APTD-1356
AN EVALUATION
OF THE STRATIFIED
CHARGE ENGINE (SCE)
CONCEPT
U.S. ENVIRONMENTAL PROTECTION AGENCY
Office of Air and Water Programs
e of Mobile Source Air Pollution Control
itomotive Power Systems Development Division
Ann Arbor, Michigan 48105
-------
AN EV ALU A TION
OF THE STRATIFIED
CHARGE ENGINE (SCE)
CONCEPT
Prepared by
L. Bogdan, H. T. McAdams, D. J. Schuring
Cornell Aeronautical Laboratory, Inc.
Buffalo, New York 14221
Contract No. 68-04-0040
EPA Proj ect Officer: J. Dillard ~1urrell
Prepared for
u. S. ENVIRONMENTAL P~OTECTION AGENCY
Office of Air and Water Programs
Office of Mobile Source Air Pollution Control
Advanced Automotive Power Systems Development Division
Ann Arbor, Michigan 48105
January 1972
APTD-1356
-------
The APTD (Air Pollution Technical Data) series of reports is issued by
the Office of Air Quality Planning and Standards, Office of Air and
Water Programs, Environmental Protection Agency, to report technical
data of mterest to a limited number of readers. Copies of APTD reports
are available free of charge to Federal employees, current contractors
and grantees, and non-profit organizations - as supplies permit - from
the Air Pollution Technical Information Center, Environmental Protection
Agency, Research Triangle Park, North Carolina 27711 or may be obtained
for a nominal cost, from the National Technical Information Service,
5285 Port Royal Road, Springfield, Virginia 22151.
This report was furnished to the U. S. Environmental Protection Agency
by Cornell Aeronautical Laboratory, Inc. in fulfillment of Contract
No. 68-04-0040 and has been reviewed and approved for publication by
the Environmental Protection Agency. Approval does not signify that
the contents necessarily reflect the views and policies of the agency.
The material presented in this report may be based on an extrapolation
of the "State-of-the-art." Each assumption must be carefully analyzed
by the reader to assure that it is acceptable for his purpose. Results
and conclusions should be viewed correspondingly. Mention of trade
names or commercial products does not constitute endorsement or
recommendation for use.
Publication No. APTD-1356
i i
-------
FOREWORD
This report was prepared by Cornell Aeronautical Laboratory,
Inc. (CAL), Buffalo, New York, under Environmental Protection Agency
(EPA) Contract No. 68-04-0040, Modification No.4.
The work was
administered under the direction of the Office of Air Programs, Division
of Advanced Automotive Power Systems Development, Mr. J. D. Murrell,
Project Officer.
This is the Phas e II
interim technical report that describes
and summarizes the results of studies conducted during the period from
October 1971 through December 1971 on the engineering and emissions
performance aspects of the stratified-charge engine concept.
The effort was performed jointly by the Vehicle Res earch
Department and the Systems Research Department of CAL.
111
-------
ABSTRACT
A descriptive aCCO\.U1t is presented of preliminary studies aimed
at establishing a rational basis for the evaluation of the stratified charge
engine as a low-emission power plant for light duty vehicle applications.
A description of the salient engineering features of the stratified charge
engine is included together with statistical analys es of the limited exhaust
emissions data that are available on vehicles equipped with hand-built
experimental engines.
For purposes of augmenting this limited data base
and for purposes of comparison and analogy, statistical analyses of
pertinent emissions data obtained for conventional-type automobile engines
are also pres ented.
To provide essential but missing data, the rationale for and the
details of an experiment design for presently available engine/vehicle \.U1its
is pres ented.
New hardware requirements are also identified.
lV
-------
I
II
III
IV
TABLE OF CONTENTS
FOREWORD
ABSTRACT
INTRODUC TION
SUMMARY
STUDY RESULTS
A.
B.
General
Survey of Available Literature on the SCE
Statistical Analysis of Selected Test Data
C.
l.
2.
Stratified Charge Engines
Conventionally-Powered Vehicles
a.
Test Error
b.
Parametric Effects and Test Procedures
c.
Product Variability
D.
E.
Statistical Modeling Cons ide rations
Generation of New Data
l.
2.
Experiment Design for Existing Hardware
Definition of New Hardware Requirements
F.
Hardware - Information Considerations
APPENDICES
A.
Summary of a Literature Survey of the Stratified
Charge Engine Concept
B.
Analysis of Emissions Test Data on Low-Emission
FCP Engine Installed in M-151 Vehicle
v
Page No.
iii
iv
1
5
7
7
7
13
13
16
16
18
19
21
24
24
26
28
A-1
B-1
-------
TABLE OF CONTENTS (cont'd)
C.
D.
Regression Analysis of Emissions Data
Analysis of Emission Test Data of Various Groups
of Automotive Vehicles
C-l
D-l
E.
Emissions Analysis of 54 Fleet-Type Passenger
Vehicles
£-1
F.
Exhaust Emis sions for Chevrolet and Ford
Automobiles
F-l
G.
The Log Normal Distribution as a Statistical Model
for Exhaust Emissions
G-)
H.
Experimental Design Considerations for Emissions
Testing of Stratified-Charge Engines
New Hardware Requirements for Assessing Exhaust
Emissions Performance of the Stratified Charge
Engine - A Statistical Consideration
H-l
1.
I-I
vi
-------
I
INTRODUCTION
This report summarizes the results of preliminary studies
conducted for the Division of Advanced Automotive Power Systems Development,
Environmental Protection Agency, and directed to the ultimate objective of
establishing the technological feasibility of the stratified charge engine (SCE)
concept meeting the 1976 Federal exhaust emission standards in light duty
vehicle applications.
In evaluating technological feasibility, such
important factors of the engine/vehicle composite as performance, drive-
ability, manufacturability, serviceability, and durability will need to be
considered in addition to the emissions aspects.
The SCE is a hybrid power plant that combines certain key
features of spark-ignition and compression-ignition engines in such a way
that some of the more desirable operational characteristics inherent in
each design philosophy are realized.
Fuel injection with charge stratification
permits operation at lean air-fud ratios (producing significant fuel economies)
and the use of high compression ratios (without correspondingly increased
fuel octane requirements) facilitates more efficient power output over a
limited speed range than the homogeneous charge (carbureted) engine.
In contrast with the diesel, positive ignition produces smooth combustion,
permitting a compact, lightweight engine design that combines low cost
with good cold-start characteristics.
Thes e features of the SCE have promising logistics potential for
military purposes and the U. S. Army Tank Automotive Command (USATACOM)
has been sponsoring the development, both by Ford and Texaco, of an SCE
replacement for the standard 4-cylinder military jeep engine.
When tests
on the early, hand-built stratified charge engines showed that low exhaust
-------
emissions were inherent in this design, the EPA contributed financial
support to the USATACOM program to help in the further development of
this engine conc ept.
While the EPA and the USATACOM share many common
objectives in the engine development program, there are some divergent goals
The EPA is concerned with the ultimate feasibility of mass producing more
powerful SCE versions (six or eight cylinder engines, for example), their
compatibility with the requirements of passenger vehicle applications, and
their durability and exhaust emissions performance in terms of compliance
with the 1976 standards.
Accordingly, the EPA is formulating an evaluation
program that embodies its particular objectives.
The study summarized in
this report repres ents an initial phas e.
While the objectives of this present
study were broadly drawn, nevertheless, the primary emphasis has been on
the emissions performance of the engine.
Specifically, the study has compris ed a literature survey desi,gned
to consolidate information on critical design factors, components, and
parameters affecting SCE performance and exhaust emissions together with
indications where trade-offs and compromises are feasible.
Included also
are statistical analyses of available emissions data on the 4-cylinder SCE
to extract such informa~lon as:
repeatability of test results, differences
between sample bags (of the 3-bag CVS test procedure), the statistical
significance of engine adjustments on engine emissions, and possible effects
of ambient test conditions.
Because of the extremely limited quantity of
data on the SCE, statistical analyses were conducted on selected emissions
data for conventional homogeneous charge, spark-ignited, internal com-
bustion engines to evaluate such factors as reproducibility of test data within
a model and within model types, deterioration of emissions with mileage,
ambient effects on emissions and, by analogy, to attempt to relate these
data to the SCE.
2
-------
With these efforts delineating the type of important data and
information that were presently unavailable, the considerations and rationale
were specified for an experiment on existing hardware to provide some of
these data.
Existing hardware in this case refers to several 4-cylinder SCE
installed in 4-speed manual shift transmission, 4-wheel drive, military jeeps
and one 2-wheel drive Postal Service van with Z-speed automatic transmission.
In addition, new hardware requirements for assessing emissions performance
of the SCE were defined from the standpoint of how many units must be built
and tested to have "reasonable assurance" that the developmental SCE will
perform satisfactorily.
3
-------
II
SUMMARY
A survey of the literature related to the Ford (FC P) and Texaco
(TCP) stratified-charge engine concepts indicates there are operational
options possible by adjusting fuel injection and spark ignition timing.
The
timing sch edule can be selected to avoid preignition and detonation, leading
to multifuel capability but poor full power output and smoke problems.
An
alternative schedule achieves excellent full power output at the expense of
susceptibility to knock and preignition.
Critical components, both from the
standpoint of durability and feasibility of mass production, are the fuel
injection pump and nozzles.
Due to the use of catalysts, lead-free fuel is
required by stratified-charge engines.
The FCP engine. installed in an Army jeep and equipped with
emission control devices, has demonstrated operation at exhaust emission
levels which at low mileage are within the 1976 goals.
This performance
has been achieved at the cost of a 30% power loss and resultant poor drive-
ability .
Analysis of limited emissions data has yielded a value of approxi-
mately 15% for the coefficient of variation associated with the "within"
engine variability for each of the three exhaust gas components.
In th e
absence of any data, a similar value (15%) has been assumed for the
coefficient of variability associated with the "between" engine variability.
To provide a broader data base from which to draw inferences
concerning the stratified-charge situation, emissions data from tests on
conventional engines were also analyzed.
These data showed that:
( 1 )
Significant differences can exist between two different makes
of vehicles although good engine-to-engine repeatability can be
achieved.
5
-------
( 2)
Conventional vehicles (with no emission controls) failed to show
any deterioration of emissions with mileage.
(3 )
A wide variation in test error was noted in results obtained at I
different laboratories and with different vehicles.
Since these
variations may mask other effects under investigation, it suggests
that the emissions measuring procedure needs to be improved.
Experiment designs appropriate to testing the limited number of
available stratified-charge vehicles have been delineated.
The tests involve
a study of durability and the effect of parametric variations on emis sions.
A limited number of additional vehicles are
recommended to establish the
engine-to-engine variability which at present is unknown.
Assuming a 15%
coefficient of variation is appropriate to both the test-to-test and engine-to-
engine variability of the SCE, a total of ten engine/vehicle units would suffice
to define mean emission data with a 95% assurance of being within -: 10% of
the expected true population mean.
Following tests of the limited quantity of hand-built engines
available and on order, it is recommended that a like quantity of soft-tooled
engines be acquired for intensive evaluation.
6
-------
III
STUDY RESULTS
A.
Gene ral
A somewhat unconventional structure is employed in this report
wherein the discussion text consists of brief, summary-type accounts of the
various technical studies conducted during the program.
Detail descriptions
of the individual studies, together with numerical results where appropriate,
are included as appendices.
Use of this report form has been occasioned by the natural
evolution of the program phase.
Documentation of the developmental studies
and test results obtained with the prototype SCE has been meager to date.
Accordingly, it was necessary, with the help of the EPA, to identify, locate,
and obtain potentially useful items of information and test data.
As thes e
separate inputs were received, analyses and studies were immediately
made and the results documented in a form that would constitute an appendix
for this report.
B.
Survey of Available Literature on the SCE
A literature survey was made with the intent of consolidating
technical information on the SCE related to critical design factors, components,
and parameters insofar as they affect operational performance and exhaust
emis s ions.
An additional function was the identification of c ompromis e
tradeoff possibilities and their consequences, together \vith an indication
of the known problem areas.
While the stratified-charge concept dates back
several decades and its history includes the developmental efforts of a
number of individuals and organizations, only two specific implementations
7
-------
have been considered here.
These include the Texaco Combustion Process
(TCP) and the Ford Combustion Process (FCP) which are currently in
competition for selection as the SCE replacement for the military L-141
jeep engine.
Physically and conceptually, the two approaches are very
similar with the differences reflecting the differences in the stated design
objectives.
While many of the objectives such as good fuel economy, for
example, were similar, the principal differenc e lay in the fact that the TC P
focus ed on multi-fuel operational capability while the FC P goals dwelt more
on low emissions, performance and smoothness of engine operation.
a r ic h,
Pr oper stratification of the fuel-air mixture, so as to achieve
easily ignitable mixture at the plug electrodes and leaner but still
combustible mixtures at the periphery of the charge, is the key factor in
the successful realization of a useful SCE.
The SCE concept requires in
its implementation the proper coordination of three basic parameters:
swirl, fuel injection, and positive spark ignition.
air
Air swirl rate is a vital parameter in the stratification process
and affects both the efficiency and the duration of combustion.
Such
geometrical devices as shaped inlet ports and shrouded intake valves are
employed to create an initial low-rate swirl in the cylinder.
A combustion-
cup-in-piston-crown configuration achieves the required high swirl rate
near the end of the compression stroke.
Experiments with various cup geometries have shown the largest
effects to be on fuel economy and combustion harshness.
Cup designs incorpor-
ating sharp edges or ridges have proven most satisfactory and the speculation
exists that these surface discontinuities promote turbulence and mixing which in
8
-------
turn creates smooth combustion.
If this supposition is correct, the
accumulation of combustion chamber deposits with usage could serve to
rOtU1d-off these surface discontinuities and produce unfavorable engine
operation.
The military development objective of maximum parts commonality
with the standard L-14l jeep engine necessitates the use of an "oversquare"
bore/ stroke ratio which is unfavorable to swirl creation.
For this reason,
as well as others discussed in Appendix A, it would be desirable to design
SCE prototypes optimized for best engine performance, unconstrained to
utilize design parameters and components of an existing engine designed for
homogeneous -charge combustion.
Another critical factor in the operation of the SCE is the fuel
injection process which must reproducibly create the proper fuel spray
pattern.
The fuel injector is the critical component here and it is susceptible
to "coking" and fouling with attendant variations in the fuel spray pattern.
Durability data are not yet available to determine the severity of these
potential problems.
Basically the SCE may be operated in an unthrottled condition
with the engine power determined by the time duration of the fuel injection.
While the earlier versions of both the TCP and FCP engines were operated
in this mode, poor idle performance has resulted in both systems incorpor-
ating some air-intake throttling over the entire speed range of the engine
In the FC P engine a more or les s conventional spark ignition
system is employed.
A spark plug with long electrodes was used in early
development engines to locate the ignition point within the fuel-rich core of
the swirling charge.
In later (PROCO) versions of the engine employing au
9
-------
throttling and exhaust gas recirculation, this ignition point appeared more
fuel-rich than desirable for flame initiation.
A short-electrode plug geometry
therefore was developed that resulted in improved operation and lower HC
and CO emission levels.
Severe spark-gap erosion has been encountered
because of the use of a high-energy ignition system needed to preclude misfire
at part-power engine operation
The TCP version uses a special transistorized ignition system
with a number of unique characteristics among which is the ability to generate
a constant-voltage, controlled-duration spark.
ignition/injection timing less critical.
This feature makes the
Control of combustion in the SCE depends on the relative timing
of spark ignition and fuel injection with respect to crank angle during the
compression stroke.
The TCP and the FCP versions of the SCE differ most
in this respect and illustrate the consequences of each philosophy.
In th e
Ford system independent advance-retard timing is employed for ignition
and fuel injection as a function of engine speed and load.
This scheme
achieves excellent air utilization at full load operation and hence realization
of full power output without the smoke problem normally associated with
tleterogeneous charge engines under these circumstances.
The pric e paid
for this advantage is that the engine is susc eptible to preignition and
detonation.
