seconds
75 to 80, regardless of whether the vehicle was fully warmed-up, NOx concentrations
computed during the acceleration from second 40 to second 45 were used in calculating
the concentration ratio between the start of the test and the end of the test, The same
portion of Phase 2 (i e , seconds 209 to 214) used for the denominator of the HC and CO
ratio was also used for the denominator of the NOx ratio
An example of a second-by-second HC concentration trace for a vehicle in the replicate
IM240 database that failed the first IM24Q but passed the second test is shown in
Figure 10 That vehicle, a 1991 Chevrolet truck, had composite IM240 emissions of 2 84
g/mi on the first test and 0 20 g/mi on the retest (This level of reduction from the first
IM240 to the second IM240 is on the high end of the spectrum for the vehicles tested in
-28-
-------
Figure 10
Second-by-Second HC Concentration
1991 Chevrolet Truck - 4.3 Liter Engine
(1st IM240 = 2.84 g/mi; 2nd IM240 = 0.20 g/mi)
Time (sec)
1st IM240 2nd IM240 ——Speed
this program, but it illustrates the extent to which preconditioning effects can impact the
test results.) Although both the first and second IM240 concentration traces are shown in
the figure, the important one is the first test, since that would be used in an operating
program to determine if the vehicle should receive a retest As observed in the first trace,
the ratio of the HC concentration during seconds 75-80 to the concentration during
seconds 209-214 is large (14 4) in this case. That information, coupled with the mean
concentration during seconds 209 - 214 being less than 1,500 ppmC, would trigger a
retest under the retest algorithm developed in the 1996 preconditioning study.
Application of Preconditioning Retest Algorithm to the Replicate IM240 and Replicate
Phase 2 Samples - The preconditioning retest algorithm developed in the 1996
preconditioning study was applied to the back-to-back IM240 data collected during this
study This included the 336 vehicles tested in the replicate IM240 phase of the program
(which were random IM240 failures based on the start-up cutpoints), and the 101 vehicles
in the replicate Phase 2 testing. (Recall that this sample of vehicles included passing and
failing vehicles that were tested on back-to-back IM240s following a 30-minute idle
period)
-29-
-------
The performance of the retest algorithm applied to the two databases generated for this
study is summarized in Table 9 As observed in that table, the retest algorithm was much
more successful at correctly identifying the failing vehicles that would benefit from an
immediate retest in the replicate Phase 2 sample than in the replicate IM240 sample In
the replicate Phase 2 sample, 88% (7 out of 8) of the failing vehicles that passed the retest
were correctly identified as needing a retest Only 60% (12 out of 20) of the vehicles
passing the retest in the replicate IM240 test sample were correctly identified as needing a
retest This result was not unexpected, however, since the 1996 preconditioning retest
algorithm was based on vehicles that had passed the start-up cutpoints and failed the final
cutpoints (i e, they were "marginal" failures under the final IM240 standards). Thus, one
would expect that algorithm to perform better on a random sample of failures than on a
sample that failed the start-up cutpoints on the initial test and passed the final cutpoints on
the retest.
Table 9
Performance of 1996 Preconditioning Retest Algorithm
in Identifying Vehicles Needing a Retest in the 1997 Databases
(1M240 Failures Based on Final Standards)
Database
Sample
Size
# Failing
1 st IM240
1st Test
Failures
Passing 2nd
Test
1st Test
Failures
Correctly DD'd
for a Retest
1st Test
Failures
Incorrectly ED'd
for a Retest
Replicate
IM240
336
336
20
12
20
Replicate
Phase 2a
101
24
8
7
3
a Note that the replicate Phase 2 database included back-to-back IM240 testing on the same vehicles following a
30-minute idle period.
Overall, the 1996 retest algorithm performed quite well, since it was designed to predict
when a vehicle failing the final standards would need a retest Assuming that roughly half
of the failures under the final standards would be "marginal" failures (i e , passing the
start-up cutpoints but failing the final cutpoints) and half would fail the start-up standards,
-30-
-------
it is estimated that 86% of those vehicles needing a retest would be correctly identified *
This is consistent with the results of the 1996 preconditioning study
Revisions to the Retest Algorithm Based on the Results of the 1997 Testing - Although
the results presented above indicate that the 1996 retest algorithm successfully identifies a
large fraction of vehicles needing a retest (i e., fail/pass vehicles) without identifying an
excessive number of fail/fail vehicles for retesting, the data collected in the 1997 test
programs made it possible to further refine the retest algorithm. This was particularly true
for light-duty trucks, which do not make up as large a fraction of the test sample (or the
fleet) as passenger cars The proposed revisions to the retest algorithm are outlined
below
• Passenger Cars
a. For vehicles failing HC and CO on the initial test, a retest is recommended if
massHC175_199 < 0 25 g and massC0175.199 < 5 0 g.
b For vehicles failing HC and NOx, a retest is noi recommended if ppmHC209-2i4
>1,200 or ppmNOx209.2i4 >1,200 (This is simply a change to the NOx
concentration from 1,000 ppm to 1,200 ppm.)
• Light-Duty Trucks
a For 1988 and later LDT2s failing NOx, a retest is recommended if
massNOx175_199 < 2 5 g.
*
Starting with 100 vehicles failing the final outpoints, approximately 50 would fail the start-up outpoints and
50 would pass the start-up outpoints (l e, these are marginal failures) Based on the results of the replicate
IM240 testing, 6% of the start-up cutpoint failures would pass the final standards on a retest (60% of these
would be identified by the retest algorithm) Based on the results of the 1996 preconditioning study, 50% of
the marginal failures would pass the final standards on a retest (88% of these would be identified by the retest
algorithm) These two sets of vehicles can be combined as follows
Vehicles Passing an Immediate Retest
Start-up Standard Failures 50 * 0 06 = 3
Marginal Failures 50 x 0 50 = 25
Vehicles Con-ectlv Identified as Needing a Retest
Start-up Standard Failures 3 * 0 60= 2
Marginal Failures 25 x 0 88 = 22
Using the above assumptions, of 100 vehicles failing the final IM240 standards on their initial test, 28 would
go on to pass an immediate retest, and 24 of those would be identified as needing a retest Thus, the 1996
retest algorithm correctly identifies 86% (24/28) of the vehicles needing a retest
-31-
-------
b For 1984 and later LDTs failing HC, a retest is not recommended if Phase 2
HC >32 g/mi
Applying the revised algorithm to the replicate IM240 sample resulted in an increase from
12 to 16 in the number of vehicles correctly identified as needing a retest, and a decrease
from 20 to 18 in the number of vehicles incorrectly identified as needing a retest (See
Table 9) These revisions caused no change to the results for the replicate Phase 2 sample
A complete summary of the IM240 retest algorithm, including the revisions recommended
above, is contained in Appendix A.
Conclusions and Recommendations from the Preconditioning
Analysis
Based on the analyses presented above, the following conclusions are reached
• The data collected in the current study support the previous conclusion that most
vehicles are not thoroughly warmed up by running Phase 1 of the IM240 test if
they have been waiting in a queue for 15-30 minutes prior to the test This is true
regardless of whether the engine is idled during queuing
• Using the current IM240 test procedures, it is estimated that 19% of the vehicles
currently failing the start-up IM24Q standards would pass those outpoints with
further preconditioning It is estimated that 6% of the vehicles failing the start-up
cutpoints would pass the final cutpoints if adequately preconditioned
• Newer (1986 and later model year) vehicles appear to be more sensitive to
preconditioning than the older vehicles
• A single Phase 2 of the IM240 test procedure is as effective as a full IM240 plus a
Phase 1 at preconditioning a vehicle that has been idling in a queue for 30 minutes.
• While Phase 2 of the IM240 is a much more effective preconditioning cycle than
Phase 1, it is unclear that I/M program managers would be receptive to eliminating
the current IM240 in favor of a Phase 2 cycle due to the high-speed operation
associated with Phase 2
It may be worthwhile to consider limiting the HC cutpoints to 1 2 g/mi for older
(i e, pre-1991) vehicles as a trade-off between cutpoint stringency and emission
benefits This would also have a beneficial impact on limiting false failures due to
inadequate preconditioning
• The results of the current study further support the modal analysis of emissions
recorded during the IM240 test as a way to identify vehicles that would benefit
-32-
-------
from further preconditioning It is estimated that the retest algorithm developed in
the 1996 study would correctly identify 86% of those vehicles needing a retest.
• Proposed refinements specific to passenger cars and light-duty trucks have been
developed for the 1996 retest algorithm
jj a a
-33-
-------
5. DEVELOPMENT OF IM240 FAST-PASS OUTPOINTS
This work assignment called for a reevaluation of IM240 fast-pass standards The task
included the evaluation of the methodological approach used in setting the current
standards, as well as an investigation of possible changes or enhancements to the previous
approach This section of the report presents an evaluation of previous methodologies
used to develop fast-pass standards and the results of the approach selected to develop
fast-pass standards under this work assignment. (Although not part of this work
assignment, the methodology used for the analysis could also be used to develop fast-fail
standards) After the methodology was developed, it was applied to the available second-
by-second IM240 data from the Arizona program to generate revised fast-pass standards
for a variety of different cutpoints.
Background
Because of the long duration of the IM240 test (relative to other I/M tests), there has been
interest on the part of states and I/M contractors in shortening the test time This interest
has grown as states have begun to focus on the need to offset the growth in test volumes
projected to result from (1) the increased failure rates that will occur as the states move
toward implementation of more stringent cutpoints, and (2) increases in the number of
vehicles subject to the IM240 testing in high growth areas such as Arizona Shortening
test times will help lessen queue time and the need to increase existing network capacity,
ultimately resulting in the need for fewer inspection lanes, and lower or at least stable
costs to motorists
One means of shortening the average test time of the IM240 is to implement "fast-pass"
and/or "fast-fail" standards in which vehicles with emissions that appear to be well below
or well above the full duration cutpoints would have a pass-fail decision made based on a
portion of the test. This possibility was recognized early in the development of EPA's
enhanced I/M regulations, and a discussion of fast-pass/fast-fail strategies was included in
the Regulatory Impact Analysis prepared for that rulemaking4 Based on an evaluation of
IM240 data available at that time, EPA estimated that roughly 33% of the vehicles tested
could be fast-passed or fast-failed based on results of Phase 1 of the IM240 (i e., the first
93 seconds of the test) EPA went on to point out that once second-by-second data were
collected, algorithms could be developed that would allow very clean cars to pass well
before the 93-second point
-34-
-------
Previous Methodology
Currently, every I/M program that has IM240 testing is using a fast-pass algorithm to
reduce average test time In addition, the Phoenix program is also using a fast-fail
algorithm The fast-pass cutpoints and methodology used in these programs are based on
work that EPA did about four years ago,* which is described in a letter from Gene Tierney
of EPA to the I/M community 5 In that method, the cumulative emissions (in grams) at
each second of the test are compared to the cutpoints (also in grams) established for each
second of the test The first point at which a vehicle can fast-pass is at second 30
Because of the "two ways to pass" criteria (i.e., different standards have been established
for the composite IM240 and Phase 2 of the test, with the thought that Phase 1 of the test
serves as a preconditioning cycle for vehicles that are not completely warmed up at the
start of the IM240), fast-pass standards have also been developed for Phase 2 of the test
Thus, it is possible for a vehicle that has cumulative IM240 emissions higher than the fast-
pass cutpoints for the composite test to fast-pass based on its cumulative emissions in
Phase 2 of the IM240 Although the original approach allowed a fast-pass determination
in Phase 2 to be made at the start of that phase (i.e, from second 94 forward), that
determination has been changed to second 109 forward in EPA's current high-tech
guidance document1 This is a positive revision, since seconds 94 to 97 are idle periods in
the IM240 Thus, making a fast-pass decision during that period is akin to spending a
significant amount of money on a very good preconditioning cycle, only to measure
emissions at idle
The fast-pass (and fast-fail) cutpoints developed by EPA were based on an analysis of
second-by-second data from 3,718 tests conducted in Arizona. According to the
documentation available on the development of the fast-pass/fast-fail cutpoints, the fast-
pass cutpoints for each second "represent the tenth lowest cumulative emission levels in
that second obtained for vehicles failing the IM240 using the two-ways-to-pass criteria."
Thus, vehicles that fall below these cutpoints have lower cumulative emissions than almost
all of vehicles that fail the full IM240 and, therefore, pass the test. Similarly, the fast-fail
standards were based on the highest cumulative emission levels from vehicles that passed
the full test. Thus, a vehicle failing these cutpoints has emissions that are higher than the
dirtiest vehicles that pass the full test
Alternative Methodologies
Although the aforementioned methodology used to develop fast-pass cutpoints provides a
workable solution to establishing fast-pass cutpoints, it does not make full use of the
information that has been collected on a second-by-second basis Rather, the fast-pass
Note that the fast-fail algorithm being used in the Arizona program is not based on the EPA cutpoints and
methodology The Arizona program uses an extremely conservative approach under which a vehicle fast-fails
only if its cumulative emissions exceed the total mass required to fail the entire test For example, if the full-
test cutpoint is 0 8 g/mi HC, then the total mass required to fail is 0 8 g/mi * 1 96 mi = I 57 grams Thus, if a
vehicle has emitted a total of 1 58 grams of HC at second 150, the vehicle fast-fails at that point
-35-
-------
decision is simply made by comparing cumulative emissions at each second to a particular
emission standard at each second With the intent of improving the performance of the
current fast-pass methodology and standards (i e, shorter test time and fewer false
passes), alternative approaches to predicting passing or failing results based on a portion
of the IM240 were investigated. The methodologies proposed by two different research
teams to predict lull-cycle IM240 scores from partial test results are summarized below.
This is followed by Sierra's recommendations for re-evaluating IM240 fast-pass outpoints
Resources for the Future fRFF^ - RFF has recently conducted an analysis6 of second-by-
second data collected in Phoenix to
• develop a methodology to use actual IM240 test data in program evaluations when
tests in the sample range in length from 31 to 240 seconds, and
• use that methodology to evaluate the accuracy of fast-pass and fast-fail algorithms.
The methodology developed by RFF, which is described below, estimates full IM240
emissions from partial test results using (1) the second at which the test ended, (2) the
cumulative emissions at that point, and (3) the model year of the vehicle
RFF developed a regression technique to predict full-test IM240 emissions from partial
test results This technique used full-duration, second-by-second data from 12,647 tests
conducted in the Phoenix program from January to June 1996 Using these data, a
regression was performed based on the following formula.
P240 = a + b*Px
+ cl*MY81 + c2*MY82 + - + cl5*MY95
+ dl*Px*MY81 + d2*Px*MY82 + - + d!5*Px*MY95
where P240 is the predicted full-duration IM240 score for pollutant P, Px is the
cumulative emissions at second x, and MY## is a dummy variable indicating whether the
test was performed on a vehicle of model year M (i e., this variable is assigned a value of
one for the model year of interest, otherwise it is zero) Overall, 627 equations were
developed by RFF based on the above methodology - one for each pollutant (HC, CO,
NOx) at each of 209 seconds (i e, tests ending after 31 seconds through 239 seconds)
Using this approach, full-cycle IM240 emissions can be predicted based on the length of
the test, the cumulative emissions at the end of the test, and the model year of the vehicle
Scatter plots showing predicted versus actual IM240 HC emissions using the RFF
technique are given in Figure 11 for tests ending at second 31, second 94, second 165, and
second 185 As expected, and as observed in that figure, the predictions improve as the
length of the test is increased.
-36-
-------
Figure 11
OJ
-J
45 degrees
• Fitted 240 Sac Cumul HC
45 degrees
• Fitted 240 Sec Cumul HC
30
20
10
I 1
10 20
Actual 240 Sec Cumul HC
—I-
30
30 -
20 -
10 "
0 -
1 1
10 20
Actual 240 Sec Cumul HC
Estimates using HC at second 31
Estimates using HC at second 94
30
45 degrees
• Fitted 240 Sec Cumul HC
45 degrees
• Filled 240 Sec Cumul HC
30
20 -
10 -
< I
10 20
Actual 240 Sec Cumui HC
Estimates using HC at second 165
30
30 -
20 -
10 -
0 -
i i
10 20
Actual 240 Sec Cumul HC
Estimates using HC at second 185
I
30
-------
New York Department of Environmental Conservation fNYDEQ - NYDEC has also
recently completed an analysis of IM240 data with the intent of predicting full-cycle scores
from partial test results 7 That work was motivated by a desire to
• reduce test time and queues (through the use of fast-pass/fast-fail algorithms),
and
• avoid the high-speed driving that occurs during the last 70 seconds of the test.
The methodology developed by NYDEC makes use of a regression technique in which
total cycle emissions are predicted from modal emissions data (rather than cumulative
emissions at each second) from the IM240 In that approach, the IM240 is broken up into
12 different modes, as shown in Figure 12, and the mass of each mode is calculated. A
regression is then performed according to the following equation:
- k * lix»£, ¦
E — E + £
IM240 IM240 c
where
Eim240 = predicted full-cycle IM240 emissions
Eim240 = actual full-cycle IM240 emissions
K = regression constant
C, = regression coefficient for mode I
E, = emissions in mode I (in grams)
j = number of modes in the model
e = error term
Using this approach, a different set of coefficients is developed based on the number of
modes included in the model (i.e., up to mode j); as the number of modes is increased, the
coefficient of determination (i.e , the t2 value) improves Because one of the goals of its
analysis was to eliminate the high-speed component of the test, NYDEC performed
regressions (and presented results) based on nine modes, or up through second 170 of the
trace That analysis was performed with data collected in the Phoenix I/M program and
EPA's Hammond, Indiana, IM240 pilot lane
-38-
-------
Figure 12
IM240 Speed-Time Trace
Modes Suggested by NYDEC
Scatter plots of predicted versus actual IM240 HC scores based on regressions performed
through mode 6 (or second 120 of the trace) are shown in Figure 13. As observed in the
figure, the model developed by NYDEC appears to be doing a very good job of predicting
full-cycle emissions from just one-half of the test, with the results from the Indiana
program being slightly better than results from the Phoenix program (in terms of the r2
value)
Overall, NYDEC concluded that the above methodology can provide both fast-pass and
fast-fail decision capability and has the advantage of providing an estimate of full-cycle
IM24G emissions, which is not possible with the EPA's current fast-pass/fast-fail
procedure. NYDEC also concluded that the minimum test time needed to make a fast-
pass/fast-fail decision using this technique is on the order of 120 to 170 seconds.
However, as discussed below, it appears that by using appropriate statistical safeguards, a
modal regression technique should be able to make a fast-pass decision with reasonable
certainty well before 120 seconds in many cases.
-39-
-------
Figure 13
IM240 G/Mi HC Emissions ~ Actual vs. Predicted
120-Second Modal Model
20
15
TJ
0)
1 10
-------
Selected Approach
In reviewing various approaches that might be used to develop a revised set of fast-pass
standards (or a revised fast-pass methodology), the method that had the most appeal (both
from an engineering and a statistical perspective) was the modal regression technique
developed by NYDEC Although the current use of outpoint tables could easily be
continued, such an approach essentially ignores valuable information collected during
different modes of vehicle operation that might allow a better prediction of how a
particular vehicle will perform on the entire test In addition, the use of modal data is
consistent with the methodology developed in the 1996 preconditioning study (described
in the previous section) to determine if a vehicle failing an IM240 test had been adequately
preconditioned. Thus, the methodology developed by NYDEC forms the basis of the
approach used to generate revised IM240 fast-pass standards under this study
The primary change that was made to the modal regression model described above was to
add modes and to shift some of the existing modes One reason for this change is that it
allows vehicles to fast-pass between the modes established in the NYDEC work, which
was expected to have a positive impact on reducing test time A total of 24 modes were
therefore selected as the basis of the regression model developed for this study These
modes are illustrated in Figure 14.
Figure 14
IM240 Speed-Time Trace
Modes Used for Development, of Fast-Pass Outpoints
-41-
-------
Because the regression statistics improve with the duration of the test (and therefore with
the number of modes included in the regression equation), a means to build a safety
margin into the fast-pass decision is desirable, especially for fast-pass decisions made early
in the test. This can be accomplished by first calculating the predicted full-cycle IM240
score from the regression coefficients, adding on a positive error term, and then
comparing the result to the full-cycle IM240 cutpoint (As written in the equation above,
the error term, e, can be positive or negative.) As the test progresses and more modes are
completed, the error term would decrease The same approach can also be taken in
determining compliance with Phase 2 of the test under the two-ways-to-pass criteria.