Hence useful compression ratios are limited and fuel octane
requirements become more critical.
To achieve and maintain the critical
ignition/injection timing schedule, Ford has developed a unitized distributor
and fuel injection pump for this purpose.
-'-
-,-
The Texaco system appears
to be timed so that the fir s t inc r e rnent
of fuel injected is always ignited.
Thus the possibility of preignition and
detonation is precluded quite independently of the fuel octane rating and a
,~
The literature is not explicit on this point.
10
-------
multi-fuel capability for the engine is achieved.
On the other hand,
air
utilization is poor at high load conditions so that maximum output is exhaust-
smoke limited.
Both the TCP and the FCP versions, as 4-cylinder, hand-built
prototypes installed in jeeps, have successfully demonstrated attainment of
th eir obj ec tives providing:
significant fuel economies, multi-fuel capability
(TCP only), cold start capability, immediate driveaway after cold start without
\,'a nnup, excellent throttle response, and torque/power output equivalent
to the L-l-U engine.
Exhaust emis sions (HC and CO) are smaller than for
a con\'entional engine since such causes of emissions as wall-wetting of
intake manifold by fuel, flame quench at combustion chamber walls, and
deficiency of air are either eliminated or minimized.
Oxides of nitrogen
are about at the same levels for the 4-cylinder SCE and the standard L-141 unit.
EPA tests on a 4-cylinder FCP installed in a military jeep have
produced low-mileage emission levels lower than the Federal 1976 emission
standards when the engine is equipped with exhaust gas recirculation (EGR) for
control of NOx and an exhaust system which includes a thermal reactor and a
catalyst for control of HC and CO. Engine adjustments to achieve this level of
emissions were such that approximately a 30% power loss was experienced with
occasional misfire at high speed. The exhaust gases of the SCE have an unpleasant
odor which can also irritate the eyes and nose. The source of this odor is believed
to be associated with aldehydes and some corrective and control measures have
been propos ed.
No demonstrable success has yet been achieved.
In addition to studies with single-cylinder and four-cylinder
prototype SCE engines, Ford has also converted some large V -8 engines to
a stratified-charge configuration.
Little detailed information is available
11
-------
slnce this latter effort constitutes company-funded research.
These engines
appeared to operate adequately but with certain demonstrated deficiencies
which would not classify them as acceptable in luxury-car applications.
These deficiencies are presumably correctable.
Recent discussions (January 1972) with Ford personnel active in
the FC P program indicate that scaling or extrapolation of performanc e or
emissions data from a single-cylinder to a four-cylinder SCE or a four-
cylinder to an eight-cylinder SCE is complex and unreliable.
One factor
is that each engine size has to be operated with its own unique injection
pump.
Attempts to use one common pump on different size engines have
not been successful.
Another factor concerns the pressure pulsations in
the intake manifold which cause
interaction between cylinders adversely
affecting swirl formation and air induction.
With the lack of adequate theory
in engine development, all approaches reduce to "cut-and-try" type operations.
If the SCE is to fill the role of a satisfactory passenger vehicle
engine and simultaneously meet the emission goals for 1976, two critical
questions must be resolved:
(1) the feas ibility of production of the SC E
in mass quantities to the required tolerances, and (2) the feasibility of the
engine/ emissions control unit maintaining tolerably-low emissions over a
5-year/50, OOO-mile endurance period.
Critical production items are the
injection pump and the injection nozzles.
Durability-related issues that are
unique to the SCE include injection pump deterioration and injector fouling
resulting in inability to operate with consistent reproducibility.
import, but still of concern, is the matter of exhaust odor.
Of lesser
A detailed review of the literature survey study on stratified-charge
engines is included in Appendix A - Summary of a Literature Survey of the
Stratified-Charge Engine Concept.
12
-------
C.
Statistical Analysis of Selected Test Data
1.
Stratified-Charge Engines
Evaluation of the feasibility of mass production of SCE-equipped
vehicles meeting the 1976 emis sion goals requires sufficient data on a
representative sample of these units so that the population distribution (i. e.,
the population mean together with some measure of the dispersion relative
to the mean) of the ultimate production fleet can be predicted with a reasonable
level of assurance.
At the present stage of the SCE development, neither
such data nor such a sample of vehicles exists.
The rationale and the design
details of structured experiments to obtain such data are discussed in
subs equent portions of this report.
Critical to the design of such an experi-
ment is a knowledge of the repeatability of the emissions performance of a
vehicle when subjected to replicative tests (i. e., the with-in vehicle/engine
variance) and the reproducibility of emissions among different units of the
same models of the vehicle/ engine (i. e., the between vehicle/ engine
varianc e).
The strategy of the experimf'l1t depends on the relative levels
of these two variances.
Should the with -in vehicle/ engine variance be
relatively the larger, then the methodology would dictate many replicative
tests on relatively few test units.
If the converse situation prevails, then
fewer tests would be performed on anyone vehicle/ engine unit but more units
would be tested.
When tests are replicated on a single vehicle/ engine unit, a
certain scatter or dispersion occurs
in the resultant emissions measurements.
The sources of this random scatter reside in the inability of the engine to
repeat itself (the with-in engine variance) plus irregularities in instrumenta-
tion, deviations from prescribed test procedures, operator variability,
variations in ambient conditions, etc.
All these sources combine to produce
13
-------
random scatter in the data and the sought-for within-engine variance is
confounded with the variances from other sources.
When tests on different
vehicle/ engine units are made, the measured results are further confounded
by the between-engine variance.
If necessary, the variance components
attributable to the individual sourc es may be separable by us e of the analysis
of variance methodology.
A limited quantity of emissions test data were available on an
FCP engine installed in a military jeep.
These data, consisting of 14 tests
conducted by EPA, Ypsilanti, and 9 tests conducted by Ford, were supplied
by the EPA.
They consisted of emissions (grams/mile) for HC, CO, and
NO as measured by the 3 -bag, CVS-CH procedures using the LA-4
x
driving schedule.
During the course of these tests several changes were
made to the EGR linkage, fuel enrichment at WOT, and deceleration fuel cut-off.
A change of catalyst was also made early in the test series.
Using statistical techniques, the entire data set was tested for
homogeneity (i. e., whether the data belonged to the same population
distribution).
The data identified as homogeneous was then pooled and the
means and standard deviations for each pollutant were computed.
From
these results, the coefficient of variation (defined as the ratio of the standard
deviation to the mean, expressed as a percentage) was computed.
For all
three exhaust gas components the coefficient of variation was essentially
the same, 16%
17%.
It will be appreciated that the standard deviations computed
include not only engine variations but also those due to test apparatus, test
procedures, etc.
Whether these other sources contribute significantly can
not be ascertained at this time.
14
-------
Details concerning the test data, calculations, and other analyses
aimed at evaluating effects of engine adjustments on emissions performance,
as well as a comparison of these data with 1976 standards, are all included
in Appendix B, Analysis of Emissions Test Data on Low-Emission FCP
Engine Installed in M-15l Vehicle.
The emissions data could have been influenced by ambient
factors such as atmospheric pressure, specific humidity, and air temperature,
as well as such imposed variables as dynamometer inertial load and horse~
powe r.
To estimate the effects of these variables on emissions, regression
analysis was applied to the data from the set of 14 tests conducted by the EPA<
A linear regression equation was fitted to these data using the least squares
method and the statistical significance was tested by the analysis of variance.
Results of this analysis indicate that the data from these 14 tests do not
establish any linear dependence of emissions on atmospheric pressure,
specific humidity, temperature, load and horsepower.
Since the combined
effects of the five variables fail to attain a statistically significant level,
it can also be inferred that any single variable would similarly fail to
demonstrate a significant effect on emissions.
Thus the coefficient of
variation calculated from the se data may be expected to be free of any
systematic influences due to these five enumerated variables.
Appendix C
contains a detailed account of the calculations conducted in performing the
regression analysis and the analysis of variance.
It is important to be aware of the fact that these emissions tests
conducted on the SCE were not structured to establish the effects of these
variable s on emis s ions.
That is, no deliberate attempt was made to vary
any of the five parameters over a range of values.
Consequently, only a
naturally occurring small variation in the ambient environmental conditions
15
-------
was experienced and the chassis dynamometer settings also encompassed
a limited range of adjustments. Consequently, the inferences drawn here
only apply to the particular set of circumstances applicable to these test
results. To prove the point, it is an established fact that specific humidity
does affect NO emissions and a correctional procedure is included for
x
these effects in the emissions test calculations applicable to 1976 standards
(the raw data used in this analysis were uncorrected data).
2.
Conventionally-Powered Vehicles
The quantity of emissions data available on stratified-charge
engines is very meager and all of the engines are different from each other
in some respect.
Consequently, this limited data base does not permit
inferences to be drawn concerning variability in emissions attributable to
differences existing among identical engines of the same make (i. e., a
measure of product variability).
To help in augmenting this limited data
base, selected emissions data that were available on current (and recent)
models of conventionally-powered vehicles were statistically analyzed.
It
was felt that this type of information would be us eful in the planning of test
programs for the SCE.
a.
Test Error
This one part of this effort focused on relating the magnitude of
the test error to emission differences ascribable to product variation and
to the other various sources of variation involved.
"Test error" is here
described as the measured fluctuations obtained by repeated tests performed
on the same vehicle by the same equipment under controlled test conditions.
The data encompassed a wide variety of vehicles: new and with accumulated
mileage, with and without emission control devices, and even conventionally-
powered military jeeps.
For comparison purposes, data on the SCE-equipped
jeeps were also included.
16
-------
The emissions data were all found to be non-normally distributed
but rather were represented by skewed distributions.
To a good approximation,
all of the data were found to follow a log-normal distribution.
These data
can be presented most easily by plotting the relative cumulative frequency of
occurrence on probability paper whose abscissa scale is logarithmically
graduated.
A log-normal distribution will plot as a straight line on such
graph paper with the slope of the line indicative of the dispersion of the data.
Data from the various sources, for each of the three components of the
exhaust gases (HC, CO, and NO ) could be represented by such graphical
x
techniques. Depending on the data sourc e, the graphical data could include
not only test error but other sources such as product variability, deterioration
of performanc e with usage, variations associated with different makes of
engines, and the like.
It would thus be logical to expect that the larger the
number of sources of variability operable within a given data set, the larger
would be the dispersion in the data.
In graphical form, this would be
equivalent to expecting the linear plot to have a shallower slope.
To test the reality of this supposition, all of the graphical data
were standardized so as to be presentable on a single sheet of graph paper
with all plots sharing a common intersection (with coordinates corresponding
to an emission of 1 gram/mile and a cumulative frequency of 50%) and
pres e rving their original slopes.
While the data in general supported
expectations, there were a significant number of exceptions.
Data sets
with few known sources of variability often exhibited a larger dispersion
than data sets with many more sourc es of variability involved.
It is suspected,
therefore, that the lack of consistency is associated with the test error, which
is present in all measurements, but apparently is subject to sufficient
variations with vehicle make, vehicle service, measurement techniques and
procedures so as to obscure other active sources of variation.
The conclusion
17
-------
is that the test error can be highly variable but evidence also exists to show
that it can be made small.
Data taken on the SCE showed a relatively small
test error but even smaller levels of test error were observed.
Hence, no
basis exists for assuming that the stratified-charge engines would show a
significantly different test error than an ordinary, carbureted engine.
A discussion of this analysis together with tabulations of the
numerical data employed and the corresponding graphical presentations
are included as Appendix D, entitled Analysis of Emission Test Data of
Various Groups of Automotive Vehicles.
b.
Parametric Effects and Test Procedures
A separate and distinct set of emissions data on fifty-four 1970
model fleet cars was also received from the EPA.
This data set included
cars of various models from different manufacturers and had varying
amounts of accumulated mileage.
Tests had been conducted using the C VS
cold-start and CVS hot-start procedures. A four-bag collection system
was used: hot start transient, hot start stabilized, cold start transi.ent, and
cold start stabilized.
Thes e data permitted the following questions to be
addressed:
does mileage affect emissions, is there a significant difference
between emissions from a cold start versus a hot start, and how are the
CVS-C and CVS-CH test procedures numerically related?
This information,
at least inferentially, was considered as instructive in planning an evaluation
program for the SCE.
An analysis of variance failed to establish any correlation between
emissions data and mileage which for these vehicles spanned the range from
4,000 to 35,000 miles. Similar analyses also failed to show any trends of
emissions with number of cylinders, displacement, and vehicle weight which
18
-------
were the other parameters given (because of commitments to hold some
aspects of the data as confidential, EPA was not at liberty to identify the
vehicles except by range of engine displacement, number of cylinders, and
approximate weight). These analytical procedures are presented in part 2
of Appendix C.
Analysis of the individually-bagged emissions established the
fact that the HC and CO contents of the first (cold start) bag are significantly
higher than the other three.
Differences among the second (cold stabilized),
the third (hot transient), and the fourth (hot stabilized) were found to be
negligibly small (for HC and CO).
Compared to the average of the subsequent
bags, the HC mean of the cold start bag is approximately 1.6 times larger;
similarly the CO mean was 3.4 times larger.
In the case of NO , the transient
x
bag emissions always exceeded those of the stable bag, irrespective of
whether the start was hot or cold.
An investigation of the effect of bag weighting on the computed
mass emission level showed that the CVS-CH procedure results in an HC
level that is 12% lower than the CVS-C method.
The CO level is lower by
29% while NO levels are virtually unaffected. Calculations leading to these latter
x
results are given in Appendix E, Emissions Analysis of 54 Fleet-Type
Passenger Vehicles.
c.
Product Variability
A report published by Automotive Research Associates (ARA)
of San Antonio, Texas, contains results of emissions tests on five 1969
model Fords and five 1969 model Chevrolets conducted at 10, OOO-mile
intervals over a total range of 100,000 miles.
Since the se data failed to
indicate a systematic trend of emissions with vehicle mileage, differences
observed in the successive (10, OOO-mile tests) can be attributed to random
19
-------
causes due to testing errors and actual changes in vehicle performance from
test to test (the "with-in" vehicle variance).
By comparing the "between"
vehicle variance and the "with-in" vehicle variance, an estimate can be made
of the. effect that individual differences among vehicles might contribute to
emissions assessment.
Since such differences might be expected, in part
at least, to reflect effects of manufacturing tolerances, the separation of
"between" and "with-in" variability could permit speculation on the effects
of manufacturing tolerances on SCE emissions.
Using analysis of variance techniques, the emissions data for
Ford and Chevrolet vehicles was used to compute two sets of F-ratios.
The
one set relates to the question of whether the "between" variance for each
make was significant at the 0.05 level.
The other set relates to the question
of whether significant differences exist between the variability of the two
makes of vehicles.
These differences are of two types: one type concerns
the ability of the vehicle to reproduce its performance on successive tests
while the other type concerns reproducibility in the vehicle-to-vehicle sense.
This latter consideration could be indicative that the manufacturing variability
(tolerances) is greater for one manufacturer than the other.
The analytical results indicate that the Chevrolets tested were
more variable than were the Fords, and appreciable emissions variations
can exist among vehicles of the same make. Between-vehicle variability thus
is not always negligibly small. The fact that one of the vehicle makes in these
tests exhibited less variability than the other is indicative that the achievement
of good engine-to-engine re: coducibility is feasible but not always encountered
in practic€. Whether this observation can be extrapolated to the situation of
the stratified-charge engine is not presently known. An account of the details
of the analysis of the ARA data, together with tabulated results, is given in
Appendix F, Exhaust Emissions for Chevrolet and Ford Automobiles.
20
-------
D.
Statistical Modeling Considerations
Exhaust emissions from stratified-charge engines depend on a
number of variables.
Some of these variables can be controlled and their
effects on emiss ions determined by systematic experimentation.
variables are of such a nature that their control is impracticaL
Othe r
The eifect
of these variables on emissions can best be assessed in statistical terms.
In order to put into perspective the various influences which
can affect emissions, it is desirable to have a conceptual model which
delineates how the various contributions to variability combine to produc e the
total variability which is obs erved when many engines are tested under many
different environments and use conditions.