The error term is measured by the confidence limits, CL, for the predicted value of
Those confidence limits (with probability 1 - a) are given by the following equation 8
^1M240 ^M240 ± *aJ2,n-j-l a
; j
1 + - + EE c- *¦ x
n
i = l*= l
« «
ik *i Xk
The new terms introduced in this equation are defined below.
ta/2,n-j-i = the ordinate of the Student's t distribution for a two-tailed significance
level, a, and n-j-1 degrees of freedom
n = the number of data points in the IM240 data set used to generate the
regression coefficients K and C,
s^ = the standard regression error of the estimated IM240 emissions
c^ = an element of the inverse matrix of the matrix of the corrected sums of
squares and cross products used in determining the regression
coefficients
< = x;-*,
X, = the value of the P emission mode during the IM240 test of the vehicle
for which a fast-pass decision is to be made
X, = the average value of the i111 emission mode in the regression data set
For large regression sample sizes, the elements of the inverse matrix, c^, are quite small
and the double summation term can be neglected This was confirmed by a review of
these terms in the SAS output for an initial sample of regressions. Since the number of
observations is very large (on the order of several thousand), the 1/n term can also be
neglected Thus, the confidence limit, CL, on the predicted IM240 emissions can be
simply written as follows
-42-
-------
Since the value of the term for a 95% confidence level is approximately two for large
sample sizes, the error term can be estimated by simply doubling the standard error
estimate calculated for each regression equation.
To test this approach, a subset of the Arizona full-cycle second-by-second data was
initially used to develop regression coefficients (for HC) based on the modal methodology
described above with the modes outlined in Figure 14 Only light-duty gasoline vehicles
(i.e , passenger cars) were included in the analysis, and three model year groups were
considered'
• 1981 to 1984,
1985 to 1989, and
1990 and later.
It is interesting to note that the r2 values increase significantly as modes are added to the
regression equation As the r2 value increases, the value of s^ decreases. This means that
there is greater confidence in the regression prediction The method used for determining
the fast-pass described below uses the value s^ (multiplied by the t distribution ordinate)
to determine the safety factor Because of this, the safety factor is automatically adjusted
as the confidence in the predicted result increases
Once the regression coefficients were developed, predicted full-duration IM240 scores
(i e, Eii^o) were generated for a separate set of full-duration IM240 test results (i e.,
these were not used in developing the regression coefficients). At each mode (beginning
with mode 4, which ends at second 32 of the IM240 test), a fast-pass decision was made
based on the following
if (E^^ + CL) £ 0.8 g/mi HC, then the vehicle passes, otherwise continue the test.*
This approach was used to investigate how well the modal regression equations predicted
fast-pass results with a small sample of second-by-second full IM240 tests from the
Phoenix program. The results are presented in Table 10 Three different approaches to
developing the regression coefficients (and the corresponding standard errors) were
investigated.
• using all data within the model year groups;
• using only test scores that complied with the 0 8 g/mi HC composite cutpoint; and
• using only test scores that were less than or equal to 1 2 g/mi.
The value and of CL is different for each mode of the test
-43-
-------
Table 10
Performance of a Regression-Based Modal Emissions Algorithm
to Determine Pass/Fail Status Under a Variety of Constraints
(Based on a 0.8 g/mi HC EVL240 Cutpoint)
Model
Year
Method
Sample
Size
Correct
Passes
Mean Pass
Time
False
Passes
1981
to
1984
Modal - All Data
206
39
192.5
0
Modal z 1.2 g/mi HC
206
39
105 5
3
Modal * 0.8 g/mi HC
206
39
73.8
11
Current Cutpoint Table
206
39
80 6
9
1985
to
1989
Modal - All Data
214
100
156.4
1
Modal s 1 2 g/mi HC
214
100
80 8
2
Modal s 0 8 g/mi HC
214
100
45 6
11
Current Cutpoint Table
214
100
82.3
4
1990
and
later
Modal - All Data
127
107
52.5
2
Modal £ 1.2 g/mi HC
127
107
41.3
4
Modal s 0 8 g/mi HC
127
107
39.1
7
Current Cutpoint Table
127
107
63 3
2
Also shown in Table 10 are the results from using the current second-by-second cutpoint
tables.
As observed in Table 10, the performance of the modal methodology is dependent on
whether all vehicles are included in the regression estimates or whether only the cleaner
cars are included If all vehicles are included, then the error term is greater, and it
generally takes much longer to pass the test. (Recall that to fast-pass, the predicted
IM240 plus the error term is compared to the full-cycle cutpoint Thus, a larger error
term results in longer average test duration.) However, it appears that through the proper
construction of a modal model, the number of false passes can be reduced without a
significant increase in test time This is particularly true of the 1981 to 1984 model year
group.
-44-
-------
Development of Fast-Pass Standards
Based on the above evaluation, it was decided to use the modal regression technique to
develop a revised set of fast-pass standards for the IM240 test Full-duration, second-by-
second IM240 data collected in the Arizona I/M program were used for this analysis.
Nearly 110,000 individual tests were in the database used in the analysis, which is
comprised of all full-duration IM240 tests conducted in Arizona from April 1995 through
April 1997 Consistent with the methodology presented above, cutpoints (or, more
correctly, regression coefficients) were generated separately for light-duty gasoline
vehicles (i e., passenger cars) and light-duty trucks for the following model year groups
• 1981 to 1984,
• 1985 to 1989, and
1990 and later
Regression coefficients were developed for HC, CO, and NOx and for both the composite
IM240 and for Phase 2 of the IM240 The Phase 2 regressions used mode 11 (see
Figure 14) as the first mode and continued through mode 23 The composite IM240
regressions used modes 1 through 23 (Although 24 modes are shown in Figure 14, if a
fast-pass decision is not made by mode 23, then the vehicle would run the full IM240. At
that point, a pass/fail decision should be made on the actual IM240 score, not the
predicted score) Finally, it is recommended that the first mode at which a pass/fail
decision should be made is mode 4 (which ends at second 32 of the IM240) for a
composite IM240 prediction, or mode 13 (which ends at second 113 of the IM240) for a
Phase 2 prediction.
The regression coefficients for a 0 8 g/mi HC composite IM240 cutpoint are given in
Appendix B, along with the coefficients for a 0 5 g/mi HC Phase 2 IM240 cutpoint. The
full series of regression coefficients developed in this effort were provided to EPA
electronically, and a listing of the corresponding cutpoints is contained in Appendix B
Although the next section of the report presents a detailed analysis of the impacts of the
revised cutpoints (and preconditioning procedures) on test time, it is useful to review
summary results at this point Using the 2% Random Sample from the Arizona program
(which consists of 26,000 records), pass/fail rates were calculated with the modal
regression procedures outlined above as well as the current fast-pass cutpoint tables This
analysis was performed using the final IM240 HC, CO, and NOx standards, and the results
are presented in Table 11
As observed in Table 11, the revised fast-pass methodology results in a lower fraction of
false passes than the current method, particularly for older cars However, this
improvement in failing vehicle identification is offset by a longer average test time for
passing vehicles in the older model year groups For newer vehicles (i e, 1990 and later),
the revised methodology results in significant improvements in average test time, without
a significant increase in the fraction of false passes
-45-
-------
Table 11
Comparison of Current and Revised Fast-Pass Methodologies
Under the Final IM240 Standards
(26,000 Vehicle Sample)
Vehicle
Class
Model
Year
Group
"True"
Failure
Rate0
Current Fast-Pass
Revised Fast-Pass
Failure
Rate
Pass Time
(seconds)
Failure
Rate
Pass Time
(seconds)
LDV
81-84
79%
76%
125
78%
157
85-89
45%
41%
130
43%
121
1990+
8%
7%
88
7%
57
LDT
81 -84
62%
51%
71
60%
113
85-89
42%
35%
70
40%
93
1990+
9%
7%
60
7%
57
a The "true" failure rate is based on full-duration 1M240 test scores
-46-
-------
6. IMPACT OF PRECONDITIONING AND
CUTPOINTS ON TEST DURATION
Task 2 of Work Assignment #2-04 called for an evaluation of the impact on test cycle
duration of implementing EPA's final IM240 cutpoints coupled with the need for
preconditioning to avoid false failures This section presents the results of an analysis of
failure rates and test cycle duration under the current start-up cutpoints, the final
cutpoints, and a set of "interim" cutpoints that fall between the start-up and final
cutpoints In addition, failure rates and test cycle duration were also evaluated for the
revised modal cutpoints described in Section 5 of this report. Finally, the impact on test
duration of implementing the preconditioning procedures described in Section 4 of this
report is presented in this section
IM240 Failure Rates Using the Current Fast-Pass Procedures
The first step in determining the effect of tighter cutpoints on test cycle duration (and the
resulting impact on lane throughput) was to evaluate IM240 failure rates under the current
program and under alternative cutpoints Failure rates directly impact the total load on an
IM240 network by virtue of the retests required - a higher failure rate translates into
additional retests In addition, there may be differences in average test time for vehicles
subject to a retest compared to the average test time of initial tests.
Table 12 summarizes the light-duty gasoline vehicle (LDGV) IM240 initial test failure
rates under the current Arizona program (which is using the start-up cutpoints outlined in
the high-tech guidance document) Those failure rates were calculated using all IM240
tests conducted from January 1996 through September 1996 Also contained in Table 12
is a summary of the IM240 failure rates based on a subset of the Arizona database that
received a full IM240 test (rather than a fast-pass or fast-fail). This subset is flagged in the
Arizona database with a '2', indicating it is the 2% random sample of vehicles that are
given a full IM240 test (The 2% Random Sample used in this analysis consisted of
vehicles tested from June 1995 to April 1997) Three sets of cutpoints were applied to
this sample
start-up cutpoints,
• "interim" cutpoints, and
• final cutpoints.
-47-
-------
Table 12
LDGV DM240 Initial Test Failure Rates in the Arizona Program
Using the Current Fast-Pass Procedures (No Preconditioning)
Model
Year
Group
Pollutant
Current
Program
(Start-Up)8
Cutpotnt Stringency
(2% Sample Results)5
Start-Up
Interim
Final
1981
to
1983
Any
45%
42%
62%
78%
HC
31%
29%
48%
65%
CO
17%
16%
26%
37%
NOx
19%
20%
31%
45%
Sample Size
28,000
1,400
1984
to
1987
Any
25%
24%
39%
55%
HC
14%
14%
28%
42%
CO
11%
11%
19%
27%
NOx
11%
12%
19%
30%
Sample Size
90,000
4,600
1988
to
1990
Any
7%
6%
14%
25%
HC
3%
3%
9%
17%
CO
2%
2%
5%
9%
NOx
4%
3%
7%
12%
Sample Size
78,000
4,000
1991
and
Later
Any
2%
2%
5%
5%
HC
1%
1%
3%
3%
CO
1%
1%
2%
2%
NOx
1%
1%
2%
2%
Sample Size
139,000
7,000
All
Vehicles
Any
13%
12%
21%
29%
HC
8%
8%
15%
22%
CO
5%
5%
9%
13%
NOx
6%
6%
10%
16%
Sample Size
335,000
17,000
a Based on an analysis of all test records from January 1996 to September 1996
k Based on an analysis of the 2% random sample of full IM240 tests from June 1995 through Apnl 1997
-48-
-------
The start-up and final cutpoints used in this analysis are those recommended by EPA in the
high-tech guidance document, while the interim cutpoints were chosen to fall between the
start-up and final cutpoints To determine the pass/fail status of each vehicle, the second-
by-second cumulative mass data from the 2% Random Sample were compared to the
existing fast-pass standards (corresponding to the appropriate composite and phase 2
IM240 cutpoints) at each second (beginning at second 30 in the trace) If a vehicle ran the
full test without being fast-passed, the pass/fail determination was made based on the
composite and phase 2 IM240 gram per mile values. The fast-pass cutpoint tables for the
current (start-up) cutpoints were obtained from Gordon-Darby, while the fast-pass tables
in the high-tech guidance document were used for the final cutpoints In some cases, the
interim cutpoints required interpolation between the start-up and final fast-pass cutpoint
tables, while in other cases cutpoint tables from the start-up or final cutpoints (from other
vehicle classes or model year groups) could be used directly Because of the low failure
rates of 1991 and later model year vehicles under the final cutpoints, the final cutpoints
were used in the interim cutpoint scenario for those model years.
The results in Table 12 indicate that the 2% Random Sample failure rates for the start-up
cutpoints are nearly identical to the failure rates observed in the entire fleet Thus, the 2%
Random Sample appears to be a very good representation of the fleet (Note that the 2%
Random Sample used in this analysis, which contains only initial tests, consisted of 26,000
vehicles The database used to determine failure rates for the current program consisted
of 522,000 initial tests) Also of note in Table 12 are two obvious findings (1) failure
rates increase for older model year vehicles, and (2) failure rates increase under the more
stringent cutpoints reflected in the interim and final cutpoints Overall, initial-test 1M240
failure rates for LDGVs are estimated to increase from 12% to 29% with the final
cutpoints if nothing is done to deal with possible false failures as a result of inadequate
preconditioning Under the interim standards evaluated in this effort, the overall LDGV
failure rates are estimated to increase from the current 12% to 21%
Similar results are also observed with light-duty gasoline trucks 1 (LDGT1, i.e, under
6,000 lbs GVWR) and with light-duty gasoline trucks 2 (LDGT2, i e, from 6,000 lbs to
8,500 lbs GVWR). A summary of the 1M240 failure rates for LDGVs, LDGTls, and
LDGT2s is given in Table 13 The sample sizes used to compute the failure rates for each
vehicle group are also shown As observed, the failure rates for the light-duty trucks are
slightly lower than for LDGVs This is partially a result of different cutpoints (and new
vehicle certification standards) that apply to passenger cars versus light trucks
-49-
-------
Table 13
1M240 Initial Test Failure Rates by Vehicle Class in the Arizona Program
Using the Current Fast-Pass Procedures
(No Preconditioning)
Vehicle
Class
Pollutant
Current
Program
(Start-Up)a
Cutpoint Stringency
(2% Sample Results)b
Start-Up
Interim
Final
LDGV
Any
13%
12%
21%
29%
HC
8%
8%
15%
22%
CO
5%
5%
9%
13%
NOx
6%
6%
10%
16%
Sample Size
335,000
17,000
LDGT1
Any
8%
7%
14%
20%
HC
4%
4%
7%
14%
CO
2%
2%
3%
6%
NOx
3%
4%
8%
10%
Sample Size
147,000
7,000
LDGT2
Any
9%
9%
19%
24%
HC
6%
6%
10%
19%
CO
4%
3%
5%
8%
NOx
3%
3%
11%
11%
Sample Size
40,000
2,000
All
Vehicles
Any
11%
11%
19%
26%
HC
7%
6%
12%
20%
CO
4%
4%
8%
11%
NOx
5%
5%
10%
14%
Sample Size
522,000
26,000
a Based on an analysis of all test records from January 1996 to September 1996
k Based on an analysis of the 2% random sample of full IM240 tests from June 1995 through April 1997
-50-
-------
IM240 Failure Rates Using Full IM240 Test Scores
To serve as a check on how well the existing fast-pass cutpoints correctly identify vehicles
passing the full EM240 test, failure rates from the 2% Random Sample were also
calculated based on using the full IM240 test scores, rather than the fast-pass procedures
outlined above The results from this analysis are presented in Table 14, which shows that
there is a moderate number of errors of omission (i.e , vehicles failing the full IM240 that
pass the fast-pass cutpoints) when using the current fast-pass cutpoints Under both the
start-up and final IM240 cutpoints, using the full IM240 test scores to determine pass/fail
status increases the overall failure rates by about 3 percentage points. Also of note in
Table 14 is that the fraction of errors of omission is higher for NOx than for HC and CO
This is likely the result of the fast-pass algorithm passing vehicles before the highest speed
portion of the IM240 is entered (i e , beginning at second 156 of the trace, the vehicle is
accelerated from 26 mph to nearly 57 mph at second 200 of the trace), where one might
expect a large contribution of NOx emissions to the overall IM240 NOx score
1
Table 14
IM240 Initial Test Failure Rates in the Arizona Program
Current Fast-Pass Methodology Versus Full IM240 Test Scores
from the 2% Random Sample Vehicles8
(26,000 Vehicle Sample with No Preconditioning)
Pollutant
Start-Up Cutpoints
Final Cutpoints
Fast-Pass
Full IM240
Fast-Pass
Full IM240
Any
11%
14%
26%
29%
HC
6%
7%
20%
21%
CO
4%
5%
11%
12%
NOx
5%
7%
14%
16%
a These results are based on an analysis of the 2% random sample of full IM240 tests from June 1995 through
April 1997
Impact of Revised Fast-Pass Procedures on IM240 Failure Rates
and Total Test Time
The revised fast-pass procedures (i e , the modal analysis methodology described in
Section 5) were applied to the 26,000 initial test scores in the 2% Random Sample
database to determine how well that procedure correctly identified passing vehicles The
-51-
-------
results of that analysis are summarized in Tables 15, 16, and 17 for the start-up, interim,
and final IM240 standards, respectively Results are presented separately for light-duty
trucks (i e, LDGT1 and LDGT2 combined) and light-duty vehicles (i e., passenger cars),
and for three model year groups 1981 to 1984, 1985 to 1989, and 1990 and later Also
shown in Tables 15, 16, and 17 are the same results based on the current fast-pass
methodology and cutpoints.
Using Table 15 as a guide, the third column (marked " 'True' P/F") is the pass/fail result
based on the full-duration test The fourth column (marked "Fast-Pass P/F") is the
pass/fail result based on the fast-pass algorithm Vehicles receiving a "P" in column 3 and
a "P" in column 4 were correctly identified as passing vehicles by the fast-pass algorithm.
Vehicles receiving an "F' in column 3 and a "P" in column 4 are failing vehicles that were
incorrectly passed by the fast-pass algorithm (i e., errors of omission or false passes).
Vehicles with an "F' in both columns ran the full IM240 under the fast-pass algorithm and
were correctly logged as failures Also shown in the tables is the average test time for the
vehicles fast-passing the test (in seconds), as well as the total test time (in hours) for all
vehicles in the sample (i e , this is simply the number of vehicles in each category
multiplied by the average test time). At the bottom of each table is the total test time
under the fast-pass scenario being investigated as well as the number of incorrect passes
and the number of true failures.
The results presented in Tables 15 to 17 indicate that the revised fast-pass algorithm does
a better job of identifying passing vehicles without increasing overall test time Although
the fast-pass test duration increases for the older model year groups, this is generally
offset by a shorter fast-pass test duration for the newer model year groups, which have a
larger number of passing vehicles. One way to look at the false passes is as a fraction of
the total "true" failing vehicles (i.e, based on full-duration IM240 scores) Thus, under
the start-up cutpoints (the results of which are summarized in Table 15), the current fast-
pass algorithm results in 745 false passes out of 3,522 "true" failures, i.e, 21% of the
failing vehicles are incorrectly fast-passed On the other hand, the revised fast-pass
algorithm results in only 10% of the failing vehicles being incorrectly fast-passed, even
though the overall test time did not increase appreciably (i e, 563 versus 566 hours of
total dynamometer time)
A summary of the false passes and the total test time for the three sets of cutpoints
analyzed in this study is shown in Table 18. It is interesting to note that the revised fast-
pass algorithm results in an approximate 50% reduction in the number of false passes
under each of the cutpoint scenarios shown in Table 18 In addition, with the exception of
the start-up standards, the revised algorithm results in an overall decrease in total test time
(or dynamometer time) relative to the current fast-pass methodology This occurs despite
the fact that there are more failing vehicles (which run the full 240 seconds) under the
revised fast-pass algorithm (Note that only fast-pass algorithms were investigated in this
study; fast-fail approaches were not evaluated)
-52-
-------
Table IS
Comparison of Current and Revised Fast-Pass Methodologies
Under the Start-Up IM240 Cutpoints
(26,000 Vehicle Sample)
Fast-Passes Based on Current Second-by-Second Cutpoints
Test Tune
Total Tune
Vehicle Class
MY Group
"True" P/P
Fast-Pass P/F
Sample Size
(sec)
(hours)
81-84
F
F
162
240
108
F
P
74
100
2 1
P
P
701
54
104
85-89
F
F
393
240
26 2
LDT
F
P
129
82
29
P
P
2,538
62
440
1990+
F
F
144
240
96
F
P
52
112
1 6
P
P
4,876
49
66 7
81-84
F
F
922
240
61 5
F
P
209
96
56
P
P
1,163
68
21 9
85-89
F
F
939
240
62 6
LDV
F
P
235
114
7 4
P
P
5,183
57
82 0
1990+
F
F
217
240
145
F
P
46
114
1 5
P
P
8,043
59
1317
Total Time (hrs)
563 0
Incorrect Passes
745
True Failures
3.522
Fast-Passes Based on Revised Modal Analysis Methodology
Test Time
Total Tune
Vehicle Class
MY Group
"True" P/P
Fast-Pass P/F
Sample Size
(sec)
(hours)
81-84
F
F
220
240
14 7
F
P
16
71
03
P
P
701
73
14 1
85-89
F
F
486
240
32 4
LDT
F
P
36
70
07
P
P
2,538
66
46 7
1990+
F
F
153
240
10 2
F
P
43
71
08
P
P
4,876
44
59 0
81-84
F
F
1,084
240
72 3
F
P
47
104
1 4
P
P
1,163
105
33 9
85-89
F
F
1,045
240
69 7
LDV
F
P
129
82
2 9
P
P
5,183
67
96 2
1990+
F
F
188
240
12 5
F
P
75
78
1 6
P
P
8,043
43
96 6
Total Time (hrs)
566 1
Incorrect Passes
346
True Failures
3,522 |
8 "True" pass/fail status is based on the full-duration IM240 score
-53-
-------
Table 16
Comparison of Current and Revised Fast-Pass Methodologies
Under the
"Interim" IM240 Cutpoints
(26,000 Vehicle Sample)
Fast-Passes Based on Current Second-by-Second Cutpoints
Test Time
Total Time
Vehicle Class
MY Group
"True" P/P
Fast-Pass P/F
Sample Size
(sec)
(hours)
LDT
81-84
F
F
339
240
22 6
F
P
132
98
36
P
P
466
60
77
85-89
F
F
707
240
47 1
F
P
206
92
53
P
P
2,147
54
32 2
199CH-
F
F
300
240
20 0
F
P
97
103
28
P
P
4,675
58
75 3
LDV
81-84
F
F
1,359
240
90 6
F
P
135
139
52
P
P
800
94
20 9
85-89
F
F
1,717
240
1145
F
P
241
142
95
P
P
4,399
92
1125
1990+
F
F
495
240
33 0
F
P
68
153
29
P
P
7,743
83
178 3
Total Time (hrs)
784 0
Incorrect Passes
879
True Failures
5.796
Fast-Passes Based on Revised Modal Analysis Methodology
Test Tune
Total Time
Vehicle Class
MY Group
"True" P/P
Fast-Pass P/F
Sample Size
(sec)
(hours)
LDT
81-84
F
F
445
240
29 7
F
P
26
111
08
P
P
466
100
12 9
85-89
F
F
869
240
57 9
F
P
44
87
1 1
P
P
2,147
84
49 9
1990+
F
F
326
240
21 7
F
P
71
88
1.7
P
P
4,675
56
72 2
LDV
81-84
F
F
1,457
240
97.1
F
P
37
128
1 3
P
P
800
130
28 8
85-89
F
F
1,802
240
120 1
F
P
156
99
43
P
P
4,399
84
103 0
1990+
F
F
466
240
31 1
F
P
97
84
23
P
P
7,743
54
1160
Total Tune (hrs)
751 9
Incorrect Passes
431
True Failures
5.796
* "True" pass/fail status is based on the full-duration IM240 score
-54-
-------
Table 17
Comparison of Current and Revised Fast-Pass Methodologies
Under the Final IM240 Cutpoints
(26,000 Vehicle San
•Pie)
Fast-Passes Based on Current Second-by-Second Cutpoints
Test Tune
Total Tune
Vehicle Class
MY Group
"True" P/P
Fast-Pass P/F
Sample Size
(sec)
(hours)
LDT
81-84
F
F
474
240
316
F
P
109
98
3.0
P
P
354
63
62
85-89
F
F
1,062
240
70 8
F
P
217
103
62
P
P
1,781
66
32.8
1990+
F
F
348
240
23 2
F
P
83
113
26
P
P
4,641
59
76 1
LDV
81-84
F
F
1,732
240
1155
F
P
88
144
35
P
P
474
122
160
85-89
F
F
2,597
240
173 1
F
P
253
162
11 4
P
P
3,507
128
124 6
1990+
F
F
606
240
40 4
F
P
81
162
36
P
P
7,619
87
185 1
Total Time (hrs)
925 7
Incorrect Passes
831
True Failures
7.650
Fast-Passes Based on Revised Modal Analysis Methodology
Vehicle Class
MY Group
"True" P/P
Fast-Pass P/F
Sample Size
Test Tune
Total Tune
(sec)
(hours)
LDT
81-84
F
F
560
240
37 3
F
P
23
114
07
P
P
354
113
11 1
85-89
F
F
1,210
240
80 7
F
P
69
101
19
P
P
1,781
93
462
1990+
F
F
371
240
24 7
F
P
60
94
1 6
P
P
4,641
57
73 5
LDV
81-84
F
F
1,791
240
1194
F
P
29
157
1 3
P
P
474
157
20 6
85-89
F
F
2,702
240
180 1
F
P
148
124
5 1
P
P
3,507
121
1174
1990+
F
F
563
240
37 5
F
P
124
100
34
P
P
7,619
56
1195
Total Time (hrs)
882 2
Incorrect Passes
453
True Failures
7,650
a "True" pass/fail status is based on the full-duration IM240 score
-55-
-------
Table 18
Summary of False Passes and Total Test Time
for the Current and Revised Fast-Pass Algorithms
(Based on 26,000 Vehicles from the Arizona 2% Random Sample)
Cutpoints
Current Fast-Pass Algorithm
Revised Fast-Pass Algorithm
False Passes®
Test Time (hrs)
False Passes1
Test Time (hrs)
Start-Up
21%
563
10%
566
Interim
15%
784
7%
752
Final
11%
926
6%
882
0 Reported as a percentage of the "true" IM240 failures based on the full-duration test scores
Test Cycle Duration for the Current Program
The analysis of test time under the current and revised fast-pass algorithms presented
above provided only an estimate of initial test gross dynamometer time for each of the
cutpoint scenarios investigated However, under an operating program, vehicles being
retested (after failing their initial test) would also be subject to fast-pass standards, and it
is not clear that those vehicles would have the same fast-pass characteristics as vehicles
undergoing their initial test. Thus, this section presents an analysis of test cycle duration
for the current program, evaluating test times for both initial tests and retests In addition,
the current program in Arizona makes use of a fast-fail algorithm as well as a fast-pass
algorithm, and the average test times for failing vehicles under the current program were
also analyzed in this effort
To determine the average test time for the current program, the January 1996 to
September 1996 database analyzed above was used Test time is included in the test
record, so it was a fairly straightforward process to determine average test cycle duration
This was estimated for passing and failing vehicles on their initial test and on retests, and
the results are presented in Table 19 by vehicle class For initial tests, the average test
time for passing vehicles in the current Arizona program is 62 seconds, and the average
test time for failing vehicles is 183 seconds On retests, the average test time for passing
vehicles is 82 seconds, and for failing vehicles it is 191 seconds
An analysis of test time was also performed with the 2% Random Sample database (which
consists of initial tests only) The results are shown in Table 19 for the start-up cutpoints,
the interim cutpoints, and the final cutpoints using the current fast-pass procedures. Only
passing vehicle test time estimates are given in Table 19 for the 2% Random Sample, since
an analysis of test times for vehicles failing the start-up and final cutpoints under the
Arizona fast-fail algorithm (based on the 2% Random Sample) indicated nearly identical
-56-
-------
Table 19
IM240 Test Time Duration Under the Current (i.e., Second-by-Second)
Fast-Pass Procedures (Seconds)
Vehicle
Type
Test
Type
Current Program®
2% Random and 3% Repair Sample
Passing Vehicles Onlyb
Failing
Passing
Start-Up
Interim
Final
LDGV
Initial
182
66
60 ^
89 1
103
Retest
191
88
81
106
125
LDGT1
Initial
187
54
47
57
63
Retest
193
71
181
74
78
LDGT2
Initial
186
58
83
66
67
Retest
191
74
176
71
76
Ail
Vehicles
Initial
183
62
59
78
88
Retest
191
82
109
94
105
a Based on a 522,000-vehicle sample
k Based on a 26,000-vehicle sample for initial tests and a 12,000-vehicle sample for retests
values (i e, 180 seconds) as failures in the current program.* For passing vehicles, the
average test time increases as the cutpoints become more stringent
Table 19 also summarizes retest duration under the three sets of cutpoints evaluated in this
analysis Those results are based on an evaluation of the second-by-second data in the 3%
Repair Sample from Arizona. (This sample consists of vehicles that fail their initial test
and receive a retest[s] after repair Vehicles in the sample are tested over the full IM240)
A comparison of test duration for all vehicles in Table 19 reveals that the average initial
test time of the current program is reasonably consistent with the results from the 2%
Random Sample However, the results for retests are quite different, particularly for light
trucks It is unclear why this occurs, although it may be due to the smaller sample sizes of
the light trucks or the fact that the 3% Repair Sample is not a completely random mix of
vehicles (Included in the 3% Repair Sample are failing vehicles that were not flagged as a
'3' at the start of the test, but became a '3' because they ran the full IM240 and failed )
Intuitively, one might expect the fast-fail test time under more stringent cutpoints to be less than under the
current cutpoints. However, the algorithm used in Arizona requires that a decision be made (either fast-pass or
fast-fail) for each pollutant before the vehicle receives a fast-pass or fast-fail for the overall test Thus,
although a vehicle might fast-fail early in the test for one pollutant, the test will continue until a decision is
made for each pollutant A review of a Limited number of test records indicated that the fast-pass decision for
one pollutant was often the limiting factor in ending a test where a vehicle fast-faded for another pollutant
-57-
-------
Because of this difference, the retest times from the current fleet were used in the analyses
that follow
Impact of Tighter Cutpoints and Preconditioning on Test Times and
Lane Throughput With the Current Fast-Pass Procedures
The failure rates contained in Table 13 were combined with the test duration estimates in
Table 19 to arrive at the overall impact that tighter cutpoints will have on IM240 test
duration and lane throughput using the current fast-pass procedures In addition, the
combined impact of more stringent cutpoints and vehicle preconditioning procedures was
also included in this analysis A description of these estimates follows.
Impact of Tighter Cutpoints - Before evaluating the impact of more stringent cutpoints on
IM240 test time, it was necessary to estimate the average test time of the current program.
This was accomplished by using 1,000 initial tests as the basis, and generating the total
test time for the following components:
• Initial test passes (890 tests x 60 seconds per test)
• Initial test failures (110 tests * 180 seconds per test)
• Retest passes (99 tests x 80 seconds per test)
• Retest fails (77 tests * 190 seconds per test)
The failure rate of 11% for the current program came from Table 13, and the test times
are from Table 19 (after rounding to multiples of 10) The fraction of vehicles subject to
retests was estimated from the January 1996 to September 1996 data, which contained
59,000 initial test failures, 52,000 retest passes, and 42,000 retest failures. Thus, it was
assumed that approximately 90% of the initial test failures ultimately pass a retest (i.e.,
52,000 - 59,000), and the total volume of retest failures is approximately 70% (42,000 -
59,000) of the initial failures In actuality, a fair number of vehicles will pass the first
retest, while others will fail several retests. It should be noted that these are only
approximations that reflect an average of the fleet tested from January to September 1996
A more rigorous analysis would track individual vehicles (e g, by VIN) from initial test
until they pass, receive a cost waiver, or remove themselves from the program
Average dynamometer test time is estimated to be 81 seconds in the current program A
total of 1,176 tests are conducted for every 1,000 initial tests (This ignores any special
testing that might be conducted and does not attempt to remove the 2% Random Sample
nor the 3% Repair Sample from the analysis.) Multiplying total tests by average test time
results in a total dynamometer testing time of 26 6 hours per 1,000 initial tests (This does
not include the time during which no vehicle is in the lane or when other functions are
occurring, e g, collection of vehicle information from the motorist.) A summary of the
above calculations is presented in the "Current Program" column shown in Table 20
-58-
-------
Table 20
Summary of IM240 Test Time Per 1000 Initial Tests
Using the Current Fast-Pass Procedures
(Fail Rates Based on 26,000 Vehicle Sample)
No Preconditioning
With Preconditioning
Current Program
Interim Cutpoints
Final Cutpoints
Interim Cutpomts
Final Cutpoints
Parameter
No of
Tests
Tune
(sec)
No of
Tests
Tune
(sec)
No of
Tests
Time
(sec)
No of
Tests
Tune
(sec)
No of
Tests
Tune
(sec)
Initial Tests
1000
1000
1000
1000
1000
Initial Pass/No Precond
890
60
810
80
740
90
810
80
740
90
Precondition Pass
...
...
...
48
350
65
350
Initial Faii/No Precond
110
180
190
180
260
180
104
240
143
240
Precondition Fail
...
—
—
38
420
52
420
Retest Pass
99
80
171
80
234
80
121
80
161
80
Precondition Pass
—
—
...
7
350
14
350
Retest Fail
77
190
133
190
182
190
73
240
100
240
Precondition Fail
—
—
...
27
430
36
430
Total Tests/Time
1176
95750
1304
137950
1416
166700
1227
163584
1312
203052
Average Test Time
81
106
118
133
155
Net Testing Time Per 1000
Initial Tests (Hours)
26 6
38 3
46 3
45 4
56 4
-------
Similar estimates were prepared for the interim outpoints and the final outpoints, and the
results are also included in Table 20 Results are shown for both sets of outpoints with
and without preconditioning 1,000 initial tests are assumed for each analysis scenario
Under each scenario, two columns are shown The first is the number of each particular
type of test (e.g, initial pass, initial fail, retest pass, etc) that would be conducted The
second column contains the average dynamometer test time (in seconds) estimated for
each type of test
For the scenarios involving preconditioning, there are additional test types that are called
either "Precondition Pass" or "Precondition Fail " The numbers shown in these rows
represent the number and overall test times of vehicles that are projected to fail an initial
IM240 and then be flagged for an immediate retest using modal data from the initial test
The times shown for these tests include both the initial test and immediate retest For
example, under the interim cutpoints it is projected that 48 out of 1,000 vehicles would fail
the initial test but then be flagged for retesting and pass an immediate retest An
additional 38 vehicles would fail and be flagged for retesting, but not pass the retest. 106
other vehicles would fail and not be flagged for retesting. The remaining 810 vehicles (out
of the initial 1,000) would pass the initial test Further details on this subject are provided
below
Average dynamometer test time (in seconds) for each scenario is shown on the next-to-
the-final row of the table The final row shows the total dynamometer test time (in hours)
associated with the 1,000 initial tests for each scenario, including all retests
Based on the analysis of the 2% Random Sample database presented m Table 19, it was
assumed that vehicles passing the initial test would run 80 seconds under the interim
cutpoints and 90 seconds under the final cutpoints The initial test failure rate was
assumed to be 19% under the interim cutpoints and 26% under the final cutpoints (from
Table 13) The same fraction of passing and failing retests (i e , 90% of initial test failures
and 70% of initial test failures, respectively) was assumed for the interim and final
cutpoints Unfortunately, no data exist with which to develop independent retest rates for
the interim and final cutpoints (which would be a function of how well the repair industry
could identify and repair vehicles failing the more stringent cutpoints) Finally, it was
assumed that the test time for vehicles failing the more stringent cutpoints would be
approximately the same as for vehicles failing the current (i e., start-up) cutpoints, which,
as described above, was verified through an analysis of the 2% Random Sample
Significant increases in average dynamometer test time, number of total tests per 1,000
initial tests, and total dynamometer testing time per 1,000 initial tests are predicted when
going to the more stringent cutpoints analyzed in this effort Implementation of the final
IM240 cutpoints, based on the current fast-pass cutpomt tables and without considering
preconditioning procedures, is estimated to result in a 20% increase in total tests, a 46%
increase in average dynamometer test time, and a 74% increase in total dynamometer
testing time
-60-
-------
Impact of Tighter Cutpoints and Preconditioning - Based on the results of the 1996
preconditioning study previously conducted by Sierra and Gordon-Darby, it became clear
that implementation of EPA's final IM240 cutpoints would result in a significant number
of false failures due to inadequate preconditioning That study also recommended several
preconditioning procedures that would minimize the fraction of false failures Based on
one of those options (i e, using modal data from a full IM240 to determine whether a
vehicle should be retested), the impact of more stringent cutpoints and preconditioning
procedures on EM240 test time was estimated.
As noted above, 1,000 initial tests were used as the basis for this calculation, the results of
which are summarized in Table 20 Using the final IM240 cutpoints as an example, the
following assumptions were made
• The initial test failure rate of 26% (from Table 13) was reduced to 19.5% based
on the finding that approximately 25% of the vehicles failing the final IM240
cutpoints would be false failures The test time for those vehicles that would
pass after preconditioning (i.e, 6 5% of the initial tests) was assumed to be the
initial 240 seconds plus 110 seconds. The 110-second test time of the second
test was based on an analysis of back-to-back IM240 tests performed on passing
vehicles that failed the first IM240. (This data set was analyzed in our previous
analysis of preconditioning )
• The test time for initial test failures was assumed to be 240 seconds for vehicles
that did not receive a second IM240 after failing the first. Based on the
algorithm developed to determine if a failing vehicle would benefit from an
immediate retest, it was found that approximately 20% of the initial test failures
flagged for a retest would not pass a second test Thus, it was assumed those
vehicles would run the full 240 seconds for the initial test and then would be
subject to a fast-fail algorithm It was assumed that the test time of the second
failing test would be equal to the average failing test in the current program
• The number of retest passes and retest fails was assumed to be the same as
described above (i e , 90% of the initial test failures and 70% of the initial test
failures, respectively). The split between those retests requiring preconditioning
and those not requiring preconditioning was assumed to be the same fraction as
observed in the initial tests (This is very simplistic and may understate the
fraction of retested vehicles needing preconditioning) Test times were
determined in the same manner as described above for the initial tests.
A similar approach was taken in estimating the test time increase for the interim cutpoints
coupled with preconditioning procedures In that case, it was also assumed that 25% of
the initial test failures would pass after a retest (see Section 4 of this report)
The implementation of more stringent cutpoints and preconditioning procedures results in
a significant increase in average test time and the total testing time per 1,000 initial tests
-61-
-------
However, the preconditioning procedures reduce the total number of tests relative to the
more stringent outpoints alone That is because there are fewer failures (and therefore
fewer retests) when the preconditioning procedures are implemented Overall,
implementation of EPA's final IM240 cutpoints with appropriate preconditioning
procedures to reduce false failures is estimated to increase the number of tests required by
12%, the average dynamometer test duration by 91%, and the total dynamometer testing
time by 112%
Overall, the net effect on test lane throughput will depend on how much time is required
for other activities at the lane position where the dynamometer test occurs and how much
time is required at the other lane positions In a three-position lane with a total of
3 minutes in each position, increasing the dynamometer testing time from 81 seconds to
155 seconds would increase the time in the position with the dyno test from 180 seconds
to 254 seconds, reducing the maximum throughput from 20 vehicles/hour to 14.2/hour A
41% increase in lane capacity would be required to maintain the same system-wide
maximum throughput It is possible that a smaller increase in capacity would be sufficient
if certain functions could be shifted from the dyno test position to another lane position
For a two-position lane with 5 minutes in the dyno test position and 3 minutes in the other
position, the same increase in test time would reduce the maximum throughput rate from
12/hour to 9.6/hour. A 25% increase in lane capacity would be needed to maintain
system-wide throughput, unless certain functions could be shifted to the other lane
position or a third position is created
Another factor that could affect the impact of the estimated increase in total testing time
on IM240 network capacity requirements under more stringent cutpoints and
preconditioning procedures is the existence of any "slack time" that can accommodate the
additional tests and increased test time estimated above According to Gordon-Darby,
however, the current Arizona IM240 test network is at or near capacity during peak
periods when queues develop Therefore, moving to more stringent cutpoints and
preconditioning procedures is expected to have an impact on EM240 network
requirements consistent with the calculations described above
Finally, it is important to note that the failure rates and test times estimated above were
based on the current fast-pass cutpoint tables, results from a similar analysis prepared with
the revised fast-pass methodology are presented below
Impact of Tighter Cutpoints and Preconditioning on Test Times and
Lane Throughput With the Revised Fast-Pass Procedures
To evaluate the impact of tighter cutpoints and preconditioning procedures on test times
under the revised fast-pass procedures, an analysis similar to that described above was
performed. This first required a more detailed analysis of mean test times with the revised
fast-pass procedures than that presented earlier in this section The results of that analysis
are shown in Table 21, which closely parallels Table 19 in format Comparing the results
-62-
-------
in Table 21 to those in Table 19, one observes that the average initial test time for passing
vehicles under the revised fast-pass procedures is 3 seconds lower than the current
procedures for the start-up cutpoints, 8 seconds lower for the interim cutpoints, and
11 seconds lower for the final cutpoints
Table 21
EM240 Test Time Duration Under the Revised (i.e., Modal)
Fast-Pass Procedures (Seconds)"
Vehicle
Type
Test
Type
2% Random and 3% Repair Sample
Passing Vehicles Only
Start-Up
Interim
Final
LDGV
Initial
57
70
81
Retest
96
104
113
LDGT1
and
LDGT2
Initial
53
67
70
Retest
66
98
101
All
Vehicles
Initial
56
70
77
Retest
85
102
108
a Initial test results based on a 26,000-vehicle sample, retest results based on a 12,000-vehicle sample
As described above, the revised fast-pass procedures result in an approximate 50%
reduction in false passes (i.e, errors of omission) Thus, the EM240 failure rate would
increase under the revised procedures relative to the current procedures at the same
cutpoint level The failure rates with the revised fast-pass procedures were therefore
assumed to be midway between the current fast-pass failure rate and the failure rate
calculated from the full-duration IM240 scores (see Table 14 of this section) These
failure rates, along with the test duration estimates shown in Table 21, were used to
generate estimates of total testing time per 1,000 initial tests for the revised fast-pass
procedures with and without a preconditioning algorithm Those results are summarized
in Table 22, which is formatted identically to Table 20.