Moreover, it is desirable to
have a rationale whereby the results from a limited number of tests can be
used as a bssis for inferring what might be observed if a larger number of
tests were conducted.
Finally, such a model is essential to the structuring
of additional tests or experiments to fill information gaps in available data.
In Appendix H, Experiment Design Considerations for Emissions
Testing of Stratified ChargL Engines, the influences of the factors which
affect emissions are treated as either fixed effects or random effects.
It
is assumed that the several contributions to variability can be considered
as if they combine in an additive manner.
A simple expres sion of this
additivity is the equation
1-..
l)
:'
r
+
(G,
J
+ E-,
lJ
One of the uses of this equation is in connection with replicate or repeat
tests on two or more vehicles.
If'
J
indexes the vehic les and
L..
indexes
the test on ea,ch vehicle, all test results can be referenc ed to their mean
21
-------
value F ' which can be regarded as a contribution common to all tests.
The J. tf> vehicle has a contribution a j peculiar to it, and, within that
.tJ,
vehicle, the repeated tests make a contribution Etj peculiar to the .0
test. A clear implication of the model is that no amount of replicate testing
on a single vehicle can provide information on the vehicle-to-vehicle van-
ability represented by the term
CLJ .
A full discussion of an approach to
the analytical treatment of sources of variability is given in Appendix H.
The implications of these sources of variability for compliance with emissions
standards requires statistical consideration.
Statistical treatment of the variability in test results from
stratified-charge engines requires the assumption of a statistical distribution
which is believed to be applicable to the data.
Classically, it is assumed
that for a random-effects model the contributions
C1....j and t::. LJ
are governed
by the normal probability distribution.
Under this assumption, the usual
procedures of analysis of variance can be employed and their findings
evaluated according to standard statistics.
As shown in Appendix D, Analysis of Emission Test Data of
Various Groups of Automotive Vehicles, emissions data tend to follow not
the normal distribution but the log-normal distribution.
Reas ons fo r this
behavior are explored in Appendix G, The Log-Normal Distribution as a
Statistical Model for Exhaust Emissions.
In this appendix it is ar gued that
emissions data can be expected to tend toward the log-normal form if
several sources of variability combine multiplicatively rather than additively.
The argument is based on the fact that if one considers not the actual con-
tributions to variability but the logarithms of these contributions, then
multiplication of contributions is converted to addition of their logarithms.
22
-------
Thyn, by virtue of the Central Limit Theorem of mathematical statistics,
it might be ar gued that the combined logarithms would tend to be normally
distributed.
Moreover, the additive model for combining sources of vari-
ability would now be applicable, provided that it is applied to the logarithms
of the emis sion data.
Considering emissions in terms of a normal distribution in
logarithm space is completely equivalent to considering the se emissions
in terms of a log-normal distribution in untransformed (antilog) space.
Indeed, there are straightforward relationships between the parameters of
the two distributions.
For example, the mean in logarithm space is analogous
to the median in antilog spac e, and the standard deviation in log space is
analogous to the "ratio standard deviation" in antilog space.
For example,
if one wishes to compute (approximately) the 95th percentile of a set of
emissions measurements, the following are equivalent procedures:
Log Space:
Mean + Standard deviation + Standard Deviation =
Mean + 2 Standard Deviation
Antilog Space:
Median x Ratio Standard Deviation x Ratio Standard
2
Deviation = Median x (Ratio Standard Deviation)
A further relationship, and one which provides a very useful
approximation, is the relationship between the ratio standard deviation and
the quantity called the coefficient of variation.
It will be recalled that the
coefficient of variation is the ratio of the standard deviation to the mean for
a set of data.
It is a useful quantity in the assessment of emissions because
the standard deviation tends to be proportional to the mean.
It is shown in
Appendix G that, if the coefficient of variation is small (say ~ 0.25), then
Ratio standard deviation
~
1 + coefficient of variation.
23
-------
If the coefficient of variation (or relative standard deviation) is used in the
making of statistical inferences, on the assumption of a normal distribution,
therefore, these inferences will often be substantially the same as thos e
evolving from the use of the log-normal assumption.
E.
Generation of New Data
Emis sion data for conventional internal combustion engines have
been examined with a view toward drawing inferences which may, to at least
a limited extent, apply to stratified-charge engines.
It must be admitted,
however, that such data, at best, aids primarily in understanding the inter-
relations of the several sources of variability.
Inasmuch as the stratified-
charge concept has no real precedent in conventional vehic les, it is es sential
to obtain a data base drawn from tests on actual stratified-charge hardware.
Only a few experimental units incorporating the stratified-charge concept have
been built to date and, for the most part, emissions tests have been conducted
by EPA on only one unit.
In order to supplement this very limited data base
in an effective way, it is essential to design the experimental tests in
accordance with known data requirements and so as to reflect postulated
statistical models.
In view of the need for better as ses sment
of differences
among individual engines or engine-vehicle combinations, an additional
consideration is the question of new hardware requirements.
The structuring
of tests for existing hardware is discussed in the next section;
following
that is a discussion of new hardware requirements to provide an adequate
data base.
1.
Experimental Design for Existing Hardware
A full discussion of a proposed testing program for existing
stratified-charge engines is pres ented in Appendix H.
The testing
24
-------
program is based on the availability of one 4-wheel drive jeep equipped with
Ford Combustion Process (FCP) engine, one 4-wheel drive jeep equipped
with Texaco Combustion Process (TCP) engine, and one 2-wheel drive postal
van,
equipped with FCP engine and automatic transmission.
A few benchmark
tests of a standard (conventional engine) .jeep are included as reference.
With such limitations in available hardware, it is important to
consider carefully which questions can be legitimately attacked by immediate
tests and which questions had best be deferred until more hardware is avail-
able.
For example, with only a few vehicles available, it is virtually
impossible to obtain much information pertaining to vehicle-to-vehicle
variations.
One can, however, consider tests aimed at evaluating mileage
degradation effects on emissions.
Tests aimed at observing the effects of
parametric changes in operating conditions can be considered, but it should
be remembered that design concepts are still in a state of flux and that
conclusions drawn from such tests should be considered tentative.
On the basis of such considerations, a test array was devised
(see Appendix H).
The parametric variations incorporated in the testing
program are directed toward at least preliminary study of the effects of
catalyst, inertial weight, EGR, and mileage on emissions.
The schedule
set forth includes both complete, cold-start tests and tests employing only
the hot-start portion of the test cycle.
In this way, it is believed that economy
of time and effort can be realized without seriously jeopardizing the validity
of the tests.
An important feature of the proposed test program is a set of
tests in which the engine is "rejuvenated" after 50, 000 miles of operation.
This rejuvenation would consist of such simple remedial measures as retuning
and replacing sparkplugs.
The intent of the tests on the rejuvenated engine
is to see how much of the observed degradation in emissions characteristics
is reversible.
25
-------
2.
Definition of New Hardware Requirements
The question of how much data is required in order to make an
assertion at a specified level of confidence is a fundamental consideration
in the definition of new hardware requirements.
The que stion has two aspects:
the statistical repeatability of emis sion measurements and the permis sible
width of a confidence interval for the expected or mean value of the emissions
measurements.
An approach to this problem is presented in Appendix I,
New Hardware Requirements for Assessing Exhaust Emissions Perforrnance
of the Stratified-Charge Engine.
The statistical approach of Appendix I is bas ed on the as sumption
that, for a given category of emission tests, the standard deviation of
emission measurements tends to be proportional to the mean value of the
measurements for that particular category.
Generally speaking, the magni-
tude of the variation will depend on the opportunity afforded for various
sources of variability to come into play.
For example, replicate tests
performed on the same engine would be expected to exhibit less variability
than an equal number of tests performed on different engines.
Data supporting
this observation is provided by Appendix D, Analysis of Emission Test Data
of Various Groups of Automotive Vehicles.
For a particular set of circum-
stances, therefore, it will be assumed that the ratio of standard deviation
to the mean is approximately constant and that this ratio, called the coefficient
of variation, is an important quantity influencing the number of vehicles
which should be procured for further testing.
In the absence of definitive measures of the coefficient of
variation for realistic sets of data for stratified-charge engines, one can
explore parametrically various as sumptions about its value.
In Appendix I,
a table is provided which tabulates the width of a 95% confidence band for
26
-------
various coefficients of variation and for various numbers of tests.
For
example, if it is assumed that test-to-test variability for a particular engine
yields a 15% coefficient of variation, then nine replicate tests would be
required to narrow the half-width of the 95% confidence interval to approxi-
mately 10% of the mean value.
An alternative statement might be that nine
replications are required to define an emission value which has 95% assurance
of being within -: 10% of the "true" or expected value.
The same type of argument can be applied to variations from one
englne to another.
Let it be assumed that for a particular engine a sufficient
number of replicate tests can be conducted to narrow the confidenc e interval
to ne gligible proportions.
Nevertheless, if a population of vehicles varies
considerably from one vehicle to another, many vehicles would need to be
tested in order to estimate the expected value for the population to within a
specified confidence interval.
Again, if vehicle-to-vehicle variation is such
that the coefficient of variation is 15%, then nine vehicles would need to be
procured in order to estimate the expected value for all the vehicles to
within -: 10% of its true value. It is on the basis of this argument that it is
recommended that no fewer than ten vehicles of each type (FC P and TC P) be
acquired to develop an emissions data base.
Needles s to say, there is no assurance that stratified-charge
engines can be built to maintain a 15% coefficient of variation.
However,
there is certainly little reason to believe that a value much lower than 15%
could be realized.
On the contrary (see Appendix F, Exhaust Emissions for
Chevrolet and Ford Automobiles), there is reason to believe that, in some
cases at least, conventional engines exhibit engine-to-engine variability
considerably exceeding this value.
These facts converge in the recommendation
for procurement of ten units.
If variability is greater than 15% of the mean,
additional vehicles could be procured as required.
On the other hand, the
27
-------
conservative value of 15% for coefficient of variation makes it unlikely that
tile recommended ten vehicles would be an excessive number.
If dl la r ge r
coefficient of variation were assumed at the outset, however, it would dictate
the procurement of a larger number of vehicles.
If this estimate of the
coefficient of variation proved to be too large, an unnecessary investment
in vehicles would have been incurred.
It is recognized, of course, that the choice of a 95% confidence
interval is somewhat arbitrary and that the specification of its half-width as
10% of the expected value is equally so.
As is pointed out in Appendix I,
however, these choices represent acceptable compromises between the
opposing factors of increased precision and cost.
F.
Hardware - Information Considerations
One aspect of an evaluation of the stratified-engine involves a
consideration of the types of data that can be most expeditiously obtained
with engines that have been (a) hand-built,
(b) built with soft-tooling
techniques, and (c) built by standard mass production techniques.
A corollary
problem conc erns the relative resour c es that should be allocated to anyone
or all of these options.
In assessing the feasibility of mass -producing stratified-charge
engines with acceptable exhaust emissions performance, the major concern is
,',
with evaluating the within-engine and between-engine variances. '
Therefore,
the relative usefulness of the type of engine manufacture (hand~uilt, soft-
tooled, or production) must be judged in relation to the capability of its
yielding the desired experimental data.
,~
An exceptionally high sensitivity of engine performance to changes in
ambient conditions (temperature, humidity, and pressure) obviously
could not be ignored.
28
-------
Extracting the components of variance associated with the engines
(between and within) involves accounting for all other (non-engine) sources
of variance.
This differentiation can te done by many repeated tests upon
a single test specimen engine (presuming the within-engine variance is small
or separately established).
There is no basis in fact to suppose that the
ability of an engine to provide consistently reproducible performance upon
repeated testing should depend upon the type of manufacture involved.
On
the other hand, the between-engine variance would be expected to strongly
depend upon manufacturing tolerances achieved which are in turn dependent
upon the type of manufacture.
The hand-built engine is used to verify and prove new concepts,
principles and designs.
Tolerances between engines would be expected to
be poorer than in either the soft-tooled or production cases.
Further,
hand-built engines generally tend to be in a continuous state of evolution with
each succ eeding unit differing in some way from the preceding one.
This is
the case with existing hand-built engines so that even an assessment of the
between-engine variability of hand-built engines is not possible unless the
family tested consists indeed of a number of carbon copies.
The major difference between soft-tooling and regular production
tooling lies with the lower initial cost and poorer durability of the former.
As far as tolerances are concerned, those obtained with soft-tooling should
be equivalent to those achieved with normal production tooling.
It is trivial to note that the most reliable data would be obtained
by making tests on samples selected from an actual population of production
engines.
Both economic and time considerations would dictate that a
commitment to even limited tooling of the stratified-charge engine would not
29
-------
be made until compelling evidence of a high probability of success had been
obtained.
Such evidence is presently not available.
Based on the considerations that have been cited in the preceding
discussion, the following recommendations are made:
(1)
Soft-tooled engines will provide the more meaningful test
results.
Since the decision between the competing FCP and
TCP engines must precede any commitment to soft-tooling,
a sufficient quantity of identical hand-built engines should be
procured in the interim to facilitate durability testing.
( 2)
When procurement is initiated for soft-tooling, it should be
sufficient to fabricate a minimum of ten and possibly up to
thirty 4-cylinder, stratified-charge engines with tolerances
equivalent to mass production standards.
( 3)
Procurement should be initiated for ten 4-cylinder, stratified-
charge engines built with the soft-tooling specified in item (2)
above.
30
-------
APPENDIX A - SCMMARY OF A LITERATURE SURVEY OF THE
STRA TIFlED C!lARGE EI\GI:'\E CONCEPT
L. Bogdan
o BJEC TIVE
The purpose of this report is to consolidate technical information,
as gleaned from a survey of the literature, on the critical design factors,
components and parameters that affect the operational performance and
exhaust emissions of the stratified charge engine (SCE).
Compromise
tradeoff possibilities are discussed with their concomitant consequences.
Problem areas are delineated.
INTRODUC nON
Historically. the concept of the stratified charge engine dates
back over forty years.
The initial concept evolved from efforts to control
detonation ("knocking") in internal-combustion, spark-ignited engines by
making the end gases in the cylinders incapable of autoignition.
One me thod
of attaining this objective is to separate or stratify the fuel-air content of the
cylinder in such a way that the end gas lacks sufficient fuel to support auto-
ignition.
Engines employing this pril1<'iple are referred to as stratified
charge engines and, in general, achie\'e stratification by fuel injection late
in the compression stroke.
The principles of the stratified charge concept have been developed
more fully by the Texas Company with exploratory research commencing in
the late 1940's and by the Ford Motor Company which first built a successfully
operating laboratory engine in 1960.
While significant research by other
organizations has also been conducted, the two aforementioned companies
dominate the field currently and have each fabricated a quantity of prototype,
hancibuilt, multicylinder stratified cha "ge engines.
This paper accordingly
will restrict its scope to a consideration of the two engine types based on the
Texaco Combustion Process (TCP) and the Ford Combustion Process (FCP).
A-I
-------
The SCE, often refer red to as a hybrid engine, combines
essential elements of both the spark-ignition and compression-ignition
internal combustion engines and also enjoys some of the better operational
consequences of both design philosophies.
Fuel injection with attendant
stratification facilitates lean operation resulting in significant fuel economies
and the use of high compression ratios, without constraints on fuel octane
requirements, permits attainment of efficient power output relative to a
carbureted, homogeneous charge engine.
Positive ignition, on the other
hand, results in a smooth combustion process that permits a compact,
lightweight engine design combining low cost with excellent cold-starting
characteristics as contrasted with the diesel engine.
These salient features of the SCE attracted the interest of the
U. S. Army Tank Automotive Command (USA TACOM) which is presently
supporting the development of the TCP and FCP concepts with the .objective
of realizing an SCE replacement'for the standard 4-cylinder L-14l jeep
engine.
The goals of the program are to develop an SCE engine having
equivalent performance characteristics to the L-14l but with significantly
better fuel economy and an ability to operate satisfactorily using a wide
range of military fuels.