In comparing Tables 20 and 22, one observes that the total testing time per 1,000 initial
tests is very similar between the current fast-pass procedures and the revised fast-pass
procedures. Although the revised fast-pass procedures result in shorter test duration for
passing vehicles, the failure rate is higher than under the current fast-pass algorithm This
results in a longer test duration for the additional initial test failures (i e, a full 240
seconds) and in more failing vehicles that must be retested
-63-
-------
Table 22
Summary of IM240 Test Time Per 1000 Initial Tests
Using the Revised (Le., Modal) Fast-Pass Procedures
(Fail Rates Based on 26,000 Vehicle Sample)
No Preconditioning
With Preconditioning
Current Program
Interim Outpoints
Final Outpoints
Interim Cutpoints
Final Cutpoints
Parameter
No of
Tests
Time
(sec)
No of
Tests
Time
(sec)
No of
Tests
Time
(sec)
No of
Tests
Time
(sec)
No of
Tests
Time
(sec)
Initial Tests
1000
1000
1000
1000
1000
Initial Pass/No Precond
875
60
795
70
725
80
795
70
725
80
Precondition Pass
—
...
...
51
350
69
350
Initial Fail/No Precond
125
180
205
180
275
180
113
240
151
240
Precondition Fail
—
—
—
41
420
55
420
Retest Pass
113
80
185
100
248
110
130
100
169
110
Precondition Pass
—
—
...
8
350
16
350
Retest Fail
88
190
144
190
193
190
79
240
106
240
Precondition Fail
—
...
...
29
430
39
430
Total Tests/Time
1201
100760
1329
138410
1441
171450
1246
165114
1330
207674
Average Test Time
84
104
119
132
156
Net Testing Time Per 1000
Initial Tests (Hours)
28 0
38 4
47 6
45 9
57 7
-------
Conclusions from the Analysis of Test Duration
Based on the results presented above, the following conclusions are reached
• Moving to more stringent cutpoints will result in a significantly higher number of
failing vehicles, which will increase the average time per 1M240 test as well as
the total number of tests (due to retests of failing vehicles).
• Implementing preconditioning procedures to reduce the number of false failures
will decrease the total number of tests. However, because the preconditioning
procedures eliminate the possibility of fast-failing vehicles on the initial test
(since the retest decision is based on a modal analysis of the entire IM240 test),
the total time per test increases significantly This results in an overall increase in
total test time per 1,000 initial tests when preconditioning procedures are
implemented.
• Use of the revised (modal) fast-pass procedures developed in this study results in
a shorter test duration and fewer false passes relative to the current procedures
The shorter test duration is offset by the higher number of failing vehicles at a
given set of cutpoints, resulting in very similar estimates of total test time per
1,000 initial tests for both the revised and current fast-pass procedures
// If a
jrrHt
-65-
-------
7. ASSESSMENT OF IM240 SPEED VARIATION
CRITERIA
In addition to preconditioning analysis, an evaluation of the effects of speed variations
currently allowed in IM240 testing was conducted. Under Tasks 3 and 4 of the Work
Assignment, modal Arizona IM240 data were analyzed to identify alternative statistical
measures that better isolated emissions-affecting speed variations in the IM240 test traces
This section of the report discusses the results of that analysis and presents recommended
alternative statistical criteria to the current standard error (SE) tolerance
Examination of Current Criteria
Background - The prescribed driving cycle for the transient IM240 test consists of varying
second-by-second speeds ranging from zero (i e., idle) to 56 7 mph, with maximum speed
changes of ±3 3 mph/sec. During actual IM240 testing, the driver watches a graphical
display of the prescribed or "reference" speed/time IM240 trace overlayed with the actual
second-by-second trace as it is being driven as an aid to following the reference trace and
anticipating upcoming speed changes (The visual display also indicates prescribed shift
points for manual transmission vehicles along the trace )
Since each vehicle has different performance characteristics, it is impossible, even for
highly skilled drivers, to precisely follow the second-by-second reference trace speeds
during actual "one-time-only" testing As a result, EPA has specified two types of speed
tolerance limits that define the "leeway" allowed to the driver in trying to follow the
reference trace for the test to be considered valid: (1) speed excursion limits, and (2)
speed variation limits Each of these criteria are described below9
Speed Excursion Limits (85.2221 (e) (4)1 - Speed excursion limits shall apply as
follows:
(I). The upper limit is 2 mph higher than the highest point on the trace within
J second of the given time.
(u). The lower limit is 2 mph lower than the lowest point on the trace within 1
second of the given time.
-66-
-------
(Hi). Speed variations greater than the tolerances (e.g., during gear changes)
are acceptable provided they occur for no more than 2 seconds on any
occasion.
(iv). Speeds lower than those prescribed during accelerations are acceptable
provided they occur for no more than 2 seconds on any occasion.
Speed Variation Limits [85.2221 (e) (5)]
(i). A linear regression of feedback value on reference value shall be
performed on each transient driving cycle for each speed using the method
of least squares, with the best fit equation having the form: y = mx+ b,
where:
(A). y = the feedback (actual) value of speed;
(B). m = the slope of the regression line;
(C). x = the reference value; and
(D). b = the y-mtercept of the regression line.
(ii). The standard error of estimate (SE) ofy on x shall be calculatedfor each
regression line. A transient driving cycle lasting the full 240 seconds that
exceeds the following criteria shall be void and the test shall be repeated:
(A). SE = 2.0 mph maximum.
(B). m = 0.96 -1.01.
(C). r2 = 0.97 minimum.
(D). b= ±2.0 mph.
EPA suspended the use of the speed variation limits on November 23, 1993, pending
further evaluation.
Analysis of Current Criteria - Under this Work Assignment, the effectiveness of the speed
variation criteria was examined (No analysis of the speed excursion limits was
performed.)
EPA's IM240 test procedures guidance includes "placeholders" for yet-to-be-defined criteria that apply to
fast-pass or fast-fail tests that end prior to 240 seconds The development of second-by-second speed variation
criteria for tests that end before 240 seconds was one of the goals of this work assignment
-67-
-------
Sierra's analysis investigated the adequacy of the linear regression-based statistical
tolerances contained in the speed variation limits to identify emissions-producing speed
variations. It was found that although the current regression statistics provide an adequate
means for ensuring that actual IM240 speeds generally follow the reference trace, they
don't adequately identify speed variations that can dramatically affect emissions Sierra
constructed a series of three hypothetical, easily driven 1M240 speed traces, all of which
complied with the current regression-based tolerances (and the speed excursion limits), as
follows.
1 Smooth - a trace that generally lagged the peaks and valleys of the
reference IM240 trace in a manner that "smoothed" their amplitudes to
allowed limits;
2. Very Aggressive - a trace that contained a number of high-frequency
oscillations (over 20) about the reference trace between the speed
excursion limits, characteristic of extremely "jerky" driving behavior, and
3 Less Aggressive - a variation of the Very Aggressive trace in which the
number of high-frequency oscillations were roughly halved and their
durations were doubled to reflect mildly aggressive behavior
Figures 15 through 17 graphically illustrate each of these traces. (The thinly dashed lines
in each of these figures show the speed excursion limits "envelope," indicating that each of
these traces also meets the speed excursion criteria) "Least-squares" linear regression
statistics (i e., SE, m, r2, and b) based on second-by-second speeds in each trace against
the reference IM240 trace were calculated and compared to the allowed speed variation
tolerances Table 23 shows these comparisons
Table 23
Regression Statistics of Speed Variation-Corn
>liant EM240 Traces
Trace
SE
(Std Error)
m
(Slope)
r2
(Correlation)
b
(Intercept)
Smooth IM240
1 99
0 97
0 98
-0 56
Very Aggressive IM240
1 86
1 00
0 99
0 11
Less Aggressive IM240
1.96
0 99
0 98
0 89
Current Allowed
Limits
<2.0
0.96 to 1.01
>0.97
<±2.0
-68-
-------
Figure 15
"Smooth" Speed Variation-Compliant IM240 Trace
' ^ i ~ xC*
_^ jr^~
V.
h
*'/j
i/
'rr\
%
V,
//y
iff
w/ *y
V-.
-J A
•It
//
A',
-=uJ
X 'J
.'XZJ,
V
0 20 40 60 SO 100 120 140 160 180 200 220 240
Time (sec)
| IM240 Reference Trace Upper Excursion Limit Lower Excursion Limit ———"Smooth Trace |
Figure 16
"Very Aggressive" Speed Variation-Compliant IM240 Trace
**
a*
tlf
f
JQ*.
\*.
rlt »
//*•
/!/
'V
\WT> il
»/
v '£/•'
%
-f
-V
0 20 40 60 80 100 120 140 160 180 200 220 240
Time (see)
| IM240 Reference Trace Upper Excursion Umrt Lower Excursion Limit Very Aggressive Trace |
Figure 17
'Less Aggressive" Speed Variation-Compliant IM240 Trace
V-,
j _i
Si*
%v
it
fr
•V,
k'
-------
As shown in Table 23, all of the linear regression-based statistics for each of these
hypothetical traces are within the allowed limits of the current speed variation criteria, and
thus would be considered valid traces if driven in an IM240 test lane.
To demonstrate the inadequacy of speed variation criteria based on these linear regression
statistics, IM240 emissions were measured for the Smooth and Very Aggressive traces as
well as the reference IM240 trace when driven over on a chassis dynamometer Each of
these second-by-second traces were input into a "driver's aid" display similar to that used
in IM240 test lanes and then driven by a single, well-trained driver for two fully warmed,
late-model vehicles in good condition - a 1997 Ford Taurus and a 1996 Honda Accord
The driver was instructed to follow each trace as well as possible
The measured emission differences between the Reference, Smooth and Very Aggressive
traces were significant, as illustrated in Figure 18 When driven aggressively but still
within current tolerance limits, average IM240 emissions for the two vehicles were 2-5
times higher than when the same vehicles were driven over the reference trace. In
addition, moderate CO and NOx differences were observed for the Smooth trace in
comparison to the Reference trace emissions.
Figure 18
Comparison of IM240 Emission Test Results for
Current Speed Variation-Compliant Traces
0 12
0 00
IM240
IM240-Smooth
Driving Trace
IM240-Aggressive
~ HC QCO + 10 ONOx
-70-
-------
The point of this limited emission testing was not to represent the magnitude of emissions
variations occurring in actual IM240 testing under the current speed variation limits, but
to merely demonstrate, by example, the inadequacy of these regression-based statistics to
identify high emissions-producing speed variations. This it does clearly
To better understand why these regression statistics cannot isolate high-emission speed
variations, Figure 19 presents a comparison of second-by-second 1M240 emissions against
speed compiled from a large sample of Arizona IM240 tests (To equate the scales of
HC, CO and NOx, the emissions shown at each second represent the sum of mean HC and
NQx emissions (in grams) plus CO divided by 10 as compiled from the Arizona test data.)
Emissions are plotted against the left axis, speed against the right
Figure 19
Mean Second-by-Second Arizona IM240
Emissions vs. Speed
(Sample Size = 16,581)
v
—T—
50
Emissions = HC + NOx + (CO/10)
100 150
Time (sec)
200
-Emissions
-Speed
250
The sample of modal Arizona IM240 data used throughout this speed variation analysis consisted of 16,581
tests conducted between January 1995 and October 1996 in which full 240 second IM240's were run
(irrespective of fast-pass status) constituting the "2% random full-test" element of the program.
-71-
-------
Figure 19 shows that the high-emission sections of the transient IM240 test (e g, seconds
10-15, 100-110 and 160-170) generally occur during periods of high acceleration (or
moderate acceleration at high speeds) This is not surprising The behavior of late-model
emission control systems under high speed and load conditions has become much better
understood in recent years10. Under high engine loads, vehicles can go into "open loop"
operation in which fuel enrichment can result in much higher emissions than under normal
engine loads, even with a fully warmed engine and catalyst.
The current speed variation statistical tolerances are based on the relationship of actual to
reference speeds and weight speed deviations equally over the range of the IM240 trace
They cannot be expected to isolate the critical "high-emissions" portions of the trace
which are more directly affected by engine load.
Figure 20 helps to illustrate this point. It shows a scatter plot of actual versus reference
IM240 speeds for the hypothetical Smooth trace described earlier, along with the
calculated "least-squares" regression line
Figure 20
Scatter Plot of Actual vs. Reference Speeds for
"Smooth" IM240 Trace
20 30 40
Reference Speed (mph)
• Actual vs. Reference Speeds —Linear (Actual vs. Reference Speeds)
-72-
-------
The standard error (SE) statistic represents the sum of the squared vertical deviations of
each of the data points to the regression line (and divided by the number of data points)'1
Mathematically, the standard error is given by the following equation
where^ are the actual data values (i e., actual IM240 speeds), ^is the regression line, and
N is the number of data points (in this case, 240). As seen in Figure 20, a given deviation
at low speeds is weighted the same as an equal deviation at high speeds in calculating the
standard error. In other words, an actual speed of 12 mph for which the reference speed
is 10 mph yields a 4 mph (22) squared deviation, which is equally weighted with a similar
deviation for an actual speed of 52 mph when the reference speed is 50 mph The
standard error statistic does not give "appropriate" higher weighting to speed deviations
occurring at the critical high-emission points along the IM240 trace, and thus (along with
the other regression statistics used in the current speed variation criteria), is ill-suited to
identifying those speed deviations that substantially affect IM240 emissions.
The Work Assignment called for an evaluation of the efficacy of reducing the allowable
standard error. However, as this analysis indicates, reducing the allowed tolerance will
not further identify high-emission speed variations Therefore, no further evaluation of
modifying the regression-based tolerances was conducted Instead, other statistical
measures were investigated, as an alternative to the current regression parameters as
explained in the following section.
Evaluation of Alternative Statistical Measures
Based upon similar work conducted by the New York Automotive Emissions Laboratory
(AEL,)12 two easily determined, alternative statistical metrics were evaluated for their
ability to identify and quantify IM240 speed variations that significantly affect emissions:
(1). DPWRSUM - the sum of absolute changes in specific power, and
(2). Positive Kinetic Energy flPKE^ - the sum of positive differences in kinetic
energy per unit distance
Each of these metrics are explained in more detail below
DPWRSUM - Specific power is defined as power per unit mass, wtuch can be restated as
follows.
„ power work Akinetic energy \ xmassxAV2
specific power = = = =
mass mass x ktime mass x Atime mass x ktime
-73-
-------
Over a transient driving cycle of second-by-second speeds, EPA defines the specific power
P at any time t (and dropping the factor of Vi) as
P,= V2-Vt-2
The absolute difference in specific power at time t can then be written as
&Pt = \P,-Pl-i\ = \Vt2-2V,-i2+ V<-22
The DPWRSUM statistic then is defined over a cycle of # seconds as
N N
DPWRSUM = 1AP, = I \V,2 - 2 V, -1! + V, - 22
/=0 f=0
PKE - Positive Kinetic Energy has been defined mathematically13 as
N
IPPl
pke=m,—
f dx
0
over a traveled driving cycle of distance x where PP is the positive specific power and is
given by
PPt=V,2-Vi-i2
when Vt > Vul, and is zero when Vt z Vt_t
Each of these metrics can be easily computed from the second-by-second speed
measurements collected in IM240 testing. In comparing their relative ability to identify
speed variations that produce high emissions, it is helpful to consider which speed
variations contribute to the value of each metric (similar to the earlier examination of the
SE statistic) over an IM240 test.
Note that although both DPWRSUM and PKE are affected by differences in specific
power or squared speeds over "adjacent" seconds of an IM240 trace, the value of
DPWRSUM is increased during decelerations as well as accelerations. PKE on the other
hand, is only increased during acceleration periods. This was an important distinction in
evaluating which metric was best-suited to "tracking" significant emission-producing
speed variations from the reference trace.
-74-
-------
Figure 21 contains a comparison of second-by-second IM240 emissions against specific
power compiled from the sample of Arizona 1M240 tests cited earlier (Figure 21 is
identical to Figure 19 presented earlier, except specific power is displayed instead of
speed)
Figure 21
Mean Second-by-Second Arizona IM240
Emissions vs. Specific Power
(Sample Size = 16,581)
Time (sec)
Emissions Specific Power j
Note that during intervals of negative specific power (i.e., decelerations), such as between
seconds 20-25, 85-95, and 195-205, mean emissions for those periods are much lower
than emissions during positive specific power intervals (e.g., between seconds 10-15, 100-
110, and 160-170). In other words, unlike DPWRSUM, the value of the PKE metric is
not "diluted" with speed variations during decelerations that have little effect on
emissions, and is only affected during accelerations Depending on the pollutant, 70%-
80% of the overall average IM240 cycle emissions from the Arizona data sample were
measured during periods of acceleration
Further evidence of the superiority of the PKE statistic over DPWRSUM is provided in
Table 24, which compares the values of both statistics for each of the hypothetical speed
variation-compliant IM240 traces described in the previous sub-section to the
corresponding values for the reference IM240 trace
-75-
-------
Table 24
Comparison of Alternative Statistics for
Speed Variation-Compliant 1M240 Traces
Trace
PKE
(miles/hr2)
DPWRSUM
(mph2/sec)
Reference IM240
3,269
6,370
Smooth IM240
3,272
5,915
Very Aggressive IM240
5,865
18,245
Less Aggressive IM240
3,987
5,815
As shown, both metrics clearly distinguish the Very Aggressive trace from the IM240
reference values The PKE statistic also "properly" represents the behavior of the Less
Aggressive trace, quantifying it between the Very Aggressive and Reference traces.
However, the DPWRSUM value for the Less Aggressive trace is even lower than that of
the Smooth trace Although the PKE statistic does not clearly distinguish the Smooth
trace from the reference, as shown earlier in Figure 18, emissions over Smooth trace are
not likely to differ notably from emissions over the Reference trace It was therefore
concluded that PKE is a better measure than DPWRSUM for identifying those IM240
speed variations from the reference trace that significantly affect measured emissions
The next sub-section describes the methodology used to develop IM240 PKE variation
limits to replace the current speed variation limits and presents the results.
Development of PKE-Based Variation Limits
The basic approach used to develop PKE-based speed variation criteria for the IM240
consisted of the following elements.
1 establishing upper and lower "composite" PKE limits for full 240-second tests
from the Arizona data sample; and
2. scaling these composite limits based on the cumulative PKE at each second of
the IM240 reference trace to produce second-by-second PKE limits.
Development of Composite Limits - As seen from the hypothetical "Aggressive" IM240
traces discussed earlier, values of PKE for a full 240-second IM240 test which
significantly exceed the reference value of 3,269 miles/hr2 may be indicative of higher
measured emissions than if the reference trace were more closely followed In turn, PKE
-76-
-------
values for full IM240 traces that are well below the reference PKE may identify "under-
measured" emissions.
Figure 22 shows the distribution of composite (i e, 240-second) PKE calculated from the
second-by-second actual speeds in the Arizona data sample, expressed as the percent
difference between actual and reference IM240 PKE As the figure shows, actual PKE
appears normally distributed, although the median PKE is approximately 1% higher than
the reference value
Figure 22
Distribution of Arizona IM240 PKE Differences
(% Difference Between Reference and Actual PKE, Sample Size = 16,581)
20% i
18%
-10% -8% -6%
Note Sierra data are normalized
-2% 0% 2% 4% 6%
Percent Difference from Reference Value
To determine how far to go along the "tails" of the PKE differences distribution to set
composite limits, PKE differences among the test lane drivers in the Arizona data were
examined. The basic concept applied in setting the PKE limits was to identify a significant
fraction of drivers who, historically, could always (or nearly always) run IM240 tests
within the selected PKE limits Given a mixture of ability among individual drivers to
follow the reference trace, Sierra sought to identify the fraction of "competent" drivers
who could follow the trace more consistently than others when conducting IM240 tests
for a range of vehicles This subset of competent drivers and tests was used to establish
composite PKE limits
-77-
-------
The Arizona IM240 data sample (16,581 tests) was first sorted into groups by individual
driver * A total of 502 drivers were found in the sample Driver groups containing fewer
than 25 tests were then discarded, this left a total of 247 driver groups, which
encompassed 86% of the tests in the total sample (i e before discarding small-sample
driver groups). Composite PKE was then calculated for each test in the remaining driver
groups. The mean and standard deviation of PKE from the tests within each driver group
were also computed. The driver groups were then ranked by increasing PKE standard
deviation and the top 50% of the drivers (based on lowest PKE standard deviation were
used to compute possible composite PKE cutpoints. Table 25 lists a series of possible
PKE cutpoints computed from the percentile PKE variance among the top 50% drivers
For example, the PKE cutpoints shown for the 2% row under the "Top 50% Percentile"
column indicates that- 96% (100% - 2 x 2%) of the tests from the top 50% drivers had
composite PKE within 3,148 and 3,522 miles/hr2.
Table 25
Preliminary PKE Cutpoints (miles/hr2) Based on Top-50% Drivers
(Sample Size = 6,749 Tests)
Top-50% Driver
Percentile
Low-End PKE
(miles/hr2)
High-End PKE
(miles/hr2)
Interval Width
(miles/hr2)
±0 5%
3,104
3,592
488
±1.0%
3,125
3,551
426
±2 0%
3,148
3,522
374
±3.0%
3,161
3,500
338
±4 0%
3,170
3,483
312
±5 0%
3,178
3,469
291
±6.0%
3,186
3,459
273
±7.0%
3,193
3,449
256
±8.0%
3,198
3,441
232
±9 0%
3,203
3,435
220
Note that these preliminary cutpoints are not centered about the IM240 reference PKE
value of 3,269 miles/hr2 (as evidenced by the PKE distribution shown earlier in Figure 22)
To generate a series of "final" cutpoints for evaluation, Sierra applied the interval widths
Identified by the 4-digit "Inspector©" field in the database
-78-
-------
shown in Table 25 to the reference PKE value to produce "centered" cutpoints about the
reference value.
Incremental abort test rates for each set of centered cutpoints were then calculated based
on both the entire 16,581 test Arizona data sample and the top-50% driver subset The
results are presented in Table 26, which lists both "simple" and "effective" abort rates
Simple abort rates represent the fraction of tests in the sample for which the composite
PKE cutpoints would have been exceeded Effective abort rates were calculated from
simple rates by subtracting the fractions of emission-pass tests that exceeded the upper
PKE cutpoint. The idea is that tests on vehicles that had passing emission scores but were
driven with high PKE should not be aborted. In addition to the calculated lest abort rates,
Table 26 also shows the percentage of drivers who are always within the limits of each set
of PKE cutpoints
Table 26
Centered PKE Cutpoints and Resulting Test Abort Rates
Top-50%
Driver
Percentile
Centered PKE Limits (mi/hi3)
Abort Rates (%) from
All Tests
Abort Rates (%) from
Top-50% Driver Tests
Percentage of All
Drivers Within Limits
Lower
PKE
Limit
Upper
PKE
Limit
Interval
Half-
Width
Simple
Abort
Rate
Effective
Abort
Rate
Simple
Abort
Rate
Effective
Abort
Rate
> Lower
Limit
< Upper
Limit
±0 5%
3,025
3,513
243 9
6 5%
1 2%
2 4%
0 3%
95 8%
48 0%
±1 0%
3,056
3,482
213 0
9 3%
1 8%
4 2%
0 6%
91 8%
40 4%
. V
>;
vCfe-
iKSt
^ :W.
±3 0%
3,099
3,438
169 2
16 1%
3 7%
9 0%
1 6%
799%
28 7%
±4 0%
3,113
3,425
156 2
18 8%
4 5%
11 3%
2 1%
75 3%
26 1%
±5 0%
3,123
3,414
145 5
21 5%
5 2%
13 6%
2.7%
71 5%
23 9%
±6 0%
3,132
3,405
136 3
24 1%
6 0%
15 7%
3 1%
67 3%
21 5%
±7 0%
3,141
3397
1281
26 7%
6 9%
18.2%
3 8%
63 3%
20 3%
±8 0%
3,147
3390
121 4
29 0%
7 6%
20 3%
4 3%
60 4%
19 1%
±9 0%
3,153
3385
1159
30 8%
8 3%
222%
4 9%
58 0%
18 5%
Sample
Size
-
-
-
16,581 Tests
6,749 Tests
502 Drivers
Based on the results given in Table 26, Sierra proposes the use of composite lower and
upper PKE limits of 3,082 and 3,456 miles/hr2, respectively (shown in the shaded row in
Table 26) As indicated in the table, these composite PKE limits would increase the abort
test rate by just under 3% based on historical test data If drivers are selected based on
their ability to perform as well as the best 50% of the current drivers, then the abort rate
-79-
-------
would drop to just over 1%. In practice, it is expected that the abort rate will increase by
less than this amount as drivers "adjust" to the new limits.
Development of Second-bv-Second PKE Limits - Using the recommended composite
PKE limits of 3,082 and 3,456 miles/hr, second-by second limits were generated by scaling
the percentage difference of these limits from the composite reference value (5 7%) to the
cumulative PKE calculated at each second from the reference trace This approach was
further modified as described below
1. Similar to fast-pass emission standards, second-by-second PKE limits were not
imposed until t=30 seconds, and
2 To further accommodate the wider variations in cumulative second-by-second
PKE at the beginning of the transient IM240 test, a "PKE Multiplier" factor
was applied that widened the allowed PKE limits. From its maximum value
beginning at t=30 seconds, the PKE multiplier factor was linearly decreased to
a value of unity (i.e., 1.0) at t=239 In other words, at the end of the test, the
PKE limits were set equal to the composite PKE limits.
Figure 23 shows the effect on the effective abort rate resulting from use of various PKE
multipliers between two and five times the composite PKE limits. Based on this analysis,
Sierra generated second-by-second PKE variation limits using a PKE multiplier of 4 0.
Figure 24 illustrates the successive narrowing of the second-by-second limits using this
approach. Appendix C provides a listing of the recommended second-by-second PKE
variation limits
Figure 23
Effect of Initial PKE Multiplier Stringency on
Effective IM240 Abort Rate
(Sample Size = 16,581)
25%
Initial PKE Multiplier
~ lower Umt Aborts Hipper Urn* Aborts
-80-
-------
Figure 24
Illustration of Second-by-Second PKE Variation Limits
Time (sec)
Reference Lower Limit Upper Unnil
Other Issues - The effects of the proposed PKE variation limits by vehicle type and
transmission type (e.g, manual versus automatic) were also investigated. Table 27
presents the effective abort rates for the Arizona data sample as a function of vehicle type
when subjected to the recommended second-by-second PKE limits listed in Appendix C
Table 27
Effective EM240 Abort Rates by Vehicle Type with
Proposed Second-by-Second PKE Variation Limits
Vehicle Type
Sample
Size
Effective Test Abort Rate (%)
Lower
Limit
Upper
Limit
Total
Automobiles
10,770
2 0%
1 9%
3 9%
Light-Duty Truck 1
4,457
1 7%
1 5%
3 3%
Light-Duty Truck 2
1,354
0 8%
19%
2.7%
All Vehicles
16,581
1.8%
1 9%
3 6%
As Table 27 shows, the recommended PKE variation limits will not cause appreciably
different test abort rates for individual light-duty vehicle types The Arizona IM240
-81-
-------
program is limited to light-duty vehicles; therefore, no evaluation of the PKE limits on
heavy-duty vehicles was performed
No evaluation of the PKE variation limits by transmission type was performed in this
analysis. In the previous New York AEL study (cited earlier), a mathematical evaluation
of manual "shift" events versus those observed from an automatic transmission was
conducted. It was found that manual shift events would not cause an appreciable change
in driving behavior statistics such as PKE (Although AEL used the DPWRSUM metric
for this analysis, its second-by-second changes are a function of velocity-squared
differences, like PKE. Therefore, Sierra believes the conclusions are the same for the PKE
statistic.)
Finally, data from the Arizona I/M program were analyzed to determine whether the
ability to meet the proposed speed variation criteria might be affected by vehicle age or
demographic differences between the population in the vicinity of various test facilities
Based on a random sample of 2% quality audit data (involving full I/M240 testing of each
vehicle, regardless of whether it qualified for a fast pass or fast fail), violations of the
proposed criteria (high and low side combined) by model year are shown below
1981 - 10 8%
1982- 12 9%
1983 - 13 8%
1984- 15 5%
1985 - 15 6%
1986 - 16.8%
1987 - 16.3%
1988 - 19.4%
1989 - 14 7%
1990 - 14.6%
1991 - 12 9%
1992 - 13.4%
1993 - 11.9%
1994- 11.4%
1995 - 9.3%
1996-9 8%
The analysis indicates that older vehicles are roughly 50% more likely to fail the proposed
speed variation criteria, as might be expected due to the increased frequency of defects in
older vehicles. Some of these defects are expected to create drivability problems
Potential demographic differences if the failure to meet the proposed speed variation
criteria were evaluated by comparing the results from test facility number 4 to test facility
number 10 Gordon-Darby suggested that these two stations would reflect the ends of the
spectrum in terms of the demographics of the surrounding areas Facility number 4 was
expected to attract an older population of vehicles that might have greater maintenance
problems
-82-
-------
Analysis of the data showed that 14 5% of the 2% audit tests at facility number 4 failed to
meet the proposed criteria compared to 10 3% of the vehicles tested at facility number 10
Much of this difference was related to the age distribution difference rather than
differences between vehicles of the same age For example, 15 2% of 1985 -7 model year
vehicles failed to meet the proposed criteria at facility number 4 compared to 14 6% of the
same age group tested at facility number 10
The analysis described above indicates that differences in the ability of different categories
of vehicles to meet the proposed speed variation criteria is not a major concern
##
-83-
-------
8. EVALUATION OF NOx CORRECTION FACTORS
Early studies on the mechanism of pollutant formation in engine combustion showed that
the formation of oxides of nitrogen had a strong temperature dependence 14 Any changes
in the initial fuel charge that reduce the peak combustion temperature will reduce the
engine-out NOx concentration This concept has been used as the basis for control of
NOx by the addition of exhaust gases as a diluent in exhaust gas recirculation (EGR) or by
the use of water injection In addition to forming the basis for controlling engine-out
emissions, the strong temperature dependence of NOx formation leads to changes in NOx
due to changes in ambient conditions A higher ambient air temperature will lead to a
higher combustion temperature which in turn will produce more NOx. Increased amounts
of water vapor in the inlet air stream will act as a diluent, which will reduce the peak
combustion temperature and reduce NOx Because of their strong temperature
dependence, small changes in ambient conditions (i e, temperature and humidity) can have
a significant impact on vehicle NOx emissions
Existing Correction Factor Equation
Because of the observed effects of initial conditions on NOx formation, several
experimenters evaluated the effects of temperature and humidity on engine NOx emissions
during the late 1960s and early 1970s These studies led to the current NOx correction
factor, which first appeared in the January 15, 1972 Federal Register and has not been
changed since that date. The same correction factor has been incorporated into EPA's
high-tech JM240 test guidance The correction factor, KH, is multiplied by the measured
NOx emissions to yield the reported results For gasoline-powered vehicles, the NOx
correction factor is given by the following equation-
K„ = ! [8-1]
H 1 - 0 0047 (H - 75)
where H is the absolute humidity in units of grains of water per pound of dry air The
absolute humidity is computed from the relative humidity, R^ the ambient (barometric)
pressure, PB, and the saturation vapor pressure of water, Pv, by the following equation"
-84-
-------
43 478 R P
H = —
PB ~
100
[8-2]
In this equation, the relative humidity is in percent and any set of consistent units may be
used for the pressures.* The correction procedure contained in EPA's high-tech guidance
for computing NOx emission results during IM240 testing caps the temperature used to
determine the vapor pressure of water at 86°F However, the value of humidity computed
with this temperature cap is less than the actual absolute humidity The effect of the
temperature cap on the correction factor is shown in Table 28 The correction factors
contained in the column for a temperature of 86°F are the maximum correction factors
allowed under the cap.
Table 28
Effect of Temperature Cap on Humidity Correction Factor
Relative
Humidity
Humidity Correction Factor without Temperature Cap for
Various Temperatures
86°F
90°F
100°F
110°F
0%
0 74
0 74
0 74
0 74
10%
0 79
0.80
0.82
0.85
20%
0 85
0 86
0 92
1.01
30%
0 92
0 95
1.05
1 24
40%
1 00
1.05
1 23
1 62
50%
1.09
1.17
1 49
2 35*
60%
1 21
1 33
1 90
4 37*
70%
1 36
1 54
261*
33 74*
80%
1 55
1.84
421*
-5 77*
90%
1.81
2.28*
11.15*
-2 63*
100%
2 18*
3 01*
-16 58*
-1 69*
'These correction factors are extrapolated beyond the region in which experimental data
were available and regression analysis for those data was conducted
*
The constant 43 47S is the ratio of the molecular weight of water to the molecular weight of air multiplied by
70 The factor of 70 is the unit conversion factor of 7000 grains per pound divided by 100 The division by
100 allows the use of relative humidity as a percent instead of a fraction This constant should actually be
43 537 if accurate molecular weights are used This error of 0 14% does not make any significant difference in
the final correction factor
-85-
-------
For certification tests, where the maximum cell temperature is 86°F, this temperature cap
is not an issue. However, ambient temperatures during I/M testing may be higher than
86°F, particular in the Arizona IM240 program If an actual temperature in excess of 86°F
were used in the equation, the calculated correction factor would be larger. The increase
in the correction factor with the use of an uncapped formula depends on the relative
humidity and temperature as shown in Table 28
At the highest relative humidity and temperature values shown in the table, the absolute
humidity exceeds 288 grains of water per pound of dry air At this humidity and above,
the correction factor equation yields a negative value The equation is obviously not valid
in this region. A review of the original data, discussed below, shows that the
measurements used to derive the correction factor did not include values above 180 grains
of water per pound of dry air. Substituting this value into equation [8-1] yields a
maximum correction factor for the range of experimental data of approximately 2.0.
K„ = - = 1 97 * 2 0 [8-3]
" 1 - 0.0047 (180 - 75)
Humidity correction factors above this value have been marked by an asterisk in Table 29
to show that they have been extrapolated beyond the range of experimental data.
If the temperature used to compute the vapor pressure of water was not capped, it would
be more difficult to pass an I/M test at higher temperatures At present the NOx
correction factor for a relative humidity of 50% and a temperature of 110°F is 1 09, this
value is found in the first numeric column of Table 28 for the capping temperature of
86°F. Without this cap, the humidity correction factor would be 2 35 at this temperature
and relative humidity This is 116% greater than the present value and could cause a
failure for a vehicle that would pass using the current correction factor
However, the capping procedure accounts in part for the lack of a temperature term in the
correction factor At constant absolute humidity, higher temperatures would be expected
to lower the correction factor. Thus, the temperature cap on the vapor pressure provides
a qualitatively correct effect that is not present in the present humidity correction formula
(equation [8-1]). a lowering of the humidity correction factor at higher temperatures.
Review of the Original Data
Sierra discussed the development of the humidity correction factor with EPA staff.15* The
derivation of the final version of the correction factor was presented in an SAE paper16
The information and references provided by Mr Don Paulsell of EPA's Testing Service Division, NVFEL,
was a great benefit to Sierra in the performance of this task His assistance is gratefully acknowledged
-86-
-------
based on the Scott Laboratories study discussed immediately below The details of this
development are discussed in the context of the Scott study
Scott Laboratories f 1970 - Under funding from the National Air Pollution Control
Administration, Scott Laboratories17 examined the effects of temperature and humidity on
"five typical American made vehicles and three foreign-made vehicles " All vehicles were
from the 1969 model year with odometer readings of 12,385 miles or less Although
individual vehicle results were analyzed for all eight vehicles, the study found that the
results from the foreign-manufactured vehicles were "considerably different" from the
domestic vehicles. Accordingly, the composite results were limited to the analysis of the
five domestic vehicles.
Experimental temperatures used ranged from 59°F to 95°F, humidities ranged from 20 to
180 grains of water per pound of dry air Tests were made on the seven-mode cycle
which was used for certification of vehicles at the time of the study Results were
presented on both a concentration basis and a gram-per-mile basis
The data analysis used linear regression to derive an additive correction factor, CFA. With
the additive correction factor, emissions at standard conditions (E0 = E[T0,H0]) are related
to emissions at the measured temperature and humidity (E[T,H]) by the following
equation.
WoyHJ = E{Tfl) + CFa = E(m + a (H - HJ + b(T - TJ [8-4]
The coefficients a and b were found by linear regression The study was done in two
phases. The same vehicles and the same set of temperature and humidity data points were
used in each phase. The report presents the data and analysis from both phases separately
because the data from the second phase were found to be more reliable.
Derivation of the Humidity Correction Factor in Equation [8-1] - The actual regression
equation from the Scott study used to derive the regulatory correction factor is shown
below
E{TJT) = 7.165 + 0 0290(7 - 78) - 0 0337 (H - 75) t8"5!
A comparison of equations [8-4] and [8-5] shows that the following dimensional variables
have been defined:
E(T,H) =
ECT.A) =
H0
T0
actual NOx emissions in grams/mile,
7 165 grams/mile NOx at the reference condition,
75 grains of water per pound of dry air (reference humidity),
78°F (reference temperature);
-87-
-------
a
b
= 0 0337 grams/mile NOx/grain of water per pound of dry air, and
= -0 0290 grams/mile NOx/°F
According to the SAE paper,15 a humidity correction factor for certification test purposes
was derived from equation [8-5] in the following manner. The reference temperature, T0,
was chosen as 78°F because it is near the midpoint of the allowed 68-86°F temperature
range for certification testing. Within this range, the contribution of the temperature term
ranged from -5.4% to +4 4% of the standard value. This effect was considered small and
the temperature term was dropped from the equation Next, a multiplicative correction
factor, Kh, was defined as the ratio of E^HJ / E(T,H). Using equation [8-5] for
E(T,H) and the value ofEfT^Hg) = 7.165 grams NOx/mile yields the humidity correction
factor used for certification testing.
K - - 7 165 _ 1 [8_6]
H E(TJi) 7 165 - 0 0337 (H - 75) 1 - 0 0047 (H - 75)
Recmalysis of Scott Data - Sierra checked this result by reanalyzing the Scott Phase II
NOx data to determine a multiplicative correction factor * The approach used in this
analysis is called the "dummy variable" or "absorption" technique This approach uses
the following regression equation.
ln£ = E o 6> + a(H - HJ + b{T - TJ t8"7!
/ = i
The dummy variable, 6J? is defined to be 1 for measurements on the j* vehicle and is zero
otherwise These dummy variables are added to the actual experimental data that give a
value for the emissions at a particular temperature and humidity. The combination of the
experimental measurements and the dummy variables are the input data for the regression
analysis. That regression analysis of equation [8-7] determines the coefficients a, b and
toaN, where N is the number of vehicles in the study The analysis below shows that the
regression coefficients a, are related to the reference emission values for each vehicle.
For any one vehicle (e g., vehicle k), the value of 6j in equation [8-5] will be zero unless
j = k. If j = k, all values of 8j will be zero except for 6t which will equal one So, for one
vehicle, equation [8-7] may be written as follows:
, [w]
*
The data used in this analysis, and some statistical results, are presented in Appendix D
-88-
-------
When T = T0 and H = Hq, the emissions are the reference emissions E0 = ECT^HJ. But,
when T = T0 and H = H„ the right hand side of equation [8-8] is ak. Thus, setting T = T0
and H = Hj, in equation [8-8] shows that the regression coefficient ak is simply the natural
logarithm of the emissions at the reference conditions for that vehicle
In En = a. 1 [8-9]
0 k \ for vehicle k
Substituting this value for a in equation [8-8] gives the following result.
[ - m*. ~ ~ *
-------
This has the same form as the existing equation for the humidity correction factor if the
temperature effect is ignored
The Phase II NOx data, on a gram-per-mile basis, for the five domestic vehicles were used
in a regression analysis with equation [8-7] as the model The reference humidity and
temperature were taken as 75 grains of water per pound of dry air and 75°F, respectively *
The results were run in three ways (1) with no temperature or humidity correction terms,
(2) with the humidity correction only, and (3) with both the temperature and humidity
correction. When only humidity is considered, the value of a is -0 0043 pounds of dry air
per grain of water; when both humidity and temperature are considered, the respective
values of a and b are -0 00498 pounds of dry air per grain of water and 0.00445/°F The
two values for the humidity coefficient bracket the value of -0 0047 pounds of dry air per
grain of water used in the equation [8-1] for KH. If the temperature term from the
Scott/EPA analysis had been retained, the equation for KH would have a temperature
coefficient of 0 0290/7 165 = 0 00405/°F, which is similar to the one found by Sierra's
analysis.
The statistical significance of adding the temperature and humidity corrections was
evaluated. With no consideration of temperature or humidity, the net effect of the
regression is to set the emissions for each data point equal to the average for the vehicle
that generated that data point. This regression has an R2 value of 0 25 Adding only the
consideration of humidity gave an R2 value of 0.76 Adding the temperature increased the
R2 value from 0 76 to 0.82. An F test on the incremental improvement obtained by each
additional step showed that the addition of both variables adds significant explanatory
power to the model. [The probability (p value) that the humidity has no effect is lxlO"24,
the p value that the addition of temperature to the humidity data has no effect is 5xl0*7 ]
Thus, the Scott Laboratories data show that both the temperature and humidity
corrections are statistically significant. The original rationale for not including the
temperature effect within the temperature range of the Federal Test Procedure (FTP) is
reasonable For applicability to IM240 testing, however, the temperature effect could be
important. Because the temperature effect is smaller than the humidity effect, it is more
difficult to detect the temperature coefficient accurately The regression equation for
temperature and humidity has the following 95% confidence intervals for the coefficients
• Humidity coefficient- 0.000641 or 13% of the coefficient (-0 00498), and
• Temperature coefficient. 0.00231 or 52% of the coefficient (0 00445)
When humidity alone (in addition to the dummy variables) was used in the regression
equation, the 95% confidence interval for the humidity coefficient was 0.000612, or 14%
of the coefficient's value of -0 00430.