A further objec tive is a maximum measure of
parts commonality with the standard L-14l engine.
Early tests conducted on the TCP and FCP versions of the
L-141 engine indicated that low exhaust emissions were also characteristic
of th is des i g n .
Consequently the Environmental Protection Agency (EPA)
is contributing financially to the support of the USATACOM development but
with the express interest of ascertaining the technological feasibility of the
SCE concept meeting the 1976 emissions and durability standards in passenger
vehicle applications.
A-2
-------
DESIGN CONCEPTS OF THE SCE
The TCP and FCP concepts have been aptly called controlled
combustion and programmed combustion processes.
Before discuss ing
the broader aspects of the design of the two approaches, which physically
and conceptually are very similar, it is well to contrast the differences as
exemplified by the stated objectives of the two companies.
It mus t be borne
in mind that Texaco is not an engine manufacturer and hence has stressed
concept unders tanding in its research efforts.
Further, Texaco's engine
development effort appears to be circumscribed by the USATACOM L-141
engine development program whereas Ford, besides its role in the L-141
project, has built and tested large-size V-8 SCE's (up to 534 in.3 displacement;
the L-14l has 141 in.3 displacement) with a view to passenger vehicle and
(possibly) truck applications.
Texaco objectives for the TCP engines, as conditioned by
USA TACOM requirements are:
.
good fuel economy
.
broad fuel tolerance
.
good dura bility
Ford objectives for the FCP engine include:
.
fuel economy
.
smoke-free power output equivalent to carbureted
engine operated on regular grade gasoline
.
cold start capability without special aids
.
parts commonality with carbureted engine
A-3
-------
.
speed/torque characteristics approximating carbureted
engine for drive train/exhaust system commonality
.
low emission characteristics
.
engine smoothness consistent with passenger vehicle
standards
The stratified charge engine, as exemplified by the TCP and
FCP, may be described a spark-ignition, internal combustion system in
which fuel-air content of the cylinder is so stratified, using fuel injection,
that the richest mixture is concentrated in the region of the spark plug
electrodes.
Thus ignition of the charge is readily initiated and operation at
lean overall fuel/air ratios Cj.t part load conditions is feasible.
Inlet air
throttling is not required with the power level regulated by the duration of
the fuel injection.
Studies have established the fact that one of the key aspects of
the stratification process, which is essential to the successful realization
of a useful engine, is the development and maintenance of a suitable swirling
motion of the air within the cylinder.
In fact, implementation of the SCE
concept is entirely dependent upon proper coordination of three primary
parameters: air swirl, fuel injection and positive spark ignition.
Each of
these parameters will be examined in turn in terms of physical implications
of design and relation to performance.
It will be appreciated that exploratory
experimentation underlying the development of useful design data has been
empiriCal in nature using trial-and-error type procedures.
A-4
-------
AIR SWIRL
Air swirl rate (rotational velocity of the air relative to crankshaft
angular velocity) has a direct bearing on the combustion duration and efficiency
serving as it does to direct the injected fuel to the flame front and to control.
the evaporation and dispersion of the fuel.
The realiza tion of a suitable
swirl rate with an attendant high volumetric efficiency over the useful operating
range of the engine constitutes one of the primary design variables.
A swirling airflow geometry free of counterflows and irregularities
is conducive to consistent engine operation from cycle to cycle.
To achieve
thls goal much experimental effort has been devoted to define geometries
for:
directional intake ports, stationary shrouds around intake valves and
earn profiles and valve lifts.
All of these factors serve to induce the initial,
low-rate swirl that exists during the early phases of the compression stroke.
The necessary high swirl rates are achieved by means of a recessed, centrally
located, combustion cup or chamber which is an integral part of the piston
crow n.
As the piston approaches the top of its compression stroke, the
swirling air is constrained to within the confines of the smaller-than-bore
diameter of the cup and the swirl rate increases in consequence of the
conservation of angular momentum.
Radial dimensions of the cup affect
swirl rate while the volume determines the compression ratio.
The cup dimensions and geometry have a significant impact 00
the combustion process for reasons that are not completely understood.
Test data have shown that the ratio of the area of the cup aperture to the
bore area represents a sensitive parameter.
If the ratio is large, fuel
economy ic; poor; when the ratio is small, combustion roughness occurs.
A typical ratio is 0.42 which suggests that charge compactness is necessary
during combustion to provide a high rate of heat release.
A-5
-------
A wide range of cup geometries has been explored:
truncated
cone, cone cup, double-cone cup, etc.
The main effect reported is on
fuel economy and combustion harshness with the double-cone cup most
successful.
This performance is attributed to the sharp-edged surface
discontinuities at the cone-cone and cone-cup intersections which create
'"
flow turbulence and mixing resulting in accelerated heat release with smooth
combustion.
This assumption has been supported by tests which demonstrated
poorer performance when the cup edges were rounded.
To a degree, the cup-in-piston concept is a consequence of the
necessity to achieve adequate swirl rate within the restrictions imposed by
the oversquare bore-to-stroke ratio of the basic L-141 engine which represents
an unfavorable geometry in tha t res pee t.
In addition, in SCE applications,
large bore-to-stroke ratios have been determined unfavorable from the
standpoint of good thermal efficiency and low hydrocarbon (HC) emissions.
To fully explore and exploit the potential of a stratified charge
engine concept, it seems unreasonable to fetter the design of prototype models
by slavish insistence upon maximum parts commonality with an existing
engine design optimized for different combustion processes.
By permitting
the bore-to-stroke ratio to be a design variable, it appears that improvements
in air swirl formation, engine thermal efficiency and HC emissions could be
realiz ed.
To the extent that such artifac ts as the combustion cup with its
surface discontinuities (ridges) are crucial to successful operation of the
SCE with the L-141 configuration, it is possible that combustion deposi.ts,
by altering critical shapes, could unfavorably affect durability.
Thus an
i.mproved bore-stroke design could also result in lesser deterioration of
engine performance (including emissions) with accumulated mileage.
A-6
-------
FUEL INJEC TION
Optimization and the consistent reproducibility of the fuel
injection process represents the very crux of the stratified charge engine
operation.
In this regard the fuel injector is the critical mechanical
component since the mixture formation is controlled primarily by the
characteristics of the fuel spray.
A large amount of empirical study has been expended in deter-
mining the proper location for the fuel injector, the angle of the fuel
injection (relative to the bore centerline), fuel injection pressure, fuel
injection rate and fuel spray characteristics (spray angle).
The objective
has been to achieve satisfactory operation over the entire speed-load range
capitalizing upon the potljntial of lean operation of the stratified charge
principle at partial loads while being able to realize the full output
capability of the homogeneous charge principles at maximum loads.
Tests have shown that at light load operation it is desirable to
confine the fuel in the proximity of the center of the bore while at high loads
good air utilization requires that more fuel be directed radially outward.
Because of mutual interaction, a compromise is effected between injector
orientation and injector spray angle.
required to achieve low penetration.
Basically, a wide angle fuel spray is
This condition may lead to cylinder
head or wall wetting by the fuel resulting in rough combustion.
A narrow
angle spray increases penetration and hence dispersion with misfire or
poor combustion characteristics at light loads.
Injection pressure is another factor affecting fuel penetration
but more significantly is its role in the atomization of fuel.
Too low injec tion
pressures appear to produce poor atomization with resultant losses in fuel
economy and increases in HC emissions.
Injector opening pressure typically
A-7
-------
range from 300 to 450 psi since higher pressures do not seem to produce
any advantages.
The fuel injection rate is constrained to a very low level so that
the duration of the injection is extended to facilitate operation and render
timing coordination less critical.
Other benefits accruing from low fuel
mass injection rates are:
lesser disturbance of air swirl pattern, better
fuel evaporation and improved fuel-air mixing.
Much effort has been expended by Ford on the design, development
and evaluation of injector nozzles to achieve reliability, consistency of
performance and freedom from fouling.
Designs inducing valve vibration
and rotation have been favored since they tend to:
promote atomization of the
full, minimize carbon buildup, provide better valve seating and maintain
better spray concentricity.
These designs seem to be susceptible to cyclic
variations in fuel delivery resulting from mechanical resonances in the
Ford h..., similarly developed specialized, multicylinder injection
pum.ps to optimize the entire fuel injection system.
valve.
Texaco has apparently been able to successfully adapt standard
diesel injection pumps and nozzles to their TCP engines.
Ford's expe rienc e
led them to reject commercially available components as being unsuitable.
POSITIVE SPARK IGNITION
Conceptually, the FCP and TCP systems employ the 'conventional
coil, distributor, spark plug system.
In practice, the differences from
convention are significant.
Rather detailed information on the FCP develop-
mental efforts in this regard is given in the open literature.
A-8
-------
Ford conducted extensive studies to determine the optimum
spark plug location for consistently reliable fuel mixture ignition over the
operating range of the engine.
In the early designs, this requirement was
best realized with the gap in the center of the fuel spray near the bore
centerline and about one-half inch below the cylinder head face.
The bas is
for this location seemed to be that the central core of the mixture cloud
rotates as a solid body without appreciable dispersion and hence is an ideal
point for flame initiation.
One problem with this gap location was that long electrodes on
the spark plug were required.
In addition, due to the asymmetrical electrode
geometry, the orientation of the plug was important and a fixed-orientation
plug installation was mandatory.
A potential difficulty with long electrodes
is their tendency to overheating with resultant preignition.
In the latest (PROCO) version of the engine which makes use of
air throttling and exhaust gas recirculation, the center of the spray mixture
was deemed to be richer than desirable for flame initiation, especially under
medium to heavy load conditions.
Consequently a new, short-electrode
spark plug has been developed with the gap located at the edge of the fuel
spray (above the spray centerline).
This location has had several beneficial
effects such as:
increased tolerance to EGR, lower specific fuel consumption,
and decreased levels of HC and CO emissions.
A slight increase in NO
emission was observed but this condition can be corrected by changes in
EGR rate and/ or ignition timing.
In the FCP engine, spark gap erosion problems have been severe
(a gap growth rate four to five times that experienced in carbureted engines).
This erosion is a consequence of the high-energy requirements of the igniti.on
A-9
-------
system to preclude part-power misfires and the high electrode temperatures.
After studying various alternatives, Ford finally resorted to riveted precious
metal (gold-palladium) inserts in the gap area.
Texaco makes no reference to the spark plugs used in its TCP
engines but rather emphasizes development of a special transistorized
ignition system with the following spark characteristicsc high energy output,
controlled duration at essentially constant voltage, restrike capability and
very fast rise time.
This system is claimed to permit operation in a
satisfactory manner at leaner mixtures than is possible with a conventional
ignition system.
A-IO
-------
INJECTION - IGNITION TIMING
The coordination of the relative timing among fuel injection,
spark ignition and crank angle on the compression stroke is vital to the
achievement of proper combustion control in the stratified charge engine.
In the FCP and TCP versions, different philosophies have been adopted with
respect to timing resulting in different operational benefits.
are made manifest in the following discussion.
The differences
The Ford system utilizes independent advance-retard timing
schedules for spark and injection as functions of engine speed and load.
light loads a well-stratified mixture is essential to achieve satisfactory
At
ignition.
To achieve this end, the fuel injection occurs late in the compression
stroke.
A small, stratified mixture cloud is formed in the proximity of the
centrally-located spark gap.
The mixture cloud gradually diffuses in the
swirling airflow pattern while remaining centered on the spark gap.
Ignition
is initiated during the period of injection.
With increased loads, injection
timing is advanced to allow more time for fuel dispersion/evaporation for
faster and more complete combustion.
At the same time, spark timing is
retarded as combustion rate increases with richer mixtures. With increasing
engine speeds, both injection and ignition timing are advanced 80 that the
mixture cloud formation and the combustion process can accommodate the
higher pis ton speeds.
Ford's timing schedule provides excellent air utilization at
high loads and thus permits the realization of full power output without
the smoke problems normally associated with heterogeneous charge engines
under these operating conditions.
On the other hand, the engine becomes
susceptible to preignition and detonation difficulties with attendant 1imitation~
on the useful ranges of compression ratios and fuel octane ratings.
A-II
-------
The relative phasing of the injection and ignition timing is
critical and the magnitudes are large.
For example, injection timing covers
a range of approximately 400 with speed and 600 with load.
Similarly, the
respective values for spark timing are 150 and 250.
To achieve and maintain
this timing schedule over the useful life of the engine by means of independent
adjustments would require exceptionally knowledgeable and skilled maintenance
personnel.
Ford is therefore opting for a unitized .injection/ignition system
incorporating the proper timing schedule.
Texaco literature lacks preClse detail concerning the injection/
ignition timing schedule used in the TCP.
From basic considerations, it is
conjectured that the fuel injection timing relative to crank angle must
approximate that of the FCP.
Ignition, however, appears to be timed so
that the firs t inc rements of the injec ted fuel are alwa ys ignited.
This
ignition time precludes the possibility of preignition and eliminates the
possibility of detonation regardless of the octane rating of the fuel.
The
realization of these possibilities has been amply demonstrated by actual
in-vehicle tests.
The cQnsequence of these real advantages is that the air
utilization at high load conditions becomes poor so that the maximum output
is exhaust smoke limited.
The relative freedom in selecting compression rates makes the
TCP a ready candidate for turbocharging.
Dynamometer tests of a turbo-
charged TCP engine (without changes to the injection system or valve train)
have shown no misfire or knock, no adverse effects on part-power fuel
economy but with a good improvement in high load performance.
Vehicle
tests demonstrated improved fuel economy (relative to a naturally aspirated
engine) at all but the very low operating speeds (for all fuels tested).
A-12
-------
The TCP and FCP demonstrate the extent to which injection/
ignition timing may be used to attain different operational objectives; in brief,
a broad fuel tolerance for the TCP and excellent, smoke-free full load operation
for the FCP.
Considerable latitude appears to exist here to explore inter-
mediate timing schedules to achieve a "best of both worlds" performance
from a stratified charge engine.
PERFORMANCE
A.
Four Cylinder SCE
Both the TCP and the FCP have fulfilled their different design
objectives while equalling or exceeding the torque/speed characteristics of
the standard carbureted L-141 engine.
Good cold weather starting has been
achieved and the ability of immediate after -start driveaway without the need
for engine warm-up is facilitated by fuel injection.
Excellent throttle response,
bordering on harshness, has been demonstrated. Over-the-road tests
provided excellent proof of fuel economy, relative to the standard engine, of
40% to 70%.
The unthrottled SCE has poor idling characteristics which may be
resolved either by employing high engine idle rpm or by incorporating air
throttling that is only operable at idle.
While the early prototypes were
unthrottled, the trend in the latest versions is to incorporate some throttling
over the entire spectrum of engine operating conditions.
In common with standard carbureted engines, the addition of
emission control devices to the SCE has resulted in a loss of performance
(power) and fuel economy.
This is especially true in the use of exhaust
gas recirculation (EGR) to control oxides of nitrogen in the exhaust.
A-13
-------
B.
~.lght Cylinder SCE
Ford has converted, on its own funding, 430 CID and 534 CID
V-8 engines to the FCP configuration using the nominal valve train and
cs.msha£u; fdund in these engines so that maximum efficiency was not realized.
In passenger' vehicle application, the smaller engine produced gains in fuel
economy oi about 30% over a wide range of operating speeds (30-70 mph).
AClc eIeI'a tion and pass ing performance was found to be somewha t defic ient
despite dYrlamometer test data which indicated power equal to (or better than)
'.he '::<3 sic standard engine.
Serious idle problems were encountered which
necess ita ted idle throttling.
The high peak instantaneous torques of the FCP
I'E'8,ulted in noise, vibration and harshness characteristics considered
uoacceptable in the luxury-type cars normally using this size engine.
In:-lp1oved engine mounting would be required to alleviate these latter
d i':r~ (ulties.
The larger engine (a truck engine) was converted principally to
obtain scaling data (Ford does not consider FCP engines as candidates for
:ynd: applications).
Compared with the 430 CID engine, test data on the
534 CLD unit showed that:
compression ratio had to be reduced to eliminate
spark knock, rnisfire was rnore prevalent, severity of spark gap erosion
increased, high-speed fuel econorny was poorer and control of hydrocarbon
e:-:iliE,sions was more difficult.
In terms of scaling effects, it was concluded that:
(a) increased
bo!';:; diameter necessitates '.lse of lower cornpression ratios which increase
t,c;ndency to misfire and ret"J.ire higher energy ignition and (b) increased bore
c:1i.a.rneter increase,'.> the volume of very lean mixture which isn't burned and
hl,once results in increased HC emissions.
Ford concluded that fuel consumption
.:J,nr.1 exhaust emission data show that good correlation exists between single and
rC:cl1lticylinde r engines.
A-14
-------
EMISSIONS
In homogeneous charge, internal combustion engines, a large
measure of the exhaust emission is associated with wall-wetting of the
intake manifold by fuel, flame quench at the walls of the combustion chamber
and a deficiency of air.
Emissions from the SCE are smaller to the extent
these conditions are eliminated.
In the SCE the principal source of hydrocarbon emissions is the
stray droplets of fuel that escape from the fuel spray and the overly-lean
mixture in the peripheral regions that does not burn or burns incompletely.
Unburned hydrocarbon content in the exhaust decreases when injection timing
i.s retarded and/or spark timing is advanced since the time for fuel dispersal
prior to flame front arrival is decreased.
Intake throttling also has a favorable
effect in decreasing HC emission since the fringe area leanness is reduced
thus facilitating outward penetration of the flame producing more thorough
combustion.
With an excess of air available at all operating conditions, carbon
monoxide emissions have assumed negligible proportions.
If additional
throttling is tobe added to the basic SCE to realize other engine operational
benefits, it is possible that CO emissions will be unfavorably affected.
The oxides of nitrogen in the exhaust of an SCE are approximately
equivalent to those found in the conventional counterpart engine.
As in
conventional engines, EGR has been found to be an effec tive control to
reducing NOx.
The stratified charge engine is at a definite disadvantage
(relative to homogeneous charge engines) in the matter of aldehyde emiss ions.
Aldehydes are believed to be the cause of the distinct odor associated with
A-l5
-------
SCE exhaust gases.
This odor is unpleasant to the sense of smell and in
sufficient concentrations irritates the eyes and nose.
Partial throttling has
been found to reduce the odor which also is affected by ignition and spark
timing.
With a compromise throttling and timing situation, a catalytic
muffler might be required for overall control.
A potential problem is the
relatively low level exhaust gas temperature found in the SCE.
consequence the catalyst efficiency could be impaired by low operating
As a
temperatures.
Recent emissions data taken by EPA on the 4-cylinder FCP
installed in a military jeep and equipped with a properly-tuned EGR system
and an exhaust catalyst have shown its ability to meet Federal 1976 emissions
s tanda rds.
To achieve these levels, however, EGR was used at wide open
throttle without simultaneous fuel enrichment.
This mode of operation
results in about a 300/0 loss in power with resultant poor acceleration and
misfire at high speeds.
Mutually conflicting requirements are therefore seen to exist
between achieving low HC/CO emissions and low NOx emissions.
A similar
situation obtains between low NOx emissions and engine full load performance
Obviously these conditions represent an area where trade-off and compromise
between emissions, engine performance and engine size (to make up for loss
of performance due to the use of emission control procedures and devices)
is both possible and desirable.
PROBLEM AREAS
As viewed from the standpoint of whether the SCE has the ability
to fulfill the role of a satisfactory passenger vehicle power plant and still
meet the Federal emission requirements for 1976, two principal questions
A-16
-------
will need to be resolved:
(a) the feasibility of mass-producing the SCE to the
necessary tolerances and (b) the capability of maintaining low emissions
performance over the necessary 5-year /50,000 mile endurance period.
Production feasibility is concerned primarily with two crucial
components; the fuel injection pump and the fuel injection nozzles (especially
in the case of the FCP).
The feasibility of mass producing a satisfactory
unitized injection/ignition system is also a vital consideration.
Durability considerations (unique to the SCE - deterioration
of emission control systems is a problem common to all internal combustion
engines) include injector coking, fouling or inability to render consistent
performance, failure of the injection pump to render consistent performance
and spark plug gap erosion.
Exhaust gas odor, characteristic of most heterogeneous charge
engines, will need to be controlled if the SCE is to be a thoroughly acceptable
vehicle power plant.
In the multifuel SCE, smoke emissions at high load
conditions represent a problem. While different methods of attack have been
put forward as possible solutions, no successful demonstrations of satisfactory
control are kuown.
RECOMMENDA TrONS
Within the limitations of the literature search conducted, it
appears that all of the multicylinder SCE engines that have been built and
tested to date represent adaptations, to a lesser or greater degree, of an
existing carbureted-engine design.
The end objec tive has generally a
maximization of parts commonality.
As discussed in the preceding sections,
this approach imposes restraints in design that preclude the full potential
of the stratified charge combustion process to be achieved.
It is therefore
A-17
-------
recommended that in an early stage of the evaluation process of the pI
available family of SCE models, if the produc ibility and durability que
are satisfactorily answered, that an engine fully optimized to the strat
charge process and configured for low emissions be designed, built an
evaluated.
A-18
-------
BIBLIOGRAPHY
1.
Ta ylor, C. F., "The Internal Combus tion Engine in Theory and
Practice - Volume II: Combustion, Fuels, Materials, Design,"
The M.L T. Press, Cambridge, Massachusetts, 1968.
2.
Bishop, 1. N. and Simko, Aladar, "A New Concept of Stratified
Cha rge Combus tion
The Ford Combustion Process (FCP),"
SAE Paper 680041, Janua ry 1968.
3.
Mitchell, E., Cobb, J. M. and Frost, R. A., "Design and
Evaluation of a Stratified Charge Multifuel Military Engine," SAE
Paper 680042, January 1968.
4.
Summerson, W. A. and Mitchell, E., "Small Military Engine
Stratified Charge Combustion Yields Multifuel Capability,"
Automotive Industries, December 1967, pp. 71-76.
5.
Haman, A. C., Cheklich, G. f-:. and Kaupe rt, A. W., "R eview
of the USA TACOM Hybrid Combustion Engine Program, "
U. S. Army
Tank A utomotive Command, no numbe r, A ugus t 1970.
6.
A Stratified Charge
"The Texaco Controlled -Combus tion Sys tem
Engine Concept, " Texaco Research Center, no number, October 1970.
7.
"The Texaco Transistorized Ignition System," Texaco Research
Center, no number, November 1970.
8.
"Engineer ing Know -How in Engine Des ign
Part 19,"
Soc iety
of Automotive Engineers, Inc., SP-365, June 1971.
A-19
-------
Bibliography (continued)
9.
"Control Techniques for Carbon Monoxide, Nitrogen Oxide, and
Hydrocarbon Emissions from Mobile Sources," U. S. Department of
Health, Education and Welfare, AP-66, March 1970.
10.
Simko, A., Choma, M. A., and Repko, L. L.,
"Exhaust Emission
Control by the Ford Programmed Combustion Proces s - PROCO, "
SAE Paper 720052, January 1972.
A-20
-------
APPENDIX B
ANAL YSIS OF EMISSIONS TEST DATA ON LOW-EI\1IS-
SIaN FCP ENGINE INSTALLED IN MI5I VEHICLE
D. J. Schuring
INTRODUCTION
The Ford Motor Company, under contract with U. S. Army
Tank Automotive Command, adapted its stratified-charge combustion
process (FCP) to the L-141 engine of the M 151 vehicle (jeep). Ford
developed two types of engines -- a low-emission engine and a best-
ecorlOmy engine -- and installed them in four vehicles, three jeeps and
one two-by-four post office truck.
Of the four vehicles, only vehicle
No.2 (2M5733), a jeep with a low-emission FCP engine, is of interest
here. The engine of this vehicle was equipped with exhaust gas recircu-
lation (EGR with heat exchanger) and a PTX-6 catalyst.
The vehicle has been repeatedly tested on a chassis dynamom-
eter at Ford and EPA using the 1975 testing and sampling procedure (CVS-
CH with three bags). (This procedure could not be completely followed,
though; the engine lagged significantly behind the prescribed acceleration
rates during the 192 215 sec. portion and the 452 470 sec. portion of
the driving cycle because, in order to keep emissions low at high accelera
tion rates,
the EGR was not closed and fuel enridunent was not applied
-,+ wide open throttle.) During the tests, a few changes of the engine and
the emission-control harcware were made, such as replacement of a
faulty catalyst, adjustment of the EGR system, and adjustment of the
deceleration fuel cutoff linkage.
The data of all tests were analysed with the objective of
clarifying the following questions.
What are the average values and the variance of the three
pollutants of the given engine?
Did the changes made during the tests significantly influence
the emission level?
To what extent did the tested vehicle meet the 1976 emis sion
standa rds?
B-1
-------
Data
The results of emission tests performed between July 22 and
September 29 (1971) were documented in three Ford Engineering Progress
Reports(l-3) and in computer printouts of EPA(4). Tables 1 through 6
list all pertinent data.
Data set No.1
(Table 1)
Data set No.2
(Table 2)
Data set No.3
(Table 3)
Data set No.4
(Table 4)
Data set No.5
(Table 5)
4 tests run between July 22 and July 29 at Ford.
Faulty PTX-6 palladium catalyst may have accounted
for high HC and CO data and for large data spread.
The Ford Progress Report remarks that the very
high CO-value of 5.2 gr /mi (July 26) was due to low
barometric pressure at the day of testing.
Since a
significant correlation betwe,en barOfil'letric pressure
and emission level could not be established, however,
(as reported elsewhere) we tend to ascribe the unusu-
iMlly high CO value to other unknown reasons. In any
event, the CO value of 5.2 gr/mi was disregarded in
furthe r data evalua tion.
5 tests between August 13 and 20 at Ford with a new
PTx-6 platinum catalyst.
5 tests between August 30 and September 3 at EPA,
Ypsilanti.
4 tests between September 7 and 9 at EPA.
Ypsilanti. Ford personnel adjusted accelerator
linkage to prevent closing of EGR at wide open
throttle.
5 tests between September 9 and 14 at EPA,
Ypsilanti. Ford personnel adjusted the deceleration
fue 1 cutoff setting.
B-2
-------
b:J
I
.....,
EGR Vacuunl Inlet Bar om.
Date HC-Fill CO-IR NOx-C 1 Orifice EGR Inertia Pres. Pres. Temp. Humidity
1971 g/mi g/mi g/mi diam. Setting Weight in Hg in Hg of %
7-22 .31-\ 1.6 . 34 29. 32
7-23 .32 1.9 . 35 0.438 .22
7-26 .44 5. 2 .28 .05
7-29 . 51 1.8 .28 i/ .28
mean 2.63
-
x .413 (1.77)* . 313
Std
Dev 1. 73
s .0,814 (. 153 )';< .0377
-- ------ - -~--
s /-;: % .. 20 :; 66
("'9 )';' - 12
TABLE 1
Emission test data of Vehicle No.2 (2M5733) tested at Ford, Dearborn
Source of Information: Ford Engrg. Progress Rpt. 14 (July 1971)
Vehicle mileage 26,949. FCP engine mileage 3502. Apparently tests were
faulty PTX-6 palladium catalyst
run with
':' without CO
5.2 g/mi.
-------
tJj
I
*"
EGR Vacuum Inlet Barom.
Date HC-Fill CO-IR N Ox - C 1 Orifice EGR Ine rtia Pres. Pres. Temp Humidity
1971 g/mi g/mi g/mi diam. Setting Weight in Hg in Hg of %
8-13 .28 .54 .42 . 375 ]3000 29.32
8-16 . 33 .32 .37 .46
.50
8-17 .35 . 21 . 35 .40
8-18 .27 . 14 . 32 .404 2000 .28
8-20 .27 .42 .41 I .25 3000 . 16
I
mean
-
x .300 .326 .374
Std.
Dev.
s .0374 . 160 .0416
s /x % "'12 :::49 :::1 11
TABLE 2
Emission test data of Vehicle No.2 (2M5733) tested at Ford, Dearborn
Source of information: Ford Engrg. Progress Rpt.15 (August 1971)
Vehicle Mileage 27,055, FCP engine mileage 3,566. Tests were run with new
platinum catalyst.
PT X - 6
-------
tJ:J
I
U1
EGR VacuuD1 Inlet Barom.
Date HC -FID CO-IR NOx-C 1 Orifice EGR Ine rtia Pres. Pres. Temp Humidity
1971 g/mi g/mi g/mi diam. Setting Weight in Hg in Hg of 0/0
8-30 .41 .98 .45 26.20 29.36 81 46
8-31 . 37 .72 . 39 . 30 .46 72 56
9 -1 .46 .84 .39 2750 .27 .42 76 60
9-2 .46 . 82 . 37 .21 .38 76 73
9-3 . 33 . 81 .49 .22 . 37 75 75
mean
- .406 .834 .418
x
1---
Std.
Dev
s .0568 .0937 . 0624
- ~ 11 =12
six % = 14
TABLE 3
Emission test data of Vehicle No.2 (2M5733) tested at EPA, Ypsilanti
Source of information: EPA Computer Data
Vehicle mileage 27,155, FCP engine mileage 3670 (estimated)
-------
tP
I
0"
EGR Vacuum Inlet Barom.
Date HC-FID CO-IR N Ox - C 1 Orifice EGR Ine rtia Pre s. Pres. Temp Humidity
1971 g/mi g/mi g/mi diam Setting Weight in Hg in H g. of %
9-7 .37 .92 . 21 2750 26.13 29.32 74 74
9-7 .47 1. 17 . 34 2750 27.66 .25 81 64
.
9-8 .39 .92 .32 3000 26. 12 .25 83 61
9-9 . 32 .83 . 31 3000 . 19 .30 76 65
mean
- .960
x .388 .295
Std.
Dev
s .0624 . 146 .0580
sf;' % -:::. 16 = 15 = 20
TABLE 4
Emission tes t data of Vehicle No.2 (2M5733) tested at EPA, Ypsilanti
Source of information: EPA Computer data
Vehicle mileage 27,227, FCP engine mileage 3730 (estimated)
EGR linkage adjusted by Ford personnel to prevent EGR cutoff at WOT
-------
IJj
!
--.j
EGR VaCUUDl Inlet. Barom.
Date HC-FID C O-IR NOx-C 1 Orifice EGR Ine rtia Pres. Pres. Temp. Humidity
1971 g/mi g/mi g/mi diam. Setting Weight in Hg in Hg OF %
----"---.- --- -- ~---~--- - - -.- ~------ -- - -- - - --"--
9-9 .27 . 58 . 28 26.16 29. 38 81 50
9-10 .30 . 55 .29 .07 . 18 74 64
3000
9-13 . 32 .78 .26 25.94 .03 70 63
9-13 . 31 1. 20 . 25 .91 .00 72 58
I
9 -14 .40 1. 24 . 35 .91 28.99 70 62 I
I
:~1= --
x 9320 .870 .286
-~ ---- I
Std. I
Dev
s . 0485 . 332 .0391
i---:--=- 01 -- I----~ -----
i S / x 10 = 15 ;:::; 38 ~ 14
-----
TABLE 5
Emission test data of Vehicle No.2 (2M5733) tested at EPA, Ypsilanti
Source of information: EPA Computer data
Vehicle mileage 27,279. FCP engine mileage 3790 (estimated)
Accelerator linkage adjusted by Ford personnel to prevent closing of EGR at WOT
Deceleration fuel cutoff adjusted by Ford personnel
-------
IJj
I
OJ
EGR Vacuum Inlet. Barom.
Date HC-FID C O-IR NOx-C 1 Orifice EGR Ine rtia Pres. Pres. Temp. Humidity
1971 g/mi g/mi g/mi diam. Setting Weight in Hg in Hg of %
9-27 .34 .36 . 31 .404 .50 29.29
9-29 .25 . 33 .30 .404 .50 29.36
mean
-
x .295 . 345 .305
Std.