Values of the humidity correction factor predicted by equation [8-12] are shown in
Table 29 The Table 29 results are presented for the same range of temperatures shown in
*
This reference temperature was chosen to have the same numerical value as the reference humidity
-90-
-------
Table 28 except that smaller temperature increments are used In addition, the bold lines
in Table 29 divide that table into the following four regions
1 The upper left-hand region has no extrapolations beyond the region of the
experimental data;
2. The upper right-hand region extrapolates temperature, but not humidity, beyond
the range of experimental data,
3. The lower left-hand region extrapolates humidity, but not temperature, beyond
the range of experimental data; and
4. The lower right-hand region extrapolates both temperature and humidity beyond
the region of experimental data.
Table 29
Temperature-Humidity Correction Factors from Equation [8-12]
Relative
Humidity-Temperature Correction Factors for Temperatures Shown
Humidity
T = 86°F
T = 90°F
T = 95°F
T =100°F
T =105°F
T = 110°F
10%
0 72
071
071
071
071
071
20%
0 79
0 79
0 80
0 82
0 84
0 86
30%
0.86
0 88
0 91
0 95
0 99
1 05
40%
0 95
0 98
1 03
1 10
1 18
| 129
50%
1 04
1 09
1 17
1 27
| 1 40
1 58
60%
1 15
1 22
1 33
1 48
1 67
1 94
70%
1 26
1.36
1 51
1 72
2 00
2 40
80%
1.39
152
| 173
201
2 40
2.97
90%
1 53
| 170
1 97
2 35
2 90
3 70
100%
1.69
1 90
2 25
2 76
3 50
4 63
Definition of
Regions in
Table 29
No
extrapo-
lations
T
extrapo-
lated
H
extrapo-
lated
T and H
extrapo-
lated
As shown in Table 29, the effect of temperature at constant relative humidity is smaller in
this case than in the results shown in Table 28. In Table 28, the correction factor at 50%
relative humidity increased from 1.09 to 2.35 as the temperature increased from 86°F to
110°F That effect was simply due to the increase in absolute humidity In Table 29, the
correction factor at 50% relative humidity increased from 1 04 to 1 58 as for the same
change in temperature. The correction factor at 50% relative humidity and 110°F is
suspect since it lies in that region of Table 29 in which both temperature and humidity
-91-
-------
have been extrapolated beyond the range of experimental data. The actual correction
factor used in practice for 50% relative humidity and any temperature at or above 85°F
would be 1 04. This is set by the formula that uses 86°F as the maximum temperature for
computing the saturation vapor pressure of water
The Scott report is very cautious about any conclusion that can be drawn from the test
fleet of five vehicles The final plots of the regression analyses for the combined fleet are
included in an attachment whose cover page has the following heading
Composite Vehicle Correction Factors
(Based on a small sample - for
illustrative purposes only)
The report concludes that
... judgement must be exercised in applying these correction factors because
they care based on a small sample of vehicles. The regression analyses show
quite clearly that additional work with a larger sample of vehicles is
desirable.
Ethyl Research Laboratories CI9701 - The Automobile Manufacturer's Association
(AMA) sponsored a study by Ethyl Research18 that tested ten vehicles Eight vehicles
(seven from the 1969 model year and one from the 1968 model year) were supplied by
AMA. The maximum odometer reading on these vehicles was 7,772 miles Two
additional vehicles from another project were tested simultaneously, but were not included
in the composite results. Tests were made on a seven-mode cycle and reported on a
concentration basis. The temperature for all runs was controlled to 76±5°F and the
humidity ranged from 20 to 120 grains of water per pound of dry air.
The Ethyl results were presented as a quadratic regression equation for the correction
factor The final equation presented by Ethyl can be modified into the form that uses
(H - 75) instead of H as the variable The Ethyl equation, modified into this form, is
shown below.
Kjj* = 1 + 0 0037 {H - 75) + 1.29x/0~5 (H - 75)2 t8"15]
This can be compared to the usual equation for the humidity correction factor by using the
series expansion for 1/(1 -x)
-92-
-------
1
= 1 + x - +
1 - x 2
[8-16]
Using this only the first two terms in this series expansion and ignoring the quadratic term
in equation [8-15], the Ethyl equation for KH** can be written approximately as follows
K,
H
1
1 - 0 0037 (H - 75)
[8-17]
This shows less of an effect of humidity as compared to the certification/IM240 formula in
which the factor multiplying the (H - 75) term is 0 0047 pounds of dry air per grain of
water (instead of 0.0037 pounds of dry air per grain of water shown in the Ethyl results)
However, both equations [8-1] and [8-17] are defined to have the same correction factor
- one - at a humidity of 75 grains of water per pound of dry air Consequently, the
difference in the humidity correction factors between these two equations is small, and the
Scott/EPA SAE paper called the NOx results from the Scott and Ethyl data "remarkably
similar."15 Sierra did not analyze the original data in the Ethyl paper However, average
data were presented in the Ethyl paper for KH at various humidity values These data
were used in a linear regression to obtain a humidity factor of 0 0036 pounds of dry air per
grain of water
The authors of the paper concluded that "[f]actors have been derived for correcting
observed concentrations of nitrogen oxides from the actual intake air moisture content at
the time of test to 75 grains of water/lb dry air at 76 F "
Ford - Ford published a study by Robison19 describing the effects of intake air humidity on
a single six-cylinder engine operated on an engine dynamometer The engine model year
was not identified in the paper, but it must have been earlier than the 1970 publication data
of the paper. This study examined steady-state NO concentrations at a variety of carefully
controlled engine conditions. The goal of this study was not to determine a fleet-average
correction factor, but rather to show how different engine operating variables interacted
with intake air humidity to affect NO concentrations. Robison used an additive correction
factor, CFA, to analyze the NO concentration data.
[NO]0 = [NO] * CFa (H - Ha) [8-18]
In this equation, [NO] represents the NO concentration in ppm and [NO]0 is that
concentration at the reference humidity, Hj, Robison found that the correction factor
increased with NO levels in most cases The exception was for rich mixtures where the
correction factor did not depend on NO concentrations He also found an effect of air/fuel
-93-
-------
ratio on the humidity correction factor He developed two empirical functions of the
air/fuel ratio, X(A/F) and Y(A/F), to express this relationship Robinson's final equation
for the additive humidity correction factor is given by equation [8-19]
CF _ XjAfF) + Y(AfF) [NO] [8_19]
A 1 - Y(A/F) H
Robison's work is based on a single engine and does not develop a correction factor that
can be applied to mass emissions in general. Instead, this work shows a wide variation in
the humidity correction factor depending on the engine operating conditions. The range
of observed values does contain the values of humidity correction found from other
studies
This work shows that humidity effects do depend on engine operating conditions To the
extent that engine operations have shifted from the air/fuel ratios and NOx levels that
existed in late-1960s model year engines, Robison's work would predict that the humidity
effects would change as well
Heaw-Dutv Correction Factor - Ethyl Research Laboratories also determined the
humidity correction factor for gasoline-powered trucks 20 The experimental plan was
similar to the one performed by Ethyl for light-duty passenger cars. Specifically, the tests
were done at a single controlled temperature while the humidity was varied. Seven
engines were tested over the nine-mode cycle used for heavy-duty vehicles at the time of
the study Ethyl presented data on a multiplicative correction factor averaged over all the
vehicles in the test fleet. Sierra used a regression analysis to convert the humidity
coefficient into the form of the present correction factor equation The value was found to
be 0.0042 pounds of dry air per grain of water This is even closer to the value of 0 0047
in equation [8-1] than the value found from the Ethyl data for light-duty vehicles This
study shows that the humidity effect on gasoline-fueled vehicles is relatively independent
of the engine classification.
Discussion of Results
The experimental studies underpinning the current humidity correction factor are generally
in good agreement However, all of those studies were conducted nearly 30 years ago
Although the fundamentals of engine thermochemistry have not changed in that period,
emission control and fuel management systems have changed significantly. As discussed
above, the study by Robison18 found that the humidity correction varied with air/fuel ratio
and NOx level As a result, the decrease in NOx emissions associated with the use of
decreased compression ratios and exhaust gas recirculation (EGR) should reduce the
impact of humidity. However, operation at leaner air/fuel ratios should increase the effect
of humidity The ultimate effect of humidity on engine-out NOx levels and the changes in
-94-
-------
those levels over a catalytic converter cannot be determined without a test program on
modern automobile engines
If extended use of a correction factor is required (i e , in operating IM240 programs)
pending completion of such a test program, the original data used to derive the humidity
correction factor could be used to develop a modified correction factor for use in the
interim. The temperature-humidity correction factor that was found from the Sierra
reanalysis of those data is shown in Equation [8-12], Using constants a = -0 004977
pounds of dry air per grain of water and b = 0.004447/°F, that equation can be rewritten
as follows
_ _ g 0 004977 (H - 75) - 0004447(T - 75) [8-20]
This equation uses temperature in °F and humidity in grains of water per pound of dry air
to produce the dimensionless correction factor The correction factor found from
equation [8-20], over the range of data used in the original study,16 is shown in Table 30
In contrast with previous tables, this table uses absolute humidity instead of relative
humidity. When the humidity values are presented in this fashion, the relatively small
effect of temperature as compared to absolute humidity is clearly seen
In actual I/M tests, there will be a large range of temperatures and humidities that will be
beyond the range in which the original measurements were obtained Some consideration
must be given to the proper extrapolation of equation [8-20] beyond this experimental-
data region. One possible form for limiting the application of this equation is shown in
Table 31. This table shows the correction factors as a function of temperature and relative
humidity. Also shown for comparison are the maximum humidity correction factors, for a
given relative humidity, computed using the formula contained in the high-tech guidance
As shown in the table, the maximum value with the current correction factor is 2 19 (at a
relative humidity of 100% and a temperature of 86°F or greater); this could be used to
limit the correction factor computed from equation [8-20], This would be consistent with
the current approach and require only a small region of temperatures and humidities
(which are shaded in the table) to be set to this maximum value
Although the correction factors in Table 31 provide a better representation of the original
high temperature data, they still have the limitation that the data were obtained on older
cars and may not be applicable to present-day automobiles. The net effect of these factors
for individuals taking an I/M test would only occur at higher temperatures above the
current cap of 86°F The modified factors would be lower (which improves the likelihood
of passing the NOx cutpoints) at relative humidities of 20% or less, and higher (more
likely to fail) at relative humidities of 40% or more Between these two relative
humidities, the change in the factor depends on the temperature
-95-
-------
Table 30
Temperature-Humidity Correction Factors from Equation [8-20]
Absolute
Humidity
Cffr/lb)
Temperature-Humidity Correction Factor for Temperatures Shown
T = 60°F
T = 70°F
T = 80°F
T = 85°F
T = 90°F
T = 95 °F
20
081
0.78
0.74
0.73
071
0 70
40
090
0.86
0 82
0 80
0 79
0 77
60
0 99
0 95
091
0 89
0 87
0 85
80
1 05
1 00
0 98
0 96
0 94
100
I 16
1 11
1 08
I 06
1 04
120
1 22
1 20
1 17
1 14
140
T emperature/humidity
1 35
1 32
1 29
1 26
160
combinations in this
1 46
1 43
1 40
180
region are not possible.
161
00
1,54
Absolute
Humidity
(RT/lb)
For comparison with previous tables, the relative humidities for the
temperatures and absolute humidity entries are shown below
T = 60°F
T = 70°F
T = 80°F
T = 85°F
I = 90°F
T = 95°F
20
26%
19%
13%
11%
10%
8%
40
52%
37%
26%
22%
19%
16%
60
78%
55%
39%
33%
29%
24%
80
73%
52%
44%
38%
32%
100
91%
65%
55%
47%
40%
120
78%
66%
56%
48%
140
This region has
90%
77%
66%
56%
160
relative humidities
87%
75%
64%
ISO
greater than 100%
98%
*1%
71%
Tern
Table 31
perature-Humidity Correction Factors beyond Experimental Range
Relative
Humidity
Temperature-Humidity Correction Factor at Temperatures Shown
Current
Factor for
T>86°F
T=40°F
T=50°F
T=60°F
T=70°F
T=80°F
T=90°F
I00°F
110°F
120°F
0%
0 80
0 77
0.74
0 70
0 67
064
0 62
0 59
0 56
0 74
10%
0.82
0 79
0.76
0 74
0 73
071
071
071
0 73
0 79
20%
0 83
081
0 79
0 78
0.78
0 79
0 82
0 86
0 94
0 85
30%
0 85
0 83
0 82
0 83
0 84
0 88
0 95
1 05
1 23
0 92
40%
0 86
0 86
0 86
0 87
091
0 98
1 10
1 29
1 61
1 00
50%
0 88
0 88
0 89
0 92
0 99
1 09
1 27
1 58
2 12
1 09
60%
0 90
0 90
0 93
0 98
1 06
1 22
1 48
1 94
' 2/82
1 21
70%
091
0 93
0 96
1 03
1 15
1 36
1 72
' 2,40;
,3.78
1 36
80%
0 93
0 95
1 00
1 09
1 25
1 52
2.01
2.97- :
5jo:
1 56
90%
0 95
0 98
1 04
1 15
1 35
1 70
• 13*-
: 643
1 82
100%
0.96
1 00
1 08
1 22
1 46
1 90
¦ %7S '
' 9M
2 19
Note: Shaded cells with correction factors would have the correction factor set to the current
maximum value of 2 19
-96-
-------
The factors in Table 31 provide a more accurate representation of the original data
However, they still have some degree of extrapolation beyond the region of the
experimental data. In addition, because these factors were developed on late 1960s
vehicles, their applicability to current automobiles is uncertain. Until additional test results
or analyses are available, the correction factor we have derived from the original data set
would appear to improve the accuracy of the NOx correction
Use of IM24Q Data
One approach to developing an improved NOx correction factor may be to use IM240
data from Arizona The main problem with such data is that they do not contain repeated
measurements on the same vehicle in the same state of repair with different values of
ambient temperature and humidity However, because of the large size of the database, it
should be possible to find a large number of tests on a vehicle with the same make and
model year If newer vehicles were selected for the study, the vehicle-to-vehicle variation
in emissions should be small This could provide a large enough sample in which the
temperature and humidity effect could be distinguished from the variation among the
different vehicles. The Arizona IM240 data base typically contains approximately 320,000
records per year Limiting the data sample to the 2% random sample of full 240-second
tests means that roughly 16,000 records/year are available for analysis * Selecting the
most popular models (e.g., Ford Taurus) in recent model years may provide a sample of
approximately 100 vehicles with the same make and model year for 10-20 combinations of
make and model year Data from these vehicles could be analyzed using the method
outlined in equations [8-7] to [8-14],
It is not clear if such an approach would work The variation from vehicle to vehicle
could overwhelm the effect of ambient conditions. However, the validity of the results
could be determined by examining their statistical significance. In particular, the
significance test could determine if the regression coefficients for humidity and
temperature effects are statistically different from zero. In addition, the standard error of
the regression coefficients could be used as an indication of the validity of this approach.
Such an approach has the value that it does not require a large and expensive test
program Furthermore, if such a method were found to work, it could be periodically
updated on newer vehicles at veiy little cost
Conclusions
The humidity correction factor presently used for certification and IM240 testing was
derived from tests on late 1960 model year vehicles. Its applicability to modern vehicles is
Until earlier this year, all ambient test temperatures in excess ofS6°F in the Arizona IM240 test program
were recorded as 86°F Because this significantly limits the available data, this analysis approach was not used
in the current Work Assignment The analysis could, however, be undertaken in future years
-97-
-------
uncertain. The initial tests were aimed at corrections for certification tests run between
68°F and 86°F. The maximum temperature and humidity in the studies was 95°F and
180 grains of water per pound of dry air The extrapolation of the original data set
beyond this region is questionable
A revised analysis, giving a multiplicative temperature and humidity correction factor that
could be applied to I/M tests at higher temperatures, was developed Although this
procedure provides a better representation of the original data to higher temperatures, the
applicability of that data set to current automobiles is uncertain There have been
significant changes in technology and a new emissions test cycle since the time that the
original data set for the correction factor was developed It is likely that new emission
control and fiiel management technologies have changed the effect of humidity on average
automobile operation over the emission test cycle For this reason, the most technically
sound approach to developing a revised NOx correction factor would be to conduct an
experimental test program involving current technology automobiles
An alternate approach to developing a temperature-humidity NOx correction factor that
involves the analysis of Arizona IM240 data could be undertaken as either a substitute or a
precursor to a detailed test program This approach would require only a small analytical
effort Although the success of this approach is uncertain, its cost would be extremely low
compared to a large experimental program It is therefore recommended that such an
IM240 analysis be completed before undertaking any experimental program to determine
temperature-humidity correction factors on current automobiles
H (IH
irtf'tt
-98-
-------
9. REFERENCES
1. "High-Tech I/M Test Procedures, Emission Standards, Quality Control
Requirements, and Equipment Specifications. IM240 and Functional Evaporative
System Tests," Revised Technical Guidance (Draft), U S Environmental
Protection Agency, EPA-AA-RPSD-IM-96-1, June 1996
2 "Preconditioning Effects on I/M Test Results Using IM240 and ASM Procedures,"
prepared by Sierra Research for the American Automobile Manufacturers
Association, Report No SR96-09-04, September 30, 1996
3. Richard J Larsen and Morris L Marx, An Introduction to Mathematical Statistics
and Its Applications (second edition), Prentice-Hall, 1986, page 362
4. "I/M Costs, Benefits, and Impacts Analysis (Draft)," U S Environmental
Protection Agency, February 1992
5 Tierney, E J , "Fast-Pass/Fast-Fail Algorithm for the IM240," U S Environmental
Protection Agency, September 13, 1993
6 Ando, A., W Harrington, and V McConnell, "An Investigation of the IM240 Fast
Pass - Fast Fail Algorithm," Resources for the Future, May 21, 1997
7. Whitby, R., C. Shih, W. Webster, and R. Card, "Enhanced Inspection and
Maintenance Program An Investigation of Real Time Data Utility and Dynamic
Quality Assurance Instrumentation - Task 1 A Predictive Modal Regression
Methodology for EM240 Fast-Pass/Fail Decisions," New York State Department
of Environmental Conservation, EPA Cooperative Agreement #X992008-01-0,
November 29, 1996.
8. Olive, Jean Dunn and Virginia A. Clark, Applied Statistics, John Wiley and Sons,
1974.
9 "High-Tech I/M Test Procedures, Emission Standards, Quality Control
Requirements, and Equipment Specifications' IM240 and Functional Evaporative
System Tests" (Revised Draft), U S. Environmental Protection Agency, EPA-AA-
RSPD-IM-96-1, June 1996
10 T C Austin, et al., "An Evaluation of Loaded Mode I/M Testing at Service
Stations," prepared for the State of California Bureau of Automotive Repair,
Sierra Research, Inc , SR-88-12-02, December 1988
-99-
-------
11 0. J Dunn and V A. Clark, Applied Statistics: Analysis of Variance and
Regression. John Wiley and Sons, Inc., 1974
12. W J. Webster amd C. Shih, "A Statistically Derived Metric to Monitor Time-
Speed Variability in Practical Emissions Testing," New York State Department of
Environmental Conservation, presented at the 6th CRC On-Road Vehicle
Emissions Workshop, March 18-20, 1996
13. E. Milkins and H. Watson, Comparison of Urban Driving Patterns, SAE Paper No
830939
14. L. S. Caretto, L. J Muzio, R F. Sawyer, and E S. Starkman, "The Role of
Kinetics in Engine Emission of Nitric Oxide," Combustion Science and
Technology, v 3, pp 53-61, (1971).
15 Telephone conversations with C Don Paulsell, Testing Services Division, System
Support Group, National Vehicle and Fuel Emissions Laboratory, U S
Environmental Protection Agency, September 1997.
16 M. J. Manos, J. W. Bozek, and T. A. Huls, "Effect of Laboratory Ambient
Conditions on Exhaust Emissions," Paper 720124 at Society of Automotive
Engineers Automotive Engineering Congress, Detroit, Michigan, January 10-14,
1972
17 "Effect of Laboratory Ambient Conditions on Exhaust Emissions," Scott Research
Laboratories, Inc, Report on NAPCA Project Number CPA 22-69-156, for
National Air Pollution Control Administration, April 24, 1970
18 W. J. Brown, S A. Gendernalik, R. V Kerley, and F J. Marsee, "Effect of Engine
Intake-Air Moisture on Exhaust Emissions," SAE Paper 700107, 1970
19 J. A. Robison, "Humidity Effects on Engine Nitric Oxide Emissions at Steady-
State Conditions," SAE Paper 700467, 1970.