Dev.
s .0636 .0212 .00706
-
six % ~ 22 -;::::6 ~2%
TABLE 6
Emission test data of Vehicle No.2 (2M5733) tested at Ford, Dearborn
Source of Information: Ford Engrg. Progress Rpt.16 (September 1971)
Vehicle mileage 27,406, FCP engine mileage 3917
-------
Data set No.6
(Table 6)
2 tests between September 27 and 29 at Ford.
Method ology
Each of the six data sets was treated as a homogeneous sample
(although small engine changes did Occur during some test series as
evidenced by alterations recorded in Table 2). The hypothesis was then
tested whether all data sets could be considered samples of one and the
same grand population.
First, the means, x
, and the standard deviations, S , were
computed for each of the three exhaust emission gases and for each of the
six sets of data, Table 1 through 6. (The coefficient of variation, six
was added in order to indicate the data spread in relation to the mean. )
Since x and s are sample means and standard deviations, we first tested
whether all sample standard deviations of a particular exhaust emission
gas could be considered estimates of the same population standard devia~
tion.
Variance ratios of all possible pairs of the 6 x 3 data sets were
computed and compared with pertinent
99% significance level.
F
N;-r) NJ -,
values on a 95% and
N.
1
number of tests of data set i
N.
J
number of tests of data set j
i :f
The results are listed in Table 7, columns "standard deviation"
Except for one weakly significant standard deviation (CO data set No.3
versus No.5), all standard deviations could be considered estimates of
the san1e true population standard deviation, which was then estimated by
B-9
-------
computing the average values of all standard variations~' for the three
gases (CO data set No.5 omitted).
average standard deviation of HC ; shc = 0.0578 gr /mi
average standard deviation of CO . s = O. 1347 gr /mi
, co
average standard deviation of NO x;snox = 0.0476 gr /mi
We then tested whether all means of a particular gas could be
considered representing the same true population mean. For all pairs of
the 6 x 3 data sets, the value
t ::.
I.; : is I i
Ni Ni
Ni ... Ni
was computed and compared with the
confidence level
t /I.. /I. -1. - value on a 95% and 99%
. J
x.
1
mean of data set i
x.
]
mean of data set j
s
average standard deviation
Results
Table 7, columns "mean" lists the results.
They show that
with respect to HC, all data sets can be considered samples of the same
population except for data set No.2 and, perhaps, No.3. Data set No.2
is weakly (95% confidence but not 990/0) different from data sets No.1 and
3, very likely a consequence of the new catalyst installed after completion
of data set No. 1. The difference between data sets Nos. 3 and 5 is very
.'.
.,'
These standard deviations encompass both the variations of the engine
and those introduced by the test apparatus and the test procedure. A
quantitative separation of the two would require additional experimen-
ta ti on.
B-10
-------
..-< ......, Standard
..... U) .....
<1> ~ <1> Mean Deviation
U) U) U)
..... ;.., ..... (t-test) (F-test)
U) <1> U)
<1> > <1>
E-t E-t HC CO NOx HC CO NOx
1-2 ,~ *,~* - - - -
1-3 - *~,* ** - - -
1-4 - ~,** - - - -
1-5 - ~~** - - - -
1-6 - ** - - - -
2-3 ~~ *** - - - -
2-4 - *~~,~ * - - -
*
2-5 - ***
(**) - - -
2..6 - - - - - -
3-4 - - ** - - -
-
3-5 (*) - ** - * -
3-6 - ** * - - -
4-5 - - - - - -
4-6 - ,~* - - - -
5-6 - *~~ - - - -
B-ll
TABLE 7
Significance test of six sets
of data
Numbers 1 to 6 in first
column refer to data sets in
Tables 1 to 6, respectively.
*
Hypothesis of equality
rejected with 95% con-
fidence
**
Hypothesis of equality
rejected with 99%
confidence
(*) }
* borderline
(**)
cases
*** Hypothesis of equality
rejected with 99.9%
confidence
-------
weak (barely 95% confidence) and can possibly be neglected. Consequently,
Ford data set No.1, the EPA data set Nos. 3, 4, and 5, and the subse-
quent Ford data set No.6 can be pooled into one data set (1+3+4+5+6).
Another possible combination is(2+4+5+6).
With respect to CO, Table 7 reveals that data set No.1 is
distinctly different from the rest (99.9% confidence), most likely a con-
sequence of the faulty catalyst used by Ford in set No. 1. Also, data set
No.2 is distinctly different (99.9% confidence) from the EPA data set
Nos. 3, 4, and 5, apparently a consequence of the new catalyst installed
after completion of tests of data set No. 1. The EPA data set Nos. 3, 4,
and 5 show no differences among thenlselves and can be considered homo-
geneous. The last Ford data set No.6, however, deviates distinctly
(99% confidence) from the EPA data and conforms with the previous Ford
data set No.2.
In summary, ~ord data set No.1 has to be considered as
an entity by itself; the EPA data sets can be pooled (3+4+5), and the Ford
data set Nos. 2 and 4 can be pooled also. With respect to NOx, we notice a
distinct difference (99% confidence) between data set Nos. 1 and 3, perhaps
due to maladjusted EGR linkage. The adjustment incurred a significant
change (99% confidence) in NOx content as evidenced by the significant
differences between EPA data set Nos. 3-4 and Nos. 3-5.
The data set
Nos. 4, 5, and 6 are homogeneous, proving that the EGR linkage adjust-
ment remained effective throughout the rest of the tests.
In order to display the results of this statistical analysis in a
more effective way than is possible in a tabulation, we plotted for each of
the three gases (HC, CO, and NOx) and for each of the six data sets the
means and their confidence limits, Tp, computed using the relation
p = s x tn/ ,.r--n , where n 5
for two combined tests) and t =
( ::: average number of degrees of freedom
2.5 (95% confidence for n =:
5 ).
The data
are shown in Figures 1, 2, and 3. The figures illustrate clearly the
impact of the three known engine changes - the replacement of a fa ulty
catalyst by a new one after completion of test set No.1, the adjustment
of the accelerator linkage to prevent closing of EGR at WOT after test
B-12
-------
set No.3, and the adjustment of the deceleration fuel cutoff after test set
No.4. The significance of these changes is evidenced by the difference
of the means and their associated confidence limits. As long as the difference
of two means is small so that their confidence ranges overlap, it cannot
be considered significant.
This situation, for instance, is true for the
difference caused by the adjustment of deceleration fuel cutoff in Figure 1;
the HC output of data set No.6 is not significantly different from that of
data set No.5. On the other hand, the replacement of the faulty catalyst
after data set No.1 caused a significant reduction of both HC and CO,
Figures 1 and 2, whereas the NOx output remained unaffected, Figure 3.
We also notice the significant reduction of NOx after adjustment of the EGR
system, Figure 3 data set Nos. 3 and 4. Figures 1 and 2 reveal also that
the drastic reduction of HC and especially of CO emissions after installment
of the new catalyst could not be maintained on the newly achieved low level.
Data set No.3 shows that HC emissions shifted back to higher levels,
although they could be again reduced in subsequent tests.
To carry the investigation somewhat further, we pooled all
data sets that showed no significant differences on a 950;" confidence level,
as indicated in Figures 1, 2, and 3, and computed their means. One could
consider these means as characteristic emission levels of test vehicle No.
2.
average HC level of data sets
(1 + 3 + 4 + 5 + 6)
0.371 gr/mi
(with faulty
catalyst)
0.323 gr/mi
(without faulty
c a ta 1 y s t )
U. S24 gr /mi
O. 29S gr/mi
average HC level of data sets (2 + 4 + 5 + 6)
average CO level of data sets (3 + 4 + 5)
average HOx level of data sets (1 + 4 + 5 + 6)
B-l3
-------
.It r t~U1~Y ~~.di:" 08'81y"
~ ~
.
" '"
I "'
..
!f ~
t- \ . homo2eneous data
)to I
He \!~ '\
tr/fn1
.3~ \ ,: !
'
.10
new platinum
catalyst
,
v
,
"""-
deceleration
fuel cutoff
adjus ted
~
'"
.10
o
2.
J
It
5
,
data set No.
l"igure 1
HC means and confidence ranges for six data sets
Data sets No.1, 2, 6
Data sets No.3, 4, 5
Ford dat a
EPA data
B-14
-------
J.D
1.0
CO
gr/mi
tD
o
Figure 2
,faulty
f
catalyst
-
~tomogeneOU8 data
-
....
\
.: f f (
J
~
~
....
"?r- --~
'\
'-
,
-
- .... . ...
"
....
..........--
t
J
.. "I
--~
I
2.
~
S'
b
If-
data set Noo
CO means and confidence ranges for six data sets
Data sets No. I, 2, 6
Ford data
Data sets No.3, 4, 5
EPA data
B-15
-------
.50
.4Q
NO
x
gr/mi
F:i.gure 3
J,
, .
.
.30 .
\
\
"
.20
. If>
o
~
....
r..
~
o
=
~
cg
~
,
"
,
,
" - -
-
- -
-
,
rmogeneous -data
j --
,
"
,
- -
o
..
3.
5
.3
data set Noo
It
,
\
)
.,
I
,
,
'"
EGR cutoff at WOT
full EGR
,
NO ITearlS and confidence ranges for six data sets
x
Data sets No.1, 2, 6
Data sets No.3, 4, 5
B-16
rord data
EPA data
-------
These levels combined with the earlier estimated standard
deviations result in the following coefficients of variation:
HC coefficient of variation
HC coefficient of variation
= (.0578/.371)100 : 16% (based on data sets
1 + 3 + 4 + 5 + 6)
= (.0578/.328)100 - 17.50/0 (based on data sets
2 + 4 + 5 + 6)
CO coefficient of variation
(.1347/.824)100 - 16 % (based on data sets
3 + 4 + 5)
NOx coefficient of variation = (.0476/.298)100 '" 16% (based on data sets
1 + 4 + 5 + 6)
The coefficient of variation for all three emission gases is approximately
the same, i. e., 16%.
In order to answer the question of whether the emission
standards of 1976 are achieved we computed the percentage of data lower
than the standards. Table 8 shows that 75% of the combined data sets
(1 + 3 + 4 + 5 + 6) are lower than the HC standard of 0.41 gr/mi.. If one
excludes the runs with faulty catalyst and combines only data sets (2 + 4
+ 5 + 6), then 92% of all data meet the HC standard. Whether these data
demonstrate full compliance cannot be decided with confidence; a higher
percentage would surely be desirable. Table 8 shows also that compliance
with the CO-standard of 3.4 gr/mi is well achieved, and that compliance
with the NOx standard of 0.40 gr/mi is within 98%.
We like to point out here that the given percentages are com-
puted from the mean values and the standard deviations. Clearly, if the
standard deviation of the HC data would have been lower, the HC standard
could have been met.
The standard deviation, as computed here,
en c om -
passes both the variations of the engine and those of the test apparatus
and the test procedure including the operations of the test driver. If the
variations of the test procedure and the test equipment playa signi-
ficant part in the overall variation, a guess we cannot substantiate without
B-17
-------
1976
Exhaust I\ a tional Combined Nurnber Mean Standard Percentage of
gas Standa rd Data of Value Deviation data below
Components g r / mi Sets Runs g r / mi gr/mi 1976 Standard
0.41 1+3+4+5+6 20 o. 37 0.058 75%
HC
2 +4+5+6 16 0.33 92%
CO 3.4 3+4+5 1 3 0.82 0.135 100%
T\Ox 0.40 1+4+5+6 20 0.30 0.048 98%
b:J
I
......
TABLE 8
Percentage of measured data below 1976 national standards
00
Data Set FORD - fa uHy cataly st
2 FORD - new catalyst
3 EPA - maladjusted EGR
4 EPA - adjusted EGR
5 EPA - adjusted accelerator linkage
6 FORD
-------
further experimentation, and if one could succeed in lowering these varia-
tions by using more accurate equipment or techniques, then compliance
with the HC standard would not be unlikely.
Summary
A M-15l (jeep) vehicle equipped with a low-emission stratified-
charge Ford engine (FCP) was emission-tested at both the Ford Company and
EPA using a chassis dynamometer in combination with the 1976 CV5-CH testing
and sampling technique.
A total of 25 test runs structured into 6 test series
were made with 2 to 5 runs per test series.
With the exceptions of a correction
of the EGR system and of the accelerator linkage, and the replacement of a
faulty catalyst, no substantial changes were made during tests.
An analysis of the measured data revealed the following:
The coefficient of variation of all three exhaust gas components
is the same, i. e.. 16 to 17 per cent.
The replacement of the faulty PTX-6 palladium catalyst by a
new PTX-6 platinum catalyst reduced HC and, especially, CO
emission substantially.
Elimination of EGR cutoff during accelerations reduced NOx
emissions substantially.
Adjustment of the deceleration fuel cutoff linkage had little
effect on emissions.
For each gas component, a number of homogeneous data were
selected as representative of the emission performance of the
tested engine-vehicle combination. Of these data sets, 100%
achieved compliance with the 1976 CO standard, 98% with the
NOx standard, and 92% with the HC standard if the runs with
faulty catalyst were excluded.
If they were included, then only
75% of all HC data achieved compliance with the HC standard.
B-19
-------
REFERENCES
1
Ford Motor Comp, Special Military Vehicle Operations, Eng.
Progress Rpt. No. 14 Project: Ford Combustion Process,
July 1971
2
Ford Motor Comp, Special Military Vehicle Operations, Eng.
Progress Rpt. No. 15 Project: Ford Combustion Process,
August 1971
3
Ford Motor Comp, Special Military Vehicles Operation, Eng.
Progress Rpt. No. 16
Project: Ford Combustion Process,
September 1971
4
Computer printout of EPA, private communication
B-20
-------
APPENDIX C - REGRESSION ANALYSIS OF EMISSIONS DATA
H. T. McAdams
1.
Regression Analysis of Emissions Tests of a Stratified Charge Engine
A number of variables can affect emissions from a stratified charge
en~ine. These include ambient environmental factors, such as barometric ~ressure,
specific humidity and temperature, as well as such other variables as dynamometer
inertia and road load horsepower. To estimate the effect of these variables on
emissions, re~ression analysis was applied to 14 tests conducted on an FCP enRine
installed in a JEEP. The results of these tests, to~ether with the associated
levels of the ambient input variables, are tabulated in Table 1.
TABLE I
ANALYSIS OF 14 TESTS 0 F A STRATI FIED-Q-IARGE ENGINE
Xl X2 X3 X4 X5 Y1 Y2 Y3
Bar. Press Spec.Hum. Temp. Road Load
(in.Hg.) (Lbs H20/ (oF) Inertia Horsepower Hydrocarbons CO NOX
Lbs.Dry Air) (lbs) (II P) ( Gms /ML) (Gms/ML) (Gms ./Mi)
29 . 3600 0.0105 81.0000 2750.0000 8.8000 0.4100 0.9800 0.4500
29.4600 0.0095 72 .0000 2750.0000 8.5000 0.3700 0.7200 0 . 3900
29.4200. 0.0115 76.0000 2750.0000 8.5000 0.4600 0.8100 0.4100
29.3800 0.0140 76.0000 2750.0000 8.5000 0.4600 0.8200 0.3700
29.3700 0.0143 75.0000 2750.0000 8.5000 0.3300 0.8100 0.4700
29.3200 0.0133 74.0000 2750.0000 8.5000 0.3700 0.9200 0.2100
29.2500 0.0147 81.0000 2750.0000 8.5000 0.4700 1. 1700 0.3400
29.2500 0.0148 '83.0000 3000.0000 8.8000 0.3900 0.9200 0.3200
29.3000 0.0125 76.0000 3000.0000 8.8000 0.3200 0.8300 0.3100
29.2800 0.0114 81.0000 3000.0000 8.8000 0.2700 0.5800 0.2800
29.1800 0.0116 74.0000 3000.0000 8.8000 0.3000 0 . 5500 0.2900
29.0300 0.0101 70.0000 3000.0000 8.8000 0.3200 0 . 7800 0 . 2600
,29.0000 0.0097 72.0000 3000.0000 8.8000 0.3100 1.2000 0.2500
28.9900 0.0099 70.0000 3000.0000 8.8000 0.4000 1 .2400 0.3500
As an approach to studying the effect of ambient conditions on ellbdons, it
can be supposed that the emissions Y, whether hydrocarbons (Y 1)' 00 (Y 2)' or NOX (Y 3)
can be expressed as a function of the variables Xl' X2' X3' X4' and Xs'
Y = f ( Xl' X2' X3' X4' Xs)
C-l
-------
The simplest relationship which can be hypothesized is a linear function
Y::bO+blXl
+ b2X2
+ b 3X 3
+ b 4X4
+
b5X5
where bO' bl' b2' b3' b4 and b5 are constants. Such a re~ression equation
can be fitted to the data of Table I by the method of least squares and its
statistical si~ificance tested by analysis of variance.
The re~ression equations computed from the data of Table I are Riven below.
Hydrocarbons:
Yl
::
8.93979 - 0.23550 Xl
- 0.94665 X2
+ 0.00590 X3
- 0.00038 X4 - 0.11858 X5
CO :
Y2
::
56.83229 - 1.78644 Xl- 5.20022 X2
+ 0.01833 X3 - 0.00150 X4 - 0.07912 X5
NOX:
Y3
z
6.45692
+ 0.18473 Xl
+ 2.83481 X2
- 0.00263 X3 - 0.00054 X4 + 0.35757 X5
For each of the regression equations, the inputted or observed values of
emission were compared with the correspondinR values computed from the reRTession
equation. For example, see Table II, in which the observed and computed values
of hydrocarbons are tabulated.
C-2
-------
TABLE II
COMPARISON OF OBSERVED AND CALCULATED HYDROCARBON EMISSIONS
Yl
Calculated
Yl
Observed
0.41000 0.40983
0.37000 0.36969
0.46000 0.40082
Q.46000 0.40788
0.33000 0.40404
CI.37000 0.41087
0.47000 0.46733
0.39000 0.34889
Q.32000 0.29799
0.27000 0.33324
o . 30000 0.31530
0.12000 0.32843
O. :31'000 0.34768
().40ooo 0.33805
C-3
-------
By computin~ the difference between the observed and calculated
quantities for each test, squaring these differences and summing over all
tests, one obtains a measure of the residual scatter unaccounted for by a
linear regression equation. This residual sum of squares can be considered
as a measure of error, on the assumption that the regression equation, as
formulated, extracts all systematic (non-random) trends in the data. An
alternative assumption is that there is no relation between the observed
emissions and the values of the ambient variables. Accordingly, one can
compute the average value of the emissions for the 14 tests and compare the
14 observed emissions with this single average value. If the deviations
from the average value are computed, squared and summed, one obtains a
residual sum of squares which remains when only a constant (the mean) is
removed from the data. This residual sum of squares can not be smaller than
the sum of squares of residuals obtained in the complete regresS10n equat10n.
The difference in magnitude of these two residual sums of squares is a
measure of the variability accounted for by adjustment for linear trends with
barometric pressure, specific humidity, temperature, load and horsepower.
Whether this difference is large enough to be considered statistically
significant can be adjudged by analysis-of-variance techniques applied to
the regression results.
Tables III, IV and V summarize the results of analysis of variance for
He, eo and NOX respectively. In each table there is displayed the sum of
squares due to bo' bl' b2' b3' b4 and b5 and the sum of squares due to the
single constant b~, the mean for all the data.. The salient feature of
each of these tables is the F-ratio displayed in the right-most column. In
order to argue that the effect of the input variables is statistically
significant at the 5% level of significance the tabulated F-ratio should exceed
the critical value
5
F (0.05) = 3.69
8
* Note that bOt is different from b in the regression equation. Though the
two quantities playa similar role, ~hat of a constant in an empirical equation,
they will be equal only if the values of the variables Xl' X2' X3' X4 and X5
constitute an orthogonal set of vectors in the matrix formulation of the
least-squares normal equations.
G-4
-------
TABLE III
ANALYSIS OF VARIANCE FOR REGRESSION OF HC EMISSIONS FOR A
STRATIFIED CHARGE ENGINE
De ~rees
of Freedom
F
Rat io
Source of
Variance
Sum of
Squares
Mean
Squares
Total 14 1.971
Due to
bO,b1 ,b2 ,b 3'b 4 ,b5 6 1.946
D.ue to bO
1
1.917
Due to
b1 ,b2,b3,b 4,b5
.029
.0058
5
Residual
8
.025
.0031
5
Fe (.95) 4 3.69
C -5
1.856
-------
ANALYSIS OF VARIANCE R)R REGRESSION ANALYSIS OF CO EMISSIONS
FOR A STRATIFIED CHARGE ENGINE
Source of
Variance
Sum of
Squares
Mean
Squares
Degrees
of Freedom
F
Ratio
Total
14
11 .488
Due to
bo,bl,b2,b3,b4,bs
6
11 .246
I
Due to bo
1
10.912
Due to
bl,b2,b3,b4,bs
.334
.0668
S
Residual
.242
g
.0302
S
Fg (.9S) = 3.69
C -6
2.208
-------
TABLE V
ANALYSIS OF VARIANCE FOR REGRESSION ANALYSIS OF NOX EMISSIONS
FOR A STRATIFIED CHARGE ENGINE
Source of
Variance
Sum of
Squares
Degrees
of Freedom
Mean
Squares
F
Ratio
Total
14
1.654
Due to
bO,bl,b2,b3,b4,h5
6
1.612
I
Due to bO
1.578
Due to
bl' b2,b3,b4,b5
5
.034
.0068
Residual
8
.042
.0053
5
F 8 (.95) = 3.69
C -7
1.295
-------
Values of F smaller than 3.69 could occur relatively often, even if there were
no effect of the input variables, simply as the result of random errors.
A value as large or lar~er than 3.69, however, would occur by chance no more
than 5\ of the time. Therefore, if the critical value is exceeded by the F-
ratio computed from the data, it can be ,rgued that the computed re~ression
equation is not the result of chance but reflects an actual relationship
between the dependent and the independent variables. It will be noted that for
neither HC, CO nor NOX did the computed F-ratio exceed the critical value.
Therefore, it must be concluded that the data from the 14 tests do not establish
any linear dependence of emissions on atmospheric pressure, specific humidity.
temperature, inertia and horsepower. Since the combined effects of these five
variables failed to reach the desired significance level, it can further be
concluded that any sin~le variable would also fail to show a significant
influence on emissions.
C-8
-------
2.
Regression Analysis of Emissions Tests of Conventional En~ines
Emissions tests are available on 54 automobiles representing various
engine sizes, vehicle weights and mileages. It was desired to determine if emission~
are related to these factors, as judged by regression analyses. For purposes of
the analysis, the engines were segregated into two groups, 8-cylinder and 6-cylinder.
Because of some questionable aspects of the data for one of the vehicles, only 53
of the 54 items were actually analyzed.
Input data for the 8-cylinder engines, 45 in number, are given in Table VI.
Regression equations for these data took the form
y =
bO
+ blXl
+ b2X2
+ b 3X3
where
x -
1
X =
2
X =
3
mileage (mi.)
weip:ht (lbs)
displacement (cu. in.)
and Y is either HC, CO or NOX in grams/mile.
The regression equations as derived from the data of Table VI are given
below.
HC:
Yl
1.05587 - 0.00000 Xl
- 0.00004 X2
+ 0.00008 X3
co:
Y2
5.27079
+ 0.00002 Xl
- 0.00106 X2
+ 0.01650 X3
NOX :
Y3
0.49340
- 0.00000 Xl +
0.00025 X
2
0.00045 X3
Analysis of variance for these three equations is given in Tables VII, VIII, and
IX respectively.
C-9
-------
TABLE VI
EMISSIONS DATA FOR 8-CYLINDER ENGINES
Xl X2 X3 Yl Y2 Y3
Mileage Weight Displacement HC m NOX
(Mi.) (lbs) (cu. in.) (gms/mi.) (gms/mi.) (gms/mi.)
14173 5000 475 0 .9800 8.4400 1.0600
23801 4500 475 0 . 79 30 6.6300 1.5300
24930 4500 360 0.8250 3.9200 1 . 1300
15427 3500 360 1.0200 8 . II 00 1.0300
15598 3500 360 1.2200 11.8700 1.1400
18923 4000 410 0.6670 8.1400 0.8200
13625 4000 360 0.7960 4.7600 1.1400
8039 4000 360 0.7800 6.4800 1 . 3100
22772 3500 290 0.7840 7.7400 1 . 3200
23957 4500 410 0.7210 2.0500 1.5500
9803 4000 360 0.8790 4.9100 1. 5400
13664 3500 290 1. 2300 5.0500 1.2800
20760 4000 360 0.7500 12.0400 1 . 1800
22188 4000 410 0.9930 11 .1700 1. 5800
14700 3500 290 0.6840 3.1900 1.0200
31888 5000 475 0.7310 5.9800 2.5000
9943 4000 360 0.8740 9.7300 1.5500
9797 3500 360 1. 0600 9.7400 0.8630
30064 5000 410 O. 7900 5.7000 1.5200
20035 4500 475 0.7960 6.0700 1.6700
17756 4000 360 0.8010 5.2200 1.2700
11 089 5000 410 1 .3200 13.2500 2.4800
29118 4500 475 1.0900 16.9900 0.5910
24963 4500 475 0.6840 8.5500 1 . 1300
18686 3000 360 1.0500 9.8800 1.2300
27205 4000 360 0.8920 8.6700 1.0500
28565 4500 475 0 . 8580 8.8700 1.2500
6078 40aO 410 0.7340 4.2100 1.4800
10229 4000 360 0.7510 5.6700 1.7800
11312 5000 410 0.7750 4.9300 1 .1600
32840 5500 475 0.8230 10.4700 1.2800
13866 3500 290 0.8520 9.6500 0.9570
15485 5500 475 0.5730 10.0200 0.7420
9351 3500 360 0.8880 11 .8900 1.0500
20831 4000 410 0.8320 8 . 1600 1.2000
20204 4000 360 0.8610 5.9600 1.1800
20037 4500 290 0.8910 7.4900 1.3700
31419 4500 475 0.6910 8 . 71 00 1.3700
13032 4500 475 0.8680 5.7300 0.9720
20206 3500 290 0.7820 4.4500 0.9520
8624 4000 360 0 . 8240 7.8600 1.1400
8991 4500 410 0.9130 7.4500 1 .7900
16698 3500 360 0.5910 2.8600 0.9570
34397 4500 290 0.7570 6. 1200 1.2700
30749 5000 475 0.8370 5.8300 1.4700
. Displacement is the median value of a class interval, the engines being
identi fied only according to a range of displacements.
C -10
-------
TABLE VI I
ANALYSIS OF VARIANCE FOR REGRESSION ANALYSIS OF HC EMISSIONS
FOR 8-CYLINDER ENGINES
Sou~ce of
Variance
Sum of
Squares
De~rees
of Freedom
Mean
Squares
F
Rat io
Total
45
33.704
Due to
bO' b1' b2' b3
4
32.677
,
Due to bO
1
32.616
Due to
b1' b2' b3
3
.061
.020
Residual
41
1.027
.025
F3
41
(.05) = 2.83
C-ll
0.800
-------
TABLE VI II
ANALYSIS OF VARIANCE FOR REGRESSION ANALYSIS OF CO EMISSIONS
FOR 8-CYLINDER ENGINES
Source of
Variance
Sum of
Squares
Degrees
of Freedom
Mean
Squares
F
Ratio
To tal
45
297,1.730
Due to
bO,bl' b2' b3
4
2605.221
Due to bO
1
2578.112
Due to
bl,b~,b3
3
27 . 109
Residual
41
369.509
F3
47
C.05) = 2.83
C -12
9.036
9.012
1.003
-------
TABLE IX
ANALYSIS OF VARIANCE FOR REGRESSION ANALYSIS OF NOX EMISSIONS
FOR 8-CYLINDER ENGINES
Source of
Variance
Degrees
of Freedom
Sum of
Squares
Mean
Squares
F
Ratio
Total
45
80.480
Due to
bO' b1,b2,b3
4
75.008
Due to bo 1 74.379
3 .629 .210 1.57
Due to
b1 ,b2,b3
Res idua1 41 5.472 .133
F3
41
(.05) = 2.83
C -13
-------
It is seen, by reference to these tables, that statistical si~ificance
at the 0.05 level is not achieved by either He, co or NOX even when the
emissions are adjusted for all three variables Xl' X2 and X3' Therefore, it
must be concluded that no one of these variables would show a satistically
significant effect in the cases analyzed.
A word of explanation is in order concerning the displacement variable, X3'
Exact displacement for each engine was unknown. Rather, each vehicle was
assigned to a class, as shown below.
Range of Engine Displacement (cu. in.)
Class
1
2
3
4
5
Less than 260
261 to 320
321 to 380
381 to 439
Greater than 440
In applying the regression analysis, mean or representative values were assumed
for the five classes. These were 290, 360, 410, and 475 for classes 2, 3, 4 and 5,
respectively. No assumption was necessary for Class I, since this included only
six-cylinder engines, which were handled separately from ei~ht-cy1inder engines.
By coding the displacement values in this way, it was possible to avoid
identifying actual makes or manufacturers of the vehicles tested.
A separate analysis was performed for the 6-cylinder engines,
eight in number. The input data for these engines is tabulated in Table X.
Inasmuch as all of the 6-cylinder en~ines belonged to the same displacement
class, displacement was not a variable for these data. Therefore, the emissions
~ere expressed as functions of only XI' mileage and X2' weight.
C-l4
-------
TABLE X
EMISSIONS FOR 6-CYL INDE R ENGINES
X X HO CO NOX
mil ea~e . 2. (gms/mi.) (gms/ mi.) (gms/mi .)
lVei r:h t
18618 3500 0.524 4.18 1.72
12320 3500 0.861 16.51 1. 21
12769 3000 0.478 3.81 1.13
4106 2500 0.533 3.95 1.04
4395 3500 0.437 4.79 1.60
10923 2500 0.731 3.89 1. 83
16183 3000 0.665 3.90 0.72
1 5124 3500 0.926 13.66 2.07
The regression equations, as derived from the data of Table X,
follow.
Hydrocarbons:
Yl
0.44245
+ O. 00001 Xl
... 0.00002 X2
co:
Y2
=-13.23511
- 0.00002 Xl
+
0.00652 X2
t-K) X :
Y -
3 -
0.39274
+ 0.00000 Xl
+ 0.00033 X2
Analysis of variance for these three equations are given in Tables XI, XII
and XIII, respectively.
It is seen, by reference to these tables, that statistical significance
at the 5% level is not achieved.
In summary, it is concluded that no statistical significance is
indicated for the regression equations computed in this Appendix. The effects of
the variables in question either are non-existent or are too small, relative
to error magnitudes, to be detectable within the limited sample of available
data.
C -15
-------
TABLE XI
ANALYSIS OF VARIANCE FOR REGRESSION ANALYSIS OF HC EMISSIONS
FOR 6-CYLINDER ENGINES
Source of
Variance
Sum of
Squares
De~rees
of Freedom
Mean
Squares
F
Ratio
Total 8 3.554
Due to 3 3.354
bo' bl,b2
I
Due to bo
I
3.322
Due to
bl,b2
2
.032
.016
Residual
5
.200
.040
F2 (.05) = 5.79
5
C-16
0.400
-------
TABLE XI I
ANALYSIS OF VARIANCE FOR REGRESSION ANALYSIS OF CO EMISSIONS
FOR 6-CYLINDER ENGINES
Source of
Variance
Degrees
of Freedom
Sum of
Squares
Mean
Squares
F
Itatio
Total
8
560.052
Due to 3 4 30 . 4 72
bO,b1 ,b2
I 1 373.874
Due to bO
Due to
b1,b2
2
56.598
28.299
1.092
Residual
5
129.580
25.916
2
F5 (.05) = 5.79
C-17
-------
TABLE XIII
ANALYSIS OF VARIANCE FOR REGRESSION ANALYSIS OF NOX EMISSIONS
FOR 6-CYLINDER ENGINES
Source of
Variance
Sum of
Squares
Mean
Squares
Degrees
of Freedom
F
Ratio
Total 8 17.493
Due to 3 16. 165
bO,b1 ,b2
Due to bb 16.018
Due to
b1,b2
2
.147
.074
Residual
5
1.328
.266
F2
5
(.05)
=5.79
C-18
0.278
-------
APPENDIX D
ANAL YSIS OF EMISSION TEST DATA OF VARIOUS
GROUPS OF AUTOMOTIVE VEHICLES
D. J Schuring
INTRODUC TION
All experimental data are subject to fluctuations ("errors" or "noise")
caused by erratic, usually small, changes of the experimental equipment
during measurements.