20 S R Krause, "Effect of Intake-air Humidity, Temperature and Pressure on
Exhaust Emissions," Paper 710835 at Society of Automotive Engineers National
Combined Fuels and Lubricants, Powerplant and Truck Meetings, St. Louis,
Missouri, October 26-29, 1971
-100-
-------
Appendix A
IM240 Retest Criteria
-------
IM240 Retest Criteria for Passenger Cars
Using the replicate IM240 data collected by Gordon-Darby, it was possible, through trial
and error, to identify criteria to determine whether a vehicle failing an initial IM240 is
inadequately preconditioned and should be tested again This analysis was performed for
each pollutant individually, and then for all pollutants combined. The evaluation followed
a step-wise progression in which the aim was to maximize the identification of vehicles
that could benefit from a second test, while minimizing retesting of vehicles likely to fail a
second test. Recommendations for passenger cars are summarized below A similar set of
conditions was also developed for light-duty trucks, which are subject to different IM240
standards than passenger cars.
HC Failures - If ppmHC209.214 is less than 1,500, a retest is recommended if any of the
following occur:
1. Phase 2 HC < 0 8 g/mi; or
2. mass£iC]7j.]gg ^ 0.2 gj or
3 (ppmHC75.80/ppmHC209.214) > 4.0.
For vehicles failing only HC, the following additional constraints are required for a vehicle
to be retested:
1. massHC175_199 < 0.3 g and (ppmHC75^0/ppmHC209.2i4) > 1.5; or
2. massHC175_199 < 0.3 g and Phase 2 HC < 1.0 g/mi.
CO Failures - For CO failures, the above criteria for HC are recommended. In addition,
the following constraints are recommended:
1. do not retest if Phase 2 CO > 20 g/mi and (Phase 1 CO/Phase 2 CO) <2; and
2. if the vehicle fails both HC and CO, retest if massHC175_199 < 0.3 g and
massC0175_199 ^ 5.0 g.
If the vehicle is a CO-only failure, then a vehicle would benefit from a retest if:
1. massC0175.199 < 6 0 g; or
2. (ppmCO75^0/ppmCO209.2i4) > 4 0, or
3. massC0175.199 < 10 g and (Phase 1 CO > 0 75 * Phase 2 CO).
NOx Failures - For vehicles failing HC or CO and NOx, a retest is recommended if the
following condition occurs.
1. massNOx^igg ^1 0 g
A-1
-------
For NOx-onlv failures, retest is recommended if the following criteria are met:
1. massNOx17J_199 < 0.9; or
2 massNOx17J.199 < 1.1 and (ppmNOx4(M5/ppmNOx209.215) > 1 5; or
3. IM240 NQx < 2.2 and (ppmNOx^yppmNOx^.^) > 1.0.
Multiple Pollutants • For multiple pollutant failures, a retest is eliminated under the
following conditions:
1. the vehicle fails for all pollutants; or
2. the vehicle fails HC and CO and (Phase 2 CO > 20 g/mi and massC0175.199 >
6.0 g); or
3 the vehicle fails HC and NOx and (ppmHC209.214 >1,200) or (ppmNOx209.2i4 >
1,200)
A - 2
-------
IM240 Retest Criteria for Light-Duty Trucks
Because they are subject to different numerical IM240 emission standards, a different set
of retest criteria were developed for light-duty trucks. These criteria are similar to those
established for passenger cars, with adjustments to account for standards differences.
HC Failures - For 1981 to 1983 model year vehicles, if ppmHC209_2i4 < 2,000 and any of
the following conditions exist, then a retest is recommended:
1. Phase 2 HC < 3 0 g/mi; or
2. massHC17J.199 < 0.8 g; or
3. (ppmHC75VPpmHC209.214)>4.0.
In addition, if the foil IM240 is less than 3 5 g/mi HC (regardless of the value of
ppmHC209_2i4). then a retest is recommended.
For 1984 and later model year vehicles, if ppmHC2o9.2i4^ ®-
In addition, if 0.42 0 (regardless of
the value of ppmHC^^) then a retest is recommended
A retest is not recommended if Phase 2 HC > 3.2 g/mi
CO Failures - For CO failures, the above criteria outlined for HC were also used. In
addition, the following conditions were also imposed to cut down on the number of
vehicles incorrectly identified as needing a retest:
1. If 1981 to 1983 model year and massC0175.199>36 g then do not retest.
2. If 1984 or later model year and massC017J_199>18 g then do not retest.
3. If Phase 2 CO >40 and Phase 2 CO > Phase 1 CO then do not retest.
NOx Failures - If the vehicle failed NOx and either HC or CO, the above criteria were
used to determine the need for a retest For LDT1 s, if the vehicle failed only NOx. then a
retest is recommended if massNOx175.l99 < 1.4 g. For 1988 and later LDT2s, a retest is
recommended only if massNOx,75.199 < 2.5 g.
A - 3
-------
Appendix B
-------
Composite IM240 HC Regression Coefficients Developed from Modal IM240 Data Analysis
1981 to 1984 Model Year Light-Duty Gasoline Vehicles - 0.8 g/mi Cutpoint
Number
RMS
Error
Reg
Constant
Regression Coefficients
C1
C2
C3
C4
C5
06
C7
C8
C9
C10
C11
C12
C13
C14
C15
C16
C17
C18
C19
C20
C21
C22
C23
1
0 301
0 566
5043
2
0 253
0 378
0187
2802
3
0 247
0 371
•0 383
2 022
2 586
4
0 232
0 337
-0 363
0 828
0 771
3 854
5
0 228
0 325
-0 454
1 046
-0 497
2884
3 156
6
0214
0 286
0 371
0566
0 327
-0 040
1 592
4650
7
0202
0 274
0697
0 468
0 136
-0 077
1 315
2 032
2 632
8
0194
0 260
0 489
0715
-0 044
-0411
1 211
2268
0 653
2410
9
0 189
0 247
0 452
0 747
0049
-0 370
1 228
2 02B
0 757
0664
2 993
10
0 185
0 242
0 163
0 947
-0 410
-0 223
0 801
1 855
0909
0 463
2 017
2 189
11
0 182
0 236
•0613
1 036
-0 076
-0 458
0 652
1 882
0 997
0312
1 900
1 397
4 130
12
0 160
0 179
0127
0 496
0 458
•0 049
0 861
0657
0 532
0 742
0 756
1 363
1 119
2 786
13
0 151
0 160
0 257
0519
0463
•0 050
0 494
0 783
0 574
0 498
0 793
1 013
1 266
2 081
2388
14
0 149
0 156
0 2B5
0 535
0 377
0080
0 324
0 741
0 578
0 480
0 878
0 729
1 310
2 069
1 748
1 228
15
0 146
0 152
0 397
0612
0 326
•0 091
0 754
0 525
0 543
0 420
0 631
0 591
0 768
1 694
1 505
0 562
2 644
IS
0 144
0 150
0 428
0 652
0 276
-0 140
0 404
0 656
0668
0 406
0 658
0 454
-0 393
1 833
1 390
0 498
1 810
1 915
17
0 140
0 142
0 463
0 579
0511
•0 148
0619
0 189
0 756
0 455
0 462
0 632
•0 086
1 551
1 247
0 516
0846
1 432
2 615
16
0 138
0 140
0 505
0566
0 462
•0 015
0 443
0386
0 676
0317
0386
0 622
•0 033
1 600
1 086
0 435
0 399
1 133
1 908
1 738
19
0 134
0 128
0506
0 529
0 820
-0 205
0294
0 539
0 735
0 259
0 147
1 098
-0 693
1 478
0 929
0 550
0693
0 252
1 463
1 S66
1 476
20
0 102
0 058
0 441
0 520
0 567
0 283
0 334
0 275
0 678
0 396
0600
1 082
0 232
0 525
0815
0 244
0831
1 083
0 721
1 244
0B09
0 931
21
0068
0 032
0 507
0 551
0 501
0508
0 307
0 466
0 393
0 487
0 542
1 195
0 446
0 329
0 563
0 546
0500
0 690
0 700
0 763
0 805
0 426
1 089
22
0 041
0013
0 518
0 516
0 564
0 503
0 329
0 622
0 398
0508
0 395
1 055
0 557
0 394
0 483
0 526
0715
0 469
0 615
0618
0 444
0 540
0 430
1 148
23
0 030
0 007
0 517
0 526
0 512
0515
0 448
0 487
0 455
0 519
0 429
0 936
0682
0 389
0 577
0 509
0500
0 654
0 403
0 575
0 456
0517
0 386
0681
0 619
Note Results for only 23 modes are shown here because if the 24th mode is completed, the actual (M240 score woutd be used rather than the predicted score
Phase 2 IM240 HC Regression Coefficients Developed from Modal IM240 Data Analysis
1981 to 1984 Model Year Light-Duty Gasoline Vehicles - 0.5 g/mi Cutpoint
Mode
Number
RMS
Error
Reg
Constant
Regression Coefficients
C11
C12
C13
C14
C15
C16
C17
C16
C19
C20
C21
C22
C23
11
0 179
0346
8 141
12
0 145
0219
2 349
3 104
13
0 136
0 198
1 727
2 209
2 823
14
0 134
0 192
1 330
2210
2034
1 470
15
0 131
0 188
0571
1 862
1 732
0 807
2665
16
0 129
0 184
-0 983
1 887
1 610
0588
1 683
2 208
17
0 125
0 171
-0 505
1 456
1 415
0 728
0712
1 820
3 193
18
0 122
0 168
-0516
1 481
1 215
0533
0 276
1 426
2 227
1 685
19
0 118
0 154
-0 904
1 361
1 041
0 803
0 703
0602
1 573
1 758
1 711
20
0066
0 076
0344
0 701
0 965
0 539
1 035
1 357
0 881
1 676
0988
1 091
21
0 061
0041
0994
0508
0691
0 796
0 735
1 173
0 792
1 146
0 925
0 640
1 372
22
0 039
0 016
1 142
0 562
0 700
0 745
1 010
0 779
0 752
0 874
0560
0 732
0 647
1 544
23
0 029
0 007
1 198
0561
0 777
0 755
0 770
1 005
0 484
0833
0607
0715
0 560
0 975
1 0S8
Note Regression coefficients are presented only (or modes 11 through 23 Mode 11 is the first mode of Pnase 2 and if tfta 21th mode is completed, the actual IM240 score
would be used rather Uian the predicted score
-------
Composite IM240 HC Regression Coefficients Developed from Modal IM240 Data Analysis
1985 to 1989 Model Year Light-Duty Gasoline Vehicles - 0.8 g/mi Cutpoint
RMS
Reg
Regression Coefficients
Number
Error
Constant
C1
C2
C3
C4
C5
C6
C7
C8
C9
C10
C11
C12
C13
C14
C15
C16
C17
C18
C19
C20
C21
C22
C23
1
0 301
0 430
5602
2
0 243
0 259
0159
3044
3
0 242
0 255
-0133
2 106
2 824
4
0 224
0 223
0012
1 015
0555
3864
5
0219
0212
-O 084
1 166
-0 763
2 919
3 395
6
0 207
0 189
0 424
0 756
•0 026
0 591
1 742
4 092
7
8
0 194
0 164
0 181
0 169
0649
0 369
0666
0 880
-0 372
-0 438
0 443
0 037
1 314
1 302
1 053
1 440
3066
0 797
2 830
9
0 174
0 157
0 342
0 927
•0 266
•0 055
1 335
1 391
0656
0 035
4 534
10
0 171
0154
0180
1 101
-0 631
0068
0 949
1 233
0 622
•0 082
3 332
2217
11
0 167
0 150
-0 536
1 168
-0 318
•0 082
0 705
1 264
0 646
-0 127
3 103
1 373
4 334
12
0 144
0 106
0170
0 593
0 221
0 266
0 883
0515
0 448
0 428
1 462
1 314
1 001
2 829
13
0 137
0 095
0 153
0 601
0 146
0 360
0 542
0548
0519
0 277
1 402
1 109
1 248
2 159
2 122
14
0 135
0 091
0169
0 563
0 126
0 499
0 226
0 547
0 530
0 265
1 482
0 760
1 225
2 163
1 358
1 490
15
0 131
0 069
0 272
0609
0 139
0 358
0 572
0388
0 544
0 195
1 174
0 605
0 985
1 914
1 217
0 696
2684
16
0 129
0 087
0 276
0663
0 092
0 322
0166
0 487
0671
0212
1 115
0 457
-0 039
1 631
1 143
0 626
1 714
2 124
17
0 125
0 081
0 346
0 641
0 274
0 241
0416
0209
0 775
0 207
0 887
0 657
0145
1 602
0 944
0 500
0 969
1 564
2 648
IB
0 122
0080
0 348
0 651
0 147
0 387
0 291
0 439
0 622
0 115
0812
0 631
0217
1 651
0 809
0 246
0 546
1 358
1 605
1 637
19
0 116
0 070
0 355
0 591
0 583
0 180
0199
0559
0 697
0 082
0 426
1 018
-0 368
1 511
0 738
0346
0 626
0 523
1 293
1 374
1 818
20
0 065
0 029
0 432
0 465
0 545
0 443
0311
0 226
0 677
0 359
0604
1 146
0 387
0 555
0 731
0187
1 034
0 908
0 755
1 211
0 916
0 953
21
0 055
0018
0 542
0 524
0 365
0 537
0 398
0546
0 373
0556
0 462
1 214
0 466
0368
0 464
0 520
0 679
0 708
0564
0 703
0 808
0 426
1 126
22
0 035
0009
0 575
0503
0 544
0511
0 339
0 596
0 434
0512
0 371
1 085
0 490
0 400
0 49B
0 500
0819
0 464
0 555
0610
0 493
0 533
0 467
1 096
23
0 024
0004
0 574
0 499
0516
0 525
0 455
0 493
0 482
0 528
0 353
0981
0668
0 420
0 558
0 534
0 542
0 577
0 422
0 576
0 522
0504
0 427
0 599
0 842
Note Results for onty 23 modes are shown here because it the 24th mode is completed, the actual IM240 score would be used rather than the predicted score
Phase 2 IM240 HC Regression Coefficients Developed from Modal IM240 Data Analysis
1985 to 1989 Model Year Light-Duty Gasoline Vehicles - 0.5 g/mi Cutpoint
Mode
Number
RMS
Error
Reg
Constant
Regression Coefficients
C11
C12
C13
C14
C15
C16
C17
C18
C19
C20
C21
C22
C23
11
0 177
0 259
9 328
12
0 140
0150
2 179
3 452
13
0 132
0 131
1 779
2 527
2 741
14
0 129
0 125
1 374
2 545
1 628
1 888
15
0125
0 121
0 661
2 070
1 367
1 001
2 991
16
0123
0 119
-0 368
2 053
f 269
0 818
2 160
2 195
17
0116
0 109
0 059
1 736
1 014
0 823
1 032
1 762
3 119
18
0115
0 107
•0 009
1 760
0 833
0 559
0556
1 401
2 128
2030
19
0 109
0094
•0 526
1 622
0 698
0 823
0 762
0636
1 569
1 573
2 175
20
0 077
0041
0 637
0716
0900
0 635
1 081
1 205
1 005
1 455
1 267
1 123
21
0 057
0025
0 945
0 562
0 700
0866
0 834
1 063
0 667
0 971
1 063
0656
1 363
22
0 037
0013
0917
0 590
0 738
0 746
1 105
0 785
0 740
0 839
0 563
0 740
0 631
1 554
23
0 027
0006
1 052
0609
0 774
0 810
0 757
0 906
0 536
0 802
0 675
0711
0 575
0 903
1 137
Note Regression coefficients are presented only (or modes 11 through 23 Mode 11 is the first mode of Phase 2 and if the 24th mode is completed, the actual IM240 score
would be used rather than the predicted score
-------
Composite IM240 HC Regression Coefficients Developed from Modal IM240 Data Analysis
1990 and Later Model Year Light-Duty Gasoline Vehicles - 0.8 g/mi Cutpoint
Mode
Number
RMS
Error
Reg
Constant
Regression Coefficients
CI
C2
C3
C4
C5
C6
C7
C8
C9
C10
C11
C12
C13
C14
C15
C16
C17
C18
C19
C20
C21
C22
C23
1
0366
0162
10 655
2
0 271
0 028
0 252
4 621
3
0 253
0036
0096
2 693
4 788
4
0 230
0 031
0 720
0 666
1 740
4 997
5
0 221
0 026
0 391
1 110
0060
3 325
5 174
6
0 207
0 021
0 662
0 623
1 005
0 397
2 743
5132
7
0 192
0 025
1 279
0 372
0 699
0 393
2 028
1 127
3 732
6
0 175
0030
\ 076
0 515
0 477
0150
1 178
1 750
0 608
3 725
9
0 162
0 032
1 111
0 490
0510
0142
1 351
1 370
0 587
0 494
5 180
10
0 155
0 034
0 742
0 835
•0133
0 470
0 754
0 829
0 980
0 287
3033
3511
11
0 149
0 037
-0 292
0916
0 270
0 141
0 342
0 923
1 049
0 326
2661
2 172
5 629
12
0 123
0 028
0 433
0 321
0 650
0 528
0818
•0 003
0 245
1 037
0 672
2 127
1 234
3 583
13
0 115
0 025
0 578
0 296
0 562
0 394
0 374
0 234
0 492
0715
0 629
1 553
1 271
2 743
2 839
14
0 113
0 025
0618
0 262
0 555
0503
-0 034
0 227
0 569
0 714
0 776
1 121
1 069
2 758
1 607
2 069
15
0 107
0 027
0 747
0 284
0 574
0 244
0 529
0 033
0 676
0 571
0 487
1 011
0912
2 369
1 258
0 591
3362
16
0 102
0 027
0 719
0 462
0 356
0 136
0065
0 303
0 778
0668
0 345
0 659
-0008
2 189
1 033
0 401
2 153
2 766
17
0 098
0 026
0 794
0 426
0518
0 062
0 242
0 168
0 7B6
0 737
0 079
0 882
0127
1 928
0 726
0 488
1 245
1 994
2 960
16
0 095
0 026
0 769
0 455
0 367
0197
0 174
0 420
0 710
0 561
0 016
0 870
0 128
2 015
0 442
0 256
0 759
1 521
1 733
2083
19
0 090
0 023
0616
0 385
0 654
0080
0083
0 642
0717
0 557
-0 358
1 269
-0 468
1 916
0 340
0 262
0619
0 688
1 357
1 477
1 846
20
0068
0009
0 575
0360
0 539
0 382
0 257
0 458
0 639
0 556
0 034
1 205
0514
0804
0 472
0 255
0 693
1 266
0 881
1 203
0 837
1 073
21
0 041
0 007
0 544
0 483
0 489
0 528
0S39
0 520
0 388
0 586
0 264
1 157
0 523
0 367
0 461
0 575
0 548
0 853
0 677
0674
0583
0 370
1 309
22
0 027
0003
0 559
0500
0535
0 538
0444
0 602
0 406
0 514
0 284
0 992
0 497
0 407
0 518
0 526
0656
0 637
0 661
0 573
0 470
0 497
0 52S
1 112
23
0019
0 002
0 5S3
0 496
0515
0 537
0 525
0518
0 438
0 535
0 349
0 870
0 650
0 429
0 549
0 567
0 521
0 624
0 554
0536
0 443
0 510
0 387
0 654
0 853
Note Results for only 23 modes are shown here because rf the 24th mode is completed, the actual IM240 score would be used rather than the predicted score
Phase 2 IM240 HC Regression Coefficients Developed from Modal IM240 Data Analysis
1990 and Later Model Year Light-Duty Gasoline Vehicles - 0.5 g/mi Cutpoint
Mode
Number
RMS
Error
Reg
Constant
Reqresston Coefficients
C11
C12
C13
C14
C15
C16
C17
C18
C19
C20
C21
C22
C23
11
0153
0 098
13 195
12
0 114
0 048
3 112
4 074
13
0 102
0 041
1 873
2680
3 765
14
0099
0039
1 406
2668
2 167
2 454
15
0 095
0 040
1 107
2 141
1 738
1 122
3 342
16
0 091
0 040
-0 030
2 065
1 365
0 858
2 119
3 157
17
0088
0038
0 165
1 791
1 043
0 931
1 028
2 401
3 357
18
0085
0038
0 093
1 843
0 719
0663
0610
1 854
2 169
2 206
19
0 081
0 035
-0 272
1 779
0514
0 829
0 678
1 356
1 669
1 515
1 952
20
0 057
0016
0 581
0 846
0 727
0616
0 627
1 640
1 313
1 345
0 656
1 195
21
0040
0010
0817
0 585
0 723
0 936
0 568
1 434
1 014
0 886
0 731
0 623
1 512
22
0 028
0005
0 823
0606
0 751
0 837
0 740
1 015
0 643
0 655
0 581
0717
0 749
1 426
23
0 020
0003
0 909
0612
0 790
0 837
0609
0 957
0714
0 775
0 632
0719
0 591
0868
1 156
Note Regression coefficients are presented only for modes 11 through 23 Mode 11 is the first mode of Phase 2 and if the 24th mode is completed the actual IM240 score
would be used rather than the predicted score
-------
Appendix C
-------
IM240 REFERENCE DATA
CUM PKE
TIME SPEED (miles/hr2)
0
00
00
1
00
00
2
00
00
3
00
00
4
00
00
5
30
10,800 0
6
59
14,080 4
7
86
15,2146
8
11 5
16,4172
9
14 3
17,001 5
10
16 9
17.079 7
11
17 3
13,902 5
12
18 1
12,336 8
13
20 7
13,263 7
14
21 7
12,284 1
15
22 4
11,261 4
16
22 5
9,964 5
17
22 1
8,890 2
18
21 5
8,046 4
19
20 9
7,366 6
20
20 4
6,805 5
21
19 8
6.336 9
22
170
5,983 3
23
14 9
5.704 2
24
14 9
5,450 1
25
15 2
5,306 1
26
15 5
5,171 6
27
160
5,103 3
28
17 1
5,213 3
29
19 1
5,599 3
30
21 1
5.990 0
31
22 7
6,242 3
32
22 9
6,014 B
33
22 7
5,745 3
34
22 6
5,500 0
35
21 3
5,287 2
36
190
5,110 8
37
17 1
4,961 9
38
158
4,831 8
39
158
4,708 3
40
17 7
4,937 5
41
198
5,220 8
42
21 6
5,450 3
43
23 2
5.638 2
44
24 2
5.685 3
45
24 6
5.592 5
46
24 9
5,481 7
47
25 0
5,332 7
48
25 7
5,321 5
49
26 1
5,245 9
50
26 7
5,216 3
51
27 5
5,230 2
52
28 6
5,307 9
53
29 3
5,296 0
54
29 8
5,246 1
55
30 1
5,155 0
56
30 4
5,068 3
57
30 7
4,985 6
58
30 7
4,848 3
59
30 5
4,719 2
60
30 4
4,597 2
61
30 3
4,481 7
PKE VARIATION CUTPOINTS (miles/hr2)
"BASE" MULT VARYING CUMULATIVE PKE
DELTA FACTOR DELTA LOW HIGH
342 3
4 000
1,369 3
4,621
7.359
356 7
3 966
1,421 8
4,820
7,564
343 7
3 971
1,365 1
4,650
7 380
328 3
3 957
1,299 3
4,446
7,045
314 3
3 943
1,239 3
4,261
6,739
302 2
3 929
1,1870
4,100
6,474
292 1
3 914
1,143 3
3,968
6,254
283 6
3 900
1,105 9
3,856
6,068
276 1
3 886
1,072 9
3,759
5,905
269 1
3 871
1,041 7
3,667
5,750
282 2
3 857
1,088 4
3,849
6,026
298 4
3 843
1,146 5
4,074
6,367
311 5
3 829
1,192 5
4,258
6,643
322 2
3 814
1,229 0
4,409
6,867
324 9
3 800
1.234 6
4,451
6,920
319 6
3 786
1.209 9
4,383
6,802
313 3
3 771
1,181 5
4,300
6,663
304 8
3 757
1,1450
4,188
6,478
304 1
3 743
1,1382
4,183
6,460
299 8
3 729
1,1178
4,128
6,364
298 1
3 714
1,107 2
4,109
6,323
298 9
3 700
1,105 9
4,124
6,336
303 3
3 686
1,1180
4,190
6,426
302 8
3 671
1,111 6
4,186
6,410
299 8
3 657
1,096 4
4,150
6,343
294 6
3 643
1,073 2
4.082
6,228
289 6
3 629
1,051 0
4,017
6,119
284 9
3 614
1,029 8
3,956
6,015
277 1
3 600
997 5
3,651
5,846
269 7
3 586
967 0
3,752
5,686
262 7
3 571
938 3
3,659
5,535
256 1
3 557
911 1
3,571
5,393
C-l
-------
IM240 REFERENCE DATA
CUM PKE
TIME SPEED (milesrtir2)
62
30 4
4.389 2
63
30 8
4,352 1
64
30 4
4,250 1
65
29 9
4,154 4
66
29 5
4,064 1
67
29 8
4,022 9
68
30 3
4,013 3
69
30 7
3,988 8
70
30 9
3,935 5
71
31 0
3.869 4
72
30 9
3,791 8
73
30 4
3.718 5
74
29 8
3,649 2
75
29 9
3,595 5
76
30 2
3,569 2
77
30 7
3,569 2
78
31 2
3.569 3
79
31 8
3.582 2
80
32 2
3,569 2
81
32 4
3.531 2
82
32 2
3,469 8
83
31 7
3,411 4
84
28 6
3,360 3
85
251
3,316 8
86
21 6
3,280 2
87
18 1
3,250 2
88
14 6
3,226 3
89
11 1
3,208 4
90
76
3,196 3
91
4 1
3,189 8
92
06
3.188 9
93
00
3.188 9
94
00
3,188 9
95
00
3,188 9
96
00
3,188 9
97
00
3,188 9
98
33
3,203 1
99
66
3,250 7
100
99
3,331 3
101
132
3,443 8
102
16 5
3,587 2
103
198
3,760 1
104
22 2
3,892 7
105
24 3
4,013 3
106
25 8
4,090 8
107
26 4
4,093 0
108
25 7
4,045 3
109
25 1
3,999 9
110
24 7
3,956 1
111
25 2
3,951 8
112
25 4
3,924 1
113
27 2
4,024 3
114
26 5
3,979 2
115
24 0
3,939 2
116
22 7
3,902 0
117
194
3,870 9
118
17 7
3,842 9
119
17 2
3,816 0
120
18 1
3,834 3
121
18 6
3,832 2
122
20 0
3,879 0
123
20 7
3,887 7
PKE VARIATION CUTPOINTS (miles/hr2)
'BASE" MULT VARYING CUMULATIVE PKE
:LTA
FACTOR
DELTA
LOW
HIGH
250 8
3 543
888 7
3,501
5,278
248 7
3 529
877 6
3,474
5,230
242 9
3 514
853 6
3,397
5,104
237 4
3 500
831 0
3,323
4,985
232 3
3 486
809 6
3,255
4,874
229 9
3 471
798 1
3,225
4,821
229 3
3 457
792 9
3,220
4,806
228 0
3 443
784 8
3,204
4,774
224 9
3 429
771 1
3,164
4,707
221 1
3 414
755 0
3,114
4,624
216 7
3 400
736 8
3,055
4,529
212 5
3 386
719 5
2.999
4,438
208 5
3 371
703 1
2,946
4,352
205 5
3 357
689 8
2,906
4,285
204 0
3 343
681 9
2,887
4,251
204 0
3 329
678 9
2.890
4,248
204 0
3 314
676 0
2,893
4,245
204 7
3 300
675 6
2,907
4.258
204 0
3 286
670 2
2,899
4.239
201 8
3 271
660 2
2,871
4,191
198 3
3 257
645 9
2,824
4,116
195 0
3 243
632 2
2,779
4,044
192 0
3 229
620 0
2,740
3,980
189 5
3 214
609 3
2,708
3,926
187 5
3 200
599 9
2,680
3,880
185 7
3 186
591 7
2,658
3,842
1844
3 171
584 7
2,642
3.B11
183 4
3 157
578 9
2,630
3,787
182 7
3 143
574 1
2,622
3,770
182 3
3 129
570 3
2.619
3,760
182 2
3 114
567 5
2,621
3,756
182 2
3 100
564 9
2,624
3,754
182 2
3 086
562 3
2,627
3,751
182 2
3 071
559 7
2,629
3,749
182 2
3 057
557 1
2,632
3,746
182 2
3 043
554 5
2,634
3,743
183 0
3 029
554 4
2,649
3,757
1858
3 014
560 0
2.691
3,811
190 4
3 000
571 1
2,760
3,902
196 8
2 986
587 6
2,856
4,031
205 0
2 971
609 1
2,978
4,196
214 9
2 957
635 4
3,125
4,396
222 5
2 943
654 7
3,238
4,547
229 4
2 929
671 7
3,342
4,685
233 8
2 914
681 3
3,409
4,772
233 9
2 900
678 3
3,415
4,771
231 2
2 886
667 1
3,378
4,712
228 6
2 871
656 4
3,344
4,656
226 1
2 857
646 0
3,310
4,602
225 8
2 843
642 0
3,310
4,594
224 3
2 829
634 3
3,290
4,558
230 0
2 814
647 2
3.377
4,672
227 4
2 800
636 7
3.342
4,616
225 1
2 786
627 1
3.312
4,566
223 0
2 771
618 0
3.284
4,520
221 2
2 757
609 9
3.261
4 481
219 6
2 743
602 4
3,241
4,445
218 1
2 729
595 0
3,221
4,411
219 1
2 714
594 8
3,240
4,429
219 0
2 700
591 3
3,241
4,423
221 7
2 686
595 4
3,284
4,474
222 2
2 671
593 5
3,294
4,481
C-2
-------
IM240 REFERENCE DATA
CUM PKE
TIME SPEED (miles/hr2)
124
21 7
3.914 4
125
22 4
3,923 5
126
22 5
3,895 8
127
22 1
3,863 1
128
21 5
3,831 7
129
20 9
3,801 8
130
20 4
3,772 9
131
19 8
3,745 4
132
170
3,722 1
133
17 1
3,703 4
134
158
3,682 2
135
158
3,661 2
136
17 7
3,720 0
137
198
3,794 6
138
21 6
3,860 2
139
22 2
3,863 3
140
24 5
3,964 6
141
24 7
3,943 1
142
24 8
3,915 9
143
24 7
3,883 2
144
24 6
3,851 1
145
24 6
3,819 6
146
251
3,817 5
147
25 6
3,815 4
148
25 7
3,789 6
149
25 4
3,758 6
150
24 9
3,728 8
151
25 0
3,704 9
152
25 4
3,698 2
153
26 0
3,702 8
154
26 0
3,673 1
155
25 7
3,644 1
156
26 1
3,637 9
157
26 7
3,643 0
158
27 3
3,648 1
159
30 5
3,812 6
160
33 5
3,978 0
161
36 2
4,133 0
162
37 3
4,172 3
163
39 3
4,282 5
164
40 5
4,330 6
165
42 1
4,412 1
166
43 5
4,477 8
167
45 1
4,561 4
168
46 0
4,584 2
169
46 8
4,598 2
170
47 5
4,603 2
171
47 5
4,546 8
172
47 3
4,492 0
173
47 2
4,438 6
174
47 2
4,386 5
175
47 4
4,352 1
176
47 9
4,343 2
177
48 5
4,342 6
178
49 1
4,342 0
179
49 5
4.324 9
180
50 0
4,316 3
181
50 6
4.316 0
182
51 0
4,299 3
183
51 5
4,291 0
184
52 2
4,299 3
185
53 2
4,332 3
PKE VARIATION CUTPOINTS (milesyhr2)
BASE" MULT VARYING CUMULATIVE PKE
ILTA
FACTOR
DELTA
LOW
HIGH
223 7
2 657
594 4
3,320
4,509
224 2
2 643
592 6
3,331
4.516
222 6
2 629
585 2
3,311
4,481
220 8
2 614
577 1
3,286
4,440
219 0
2 600
569 3
3,262
4,401
217 3
2 586
561 B
3,240
4,364
215 6
2 571
554 4
3,219
4,327
214 0
2 557
547 3
3,198
4,293
212 7
2 543
540 9
3,181
4,263
211 6
2 529
535 1
3,168
4,239
210 4
2 514
529 1
3,153
4,211
209 2
2 500
523 1
3,138
4,184
212 6
2 486
528 4
3,192
4,248
216 9
2 471
535 9
3,259
4,330
220 6
2 457
542 1
3,318
4,402
220 8
2 443
539 3
3,324
4,403
226 6
2 429
550 2
3,414
4,515
225 3
2 414
544 0
3,399
4,487
223 8
2 400
537 1
3,379
4,453
221 9
2 386
529 4
3,354
4,413
220 1
2 371
521 9
3,329
4,373
218 3
2 357
514 5
3,305
4,334
218 2
2 343
511 1
3,306
4,329
218 0
2 329
507 7
3,308
4,323
216 6
2 314
501 2
3,288
4,291
214 8
2 300
494 0
3,265
4,253
213 1
2 286
487 1
3,242
4,216
211 7
2 271
480 9
3,224
4,186
211 3
2 257
477 0
3,221
4,175
211 6
2 243
474 6
3.228
4,177
209 9
2 229
467 8
3.205
4,141
208 3
2 214
461 1
3,183
4,105
207 9
2 200
457 4
3,181
4,095
208 2
2 186
455 0
3,188
4,098
208 5
2 171
452 7
3,195
4,101
217 9
2 157
470 0
3,343
4,283
227 3
2 143
487 1
3,491
4,465
236 2
2 129
502 8
3,630
4,636
238 4
2 114
504 1
3,668
4,676
244 7
2 100
5139
3,769
4,796
247 5
2 086
516 2
3,814
4,847
252 1
2 071
522 3
3,890
4,934
255 9
2 057
526 4
3,951
5,004
260 7
2 043
532 5
4,029
5,094
262 0
2 029
531 4
4,053
5,116
262 8
2 014
529 3
4,069
5,127
263 1
2 000
526 1
4,077
5,129
259 8
1 986
516 0
4,031
5,063
256 7
1 971
5061
3,986
4,998
253 7
1 957
496 4
3,942
4,935
250 7
1 943
487 0
3,899
4,874
248 7
1 929
479 7
3,872
4,832
248 2
1 914
475 1
3,868
4,818
248 2
1 900
471 5
3,871
4,814
248 1
1 886
467 9
3,874
4,810
247 2
1 871
462 5
3,862
4,787
246 7
1 857
458 1
3,858
4,774
246 7
1 843
454 5
3,861
4,771
245 7
1 829
449 3
3,850
4,749
245 2
1 814
444 9
3,846
4,736
245 7
1 800
442 3
3,857
4,742
247 6
1 786
442 1
3,890
4,774
C-3
-------
IM240 REFERENCE DATA
CUM PKE
TIME SPEED
-------
Appendix D
-------
Original Scott Laboratories Data Used in Regression Analysis
Vehicle
Temperature
(°F)
Humidity
Grains
H20
per pound
of dry air
Measured
NOx
(grams/
mile)
NOxfrom
Regression
with T & H
(grams/mile)
NOx from
Regression
with H only
(grams/mile)
NOx from
Regression
without T or H
(grams/mile)
I
59
20
7 62
6 78
7 19
5 65
1
59
40
6 81
614
6 60
5 65
1
59
60
6 54
5 56
6.05
5 65
1
71
20
6 92
7 15
7 19
5 65
1
71
40
6 90
6 47
6 60
5 65
1
71
60
6 48
5 86
6 05
5 65
1
71
80
5 39
5 30
5 55
5 65
]
83
20
7 42
7 54
7 19
5 65
1
83
60
5 26
6 18
605
5 65
1
83
80
6 06
5 60
5 55
5 65
I
83
100
4 24
5 07
5 09
5 65
1
83
120
3 39
4 59
4 67
5 65
I
83
140
3 35
4 15
4 28
5 65
1
95
40
751
7 20
660
5 65
I
95
60
7 04
6 52
605
5 65
1
95
80
4.69
5 90
5 55
5 65
1
95
100
5 34
5 34
5 09
5 65
1
95
140
4 62
4 38
4 28
5 65
1
95
180
4 94
3 59
3 60
5 65
2
59
20
821
6 72
7 14
5 66
2
59
40
5 92
6 09
6 55
5 66
2
59
60
5 60
551
601
5 66
2
71
20
6.69
7 09
7 14
5 66
2
71
40
6 60
6 42
6 55
5 66
2
71
60
6 46
5 81
601
5 66
2
71
80
5 15
5 26
551
5 66
2
83
20
7 12
7 48
7 14
5 66
83
40
6 37
6 77
6 55
5 66
83
60
6 44
6 13
601
5 66
n
83
80
4 26
5 55
551
5 66
2
83
100
4 94
5 02
5 05
5 66
2
83
120
4 61
4 55
4 64
5 66
2
83
140
3 94
4 12
4 25
5 66
2
95
40
7 26
7 14
6 55
5 66
2
95
60
6 80
6 47
601
5 66
-------
Original Scott Laboratories Data Used in Regression Analysis
Vehicle
Temperature
(°F)
Humidity
Grams
H20
per pound
of dry air
Measured
NOx
(grams/
mile)
NOxfrom
Regression
with T & H
(grams/mile)
NOx from
Regression
wilh H only
(grams/mile)
NOx from
Regression
without T orH
(grams/mite)
2
95
80
5 42
5 85
5 51
5 66
2
95
100
5 59
5 30
5 05
5 66
2
95
J 40
4 38
4 34
4 25
566
2
95
180
384
3 56
3 58
566
3
59
20
831
8 99
9 55
7.56
3
59
40
7 74
8.14
8 76
7 56
3
59
60
7 89
7 36
8 03
7 56
3
71
20
8 33
948
9 55
7.56
3
71
40
8 89
8 58
8.76
7 56
3
71
60
5 82
777
8.03
7.56
3
71
80
5 58
7 03
7 37
7 56
3
83
20
10 60
10.00
9 55
7 56
3
83
40
845
905
8 76
7 56
3
83
60
934
8 19
8 03
7 56
3
83
80
908
7 42
7 37
7 56
3
83
100
7 36
6 72
6 76
7 56
3
83
120
6.6 \
608
6 20
7 56
3
83
140
586
5 50
5 68
7 56
3
95
40
9 84
9 55
8 76
7 56
3
95
60
9 28
864
8 03
7 56
3
95
80
7 17
7 82
7 37
7.56
3
95
100
7 12
7 08
6 76
7 56
3
95
140
5 83
" 580
5 68
7 56
3
95
180
5 20
4 76
4 78
7 56
4
59
20
7 12
7 22
7 68
601
4
59
40
5 98
6 54
7 04
6 01
4
59
60
608
5 92
6 46
601
4
71
40
806
6 90
7 04
601
4
71
60
7 65
624
6 46
6.01
4
71
80
4 96
5 65
5 92
6 01
4
83
20
804
804
7 68
6 01
4
83
40
6 02
7 28
7 04
6 01
4
83
60
7 70
6 59
6 46
6 01
4
83
80
6 46
5 96
5 92
601
4
83
100
5 56
5 40
5 43
601
4
83
120
4 86
4 89
4 98
601
4
83
140
4 57
4 42
4 57
601
4
95
40
950
7 67
7 04
601
4
95
60
7 77
6 95
646
601
-------
Original Scott Laboratories Data Used in Regression Analysis
Vehicle
Temperature
(°F)
Humidity
Grains
H20
per pound
of dry air
Measured
NOx
(grams/
mile)
NOxfrom
Regression
with T & H
(grams/mile)
NOxfrom
Regression
with H only
(grams/mile)
NOxfrom
Regression
without T or H
(grams/mile)
4
95
80
6 67
6 29
5 92
601
4
95
100
6 24
5 69
5 43
601
4
95
140
3 93
4 67
4 57
601
4
95
180
3 85
3 82
3 84
601
5
59
20
801
884
9 39
7 43
59
40
7 74
8 00
861
7 43
5
59
60
7 89
7 24
790
7 43
s
71
20
9 97
9 32
9 39
7 43
s
71
40
9 22
8 44
861
7 43
5
71
60
6 44
7 64
790
7 43
5
71
80
7 39
691
7 24
7 43
5
83
20
971
9 83
939
7 43
5
83
40
9 35
8 90
861
7 43
5
83
60
7 74
8 06
7 90
7 43
S
83
80
7 60
7 29
7 24
7 43
5
83
100
6 89
6.60
6 64
7 43
5
83
120
621
5 98
6 09
7 43
5
83
140
5 62
541
5 59
7 43
5
95
40
10 40
9 39
861
7 43
5
95
60
8 89
8 50
7 90
7 43
S
95
80
7 34
7 69
7 24
7 43
5
95
100
6 70
6 96
664
7 43
5
95
140
4 97
571
5 59
7 43
5
95
180
4 28
4 68
4 70
7 43
Mean NOx
6 67
NOx Standard Deviation
I 68
NOx total sum of squares
275 40
Number of data points
98
98
98
Number of regression parameters
7
6
5
Degrees of freedom
91
92
93
Standard Error of NOx estimate for original data.
0 74
0 84
I 49
Residual sum of squares for original NOx data
49 51
65 38
205 88
RJ for original NOx data
0 82
0 76
0 25
Difference in sum of squares from column to left
15 87
140 50
Additional parameters for reduced sum of squares
I
1
Mean square for added parameter
15 87
140 50
F ratio for added parameter
29 17
197 71
Probability that added parameter is not significant
5 25x10"7
1 21x10"24
------- |