In emission testing, the fluctuations are often large,
that is, not much smaller than the differences ("signals") we wish to study,
so that statistical techniques must be employed to separate the signals from
the nois e.
Emission fluctuations originate from many sources -- fuel injection,
spark plugs, air temperature, humidity, air-fuel mixture, compression
ratio, combustion process, valves, linkages, collecting and diagnostic
equipment, and the human operator -- to name a few.
The fluctuations
issuing from these and many other sources are by necessity confounded;
they can hardly be separated except for the hypothetical case of all but one
of the errors being negligibly small.
The fluctuations of an engine (taken
as a unit source of error), for instance, could be isolated only if the error
of the measuring equipment would be negligibly small, if each measurement
would follow exactly the same procedure, and if the initial engine and vehicle
conditions before each repetition, and the temperature, humidity, and
pressure of the air, as well as all other external variables, would remain
constant.
Only under thes e (hypothetical) conditions would the tested engine
reveal its inherent noise level.
In reality, of course, irregularities of the measuring equipment,
deviations from the prescribed test procedure, variations of the atmospheric
state, and changes of the initial vehicle conditions are inevitable,
H enc e,
each repetition yields a different result.
The fluc tuations of oba ervations
performed repeatedly on the same car with the same equipment under
D-l
-------
identically c0ntrolled test conditions will be denoted by the term "test
error".
The test error masks the emission differences we are interested in ~-
differences between cars at the end of the production line, between perform~
ances of the same car at various stages of its life, between cars of different
make, between cars with and without emission control, and the like.
The
question then arises of how many cars have to be tested in order to establish
the desired information with confidence.
The following study attempts to
shed some light on this problem.
It tries a first cut at the relation between
the test error, which is present in all measurements, and the emission
differences between ca,rs.
Unfortunately, the available pertinent information
is far from complete.
The data used here are gleaned from some EPA
reports, some company presentations, and from communications of auto~
mobile manufacturers.
Since the latter were confidential, they could not
be fully utilized. Furthermore, the size of available emis sion samples was
usually small ~~ frequently not larger than five to ten observations per
sample.
Under thes e circumstanc es, lacking statistical evidenc e had to be
filled in by engineering judgment.
In addition, statistical data for very low
emission engines are practically non~existent because these engines are
still under development.
Notwithstanding all these restrictions, we felt
that any, however preliminary, information on the fluctuations of emission
measurements would be helpful in planning tests for the stratified-combustion~
process engines.
Hence, this study.
D~2
-------
EMISSION DISTRIB UTIONS
The test error, that is, the distribution of all observations performed
repeatedly on the same car with the same equipment under identically
controlled test conditions, can be characterized statistically in various ways.
Since emission distributions are usually not normal, a characterization by
the mean and the standard deviation is often not sufficient.
A more complete
picture e\'oh'es from the cumulative relative frequency distribution:
are grouped into equidistant classes and their relative frequency, i. e., the
data
relative number" of data in each class, is associated with the respective
upper class level.
Plotted on probability paper (on which the ordinate scale
is graduated according to the area under a normal distribution), the cumu-
lative data, if essentially normally distributed, approximate a straight line.
The mean value appears then at the cumulative relative frequency of 50%,
the standard deviation as difference of the emission values at the frequencies
of 8~r, and 50":, (or 160"'0 and 50%).
Emission data are not normally distributed, however, as indicated
before.
They are skewed because emissions cannot surpass a certain mini-
rnlLD1 value (ultimately, the zero vi11ue).
If the mean value of an emis sion
distribution is large (as is true for an uncontrolled engine), the skew may
be slight and can perhaps be neglected.
Then, a normal distribution would
fit the data best.
If the mean value is clos e to the zero point, however, as
realized for an emission-controlled engine, the emission data are compressed
towards smaller values, and expanded towards larger values.
Skewed
distributions of this kind can often be turned into normal distributions by
cOI1'.'erting the measured data into their log values before processing them
statistically.
The transformation is most easily achieved by plotting the
relative accumulative frequencies on probability paper with a logarithmically
D-3
-------
graduated abscissa scale.
If the log values of the emission data are normally
distributed, they will approximate a straight line on this paper.
The
emission value at the cumulative frequency of 50% signifies the median rather
than the mean (which is somewhat larger than the median).
Also, the standard
deviation cannot be directly obtained from this log plot.
Instead, the percentage
of emission data smaller (or larger) than any given emission value can be
predicted with eas e.
As an example, we present HC and CO emission data of a large number
of 1970 General Motors cars.
The cars had been in actual customer use and
were tested by the California Air Resources Board using the FTP hot-cycle
proc edure.
Figures 1 and 2 show the data as pres ented by GM.
We subdivided
the data into 12 clas s es (each O. 50 gr / mi wide), counted the number of
observations within each class, computed their relative accumulated frequencies,
plotted them at the upper level of each class on log-probability paper, as
shown in Figure 3 for HC and Figure 4 for CO, and faired each plot by a
straight line signifying a logarithmic distribution of emission data.
The
data at the ends are somewhat erratic, which must be attributed to the small
number of observations at high emission levels, see Figures I and 2.
Sine e thes e and other data (s ee Figures 5, 6, 7) indicate strongly
that emissions data generally conform to the normal log-distribution, we
plotted all emission distributions on log-probability paper and faired them
by straight lines even if the scatter was large.
The scatter was always
attributed to the small number 'f observations rather than to a significant
deviation from the log-distribuc ~on.
We would like to note that the emissicm
data in Figures 1 and 3 were obtained from many vehicles instead of just
one;
they exhibit the combined effect of the test error and the variance
D-4
-------
1.00
6.00
5.00
He
4.00
gr (IY\;
3.00
.
.
.
----.- -- - -
N L 1/ ~'/ ('0
---- +-.
157 I DO
I
. :
.
. .
-.----...---- ._--~---- . ---------
. .
.
.
.
.
..
.
.
. .
. .. e
- .\--- 0'- .
"o' .
2.00: ':. .:.. .'
--:- . ~.
. .
1.00: .:: .
.
. .
o
o
Figure 1 -
--- -.- ---
-----~_.
-------_..____n____"-- -- -. -.._-----r----
.
"------- -- -...--
.
.
. .
. .
.
. i'- -
--'------'-~---- -~.
.
.
.
- - .
..
. ..
.....- . .
-....--- -- - .--.-.'--
. .
. . . .
.. .
..........'----
..
.
.
.
.
.
.
5,000
.
. __n______------"--- -.-. --.
.
-.---
.
.
.
4
----~_. --
/~ & qq if
----
Ie 2 q~. t
118 Cf If, 2-
! If (, Q3.0
lLi( ere) 0
{~I S~.;
1% 67
)8 T;'
36 27
~ :.7
o 0
- -- ---t-- - --
o 15,000
MILES
o
20,000
!f
2
(;
IC'
2S
~[;
42
27
q
HC Emissions of GM Cars, Model 1970, Tested by
California Air Resources Board (FTP Hot-Cycle)
Adapted from Reference 1
D-5
-------
70.00 .
.
60.00 .
.
50.00 . .
.
.
CO .
.
~rf~; 40.00 ~ .
. . .
I o. ".
30.00 .
. ' '.
: I .
. I. . . .
0 iI" 01 . -. .
..
. ' .'
. .
. .
I ..
'0 0 ' . .
0 . .
." .
,.. . . . . .
.
I " I I
0 "0 . .
" . .
I I ". ..
.
,..' " I . , I .
.-..- .' ' " .
. ... . ' .
0 5,000 10,000 15,000 20,000
MilES
Figure 2 -
CO Emissions of GM Cars, Model 1970, Tested by
California Air Resources Board (FTP Hot-Cycle)
Adapted from Reference 1
D-6
-------
«ft.'
10
-------
q,.q
ro
~
ew8
'QS
'0
,
/
80
70
50
30
.,
'9
S-
: '0
l.a
,
~o
(0
qy (Wli
,(
~ '
,
50
/
"
/
" .
"
80
Figure 4 - Cumulative Relative Frequency Distribution of CO Emissions
of GM Cars, Model 1970, Tested by California Air Resources
Board (FTP Hot-Cycle) - Reference 1
D-8
-------
between vehicles.
Unfortunately, no single vehicle measurements with a
large number of repetitions were available.
We will assume, however, that
all emission data, whether obtained from one engine or a large number of
engines, are distributed logarithmically.
In the following,a number of emission measurements gleaned from
.',
papers, company reports, EPA evaluations, and NRC"presentations are
plotted on log-probability paper.
In instanc es where only the mean and the
standard deviation were available instead of the raw data, the following
transformation was made.
If the mean, X, and the standard deviation, s, are given of a set of
observations, x" whose log-distribution we know (or suspect) to be a
1
normal one, we can compute three characteristical values: the median, x50'
(which is defined as the upper bound of 50% of all observations), the
value x84' bounding 84% of all observations, and the value x16' 16%. The
formulas for computing thes e three values from x and s are given below.
The derivation is presented in Appendix G.
-1-
-X
median
Xso =
v:xl-
~
).
.s
",,,,;:'0:::
84% limit""
")(,1+ = X.SrJ
tX ~ VL... [ ( ;) \ ! J
16 % limit
X,. r
Xljo
I.
XSo
t~~ I i [( :) \ (1
):('i
.',
','
National Research Council - National Academy of Engineering -
Committee on Motor Vehicle Emissions
or
€)(P r ~ e
..I~ ..f...
',"',"
D-9
-------
Thes e three data fix three points on log-probability paper at 50%, 84%,
and 16% of the cumulative relative distribution scale.
A (straight) line
through these points establishes the log-distribution, from which any data
of interest can be determined easily.
D-10
-------
DATA
(1)
In a Progress Report to EPA~l) the General Motors Corporation
published HC emissions and CO emissions of a large number of 1970
GM cars tested by the California Air Resources Board using the
FTP hot-cycle test procedure.
In this report the emission data were
displayed as function of mileage in two graphs, Figures 1 and Z.
The
scatter is very large; hence, an
influenc e of mileage on emis s ion is
unlikely, and we considered all data as drawn from a homogeneous
population of cars.
The number of observations is approximately 160.
A visual inspection
of the distribution of data points suggests that repetitive measurements
had not been made so that each point was assumed to represent the
emission of one car measured only one time at given mileage.
A log-distribution plot of all CO and HC data is presented in Figures
3 and 4.
(2)
Fifty-four fleet cars of different makes and mileage were tested by
EPA(Z) using the CVS-CH cycle (41 minutes). The identity of the
vehicles was not disclosed.
An analysis of variance (discussed else-
where) did not disprove the assumption of all cars representing a
sample of a homogeneous population.
Hence, all HC, CO, and NOx
data were lumped together and their distribution plotted, Figures 5,
6, and 7.
D-ll
-------
It If. If
'0
"
U
qs
'0 - .
/
8i> .
/
70
.
50 I
30 .
/
.
. I
.
.~
1.5'
.5"
HC -
. I I
." (.0
Q'" f f44 "
Figure 5 - Cumulative Relative Frequency Distribution of HC Emissions
of 54 Fleet Cars Tested by EPA (CVS-CH Cycle)-Reference 2
D-12
-------
1'.' -
",
q8 .
1.
q5" -
Clo -
.
gO /
. '
10
50
~D
.
, j . , I
2. 3 5" '0
CO~ ~.r r...i
Figur e 6
r
20
Cumulative Relative Frequency Distribution of CO Emissions
of 54 Fleet Cars Tested by EPA (CVS-CH Cycle) - Reference 2
D-l3
-------
.",1
1.
11
18 j.
4" :
.
-------
(3)
( 3)
EPA reported on emissions of two new (less than 60 miles) M-151
Jeeps, production 1971, tested according to the 1972 CVS-C test
procedure (23 minutes).
The raw data are shown in Table 1.
At-test
and a F-test revealed no significant differences between the two data
sets,
neither in the varianc e nor in the means.
Consequently, all
data were combined and their distribution plotted, Figure 8.
( 4)
A company whos e identity cannot be disclos ed (4) established limits
of mean and standard deviations that, in the company's judgment,
would have to be pas sed by all cars if they were to comply with the
1976 emission standards.
Table Z shows recommended mean and
standard deviations together with tile 1976 standards.
F rom them,
we computed the median, x50; and x16 and x84 of the log-distribution.
Figure 9 shows the results on log-probability paper.
( 5)
(5)
FPA measured the exhaust emissions of nine Rebels and nlne Falcons,
all models 1970, using the CVS-C sampling technique in combination with
the 23-minute driving schedule as specified for 1972.
Tables 3 and 4
list the results.
Figures 10 and 11 picture their log-distribution.
(6)
Another company whos e identity_cannot be disclos ed( 6) communicated
emission results obtained from two unidentified cars by the CVS-C
test procedure (23 minutes).
Both cars were equipped with experi-
mental emission control systems.
Emissions were measured at
various mileages, Table 5, but no trend was discernible.
Cons equently,
the six observations per emission constituent and car were considered
repetitions.
The log plot of their distribution is shown in Figure 12.
The same company revealed emission data of a large group of cars of
the same make tested by the hot 7-mode cycle.
evaluation are plotted in Figure 13.
The results of their
D-15
-------
Table 1
1972 Federal Emission Test of 1971 Army M-151
Performed by EPA
All Numbers in Grams per Mile
(Reference 3)
Jeep
HC CO NOx
5. 0 150+ 0.6
5. 5 128 L9
V ehic Ie 3. 9 101 2. 1
No. 1 7.5 150+ 1.2
4.4 110 2. 1
5. 0 109 2. 1
3. 2 87 1.3
6. 2 103 1.8
Vehicle 4. 8 115 3.0
No. 2 7.3 137 2. 3
5. 0 103 1.6
5. 9 102 1.8
D-16
-------
qq
qB He co NO..
tf~
10
fO
10 IJ)
50
~l>
I»
)(
, ,
~
HC- 'f(~;
I
3
,
. .5'
: I I
"'
I & q
'0
100 ,'" (>
CD - ~ ' f tift;
! , , ;---1-- t
1,0 1.0 3.0
NO - ~'f/h\i
,..
Ilgure 8 -
Cumulativp Relative Frequency Distributions of Emission
Constituents of Two New M-151 Jeeps, Production 1971,
Tested by EPA (C VS~C Cycle) - Referenc e 3
D-17
-------
tJ
I
I--'
Manufacturer Recommended Computed from x and s
Mean and Standard Deviation 1976
Exhaust Standard
Cons tituent Standard Median
Mean Deviation x99.9 x50 x84 x16
x s
HC . 183 .07 . 41 . 171 .247 . 118
CO .99 .52 3.4 .876 1.435 . 535
NO .27 .0523 .40 .265 . 321 . 219
x
())
Table 2
Data x50' x84' and xl 6 of Log-Distribution, Computed from
Emission Data x and s Recommended by Unidentified Manufacturer
-------
~ctq
oft>
qg
qS"
qo
50
~o
80
70
NOv
He
/
I: I
,~
. I
.l .~.~
1-1 c) 1J0'lr - gd....i
.5"
1.0
lD
-r-
2.0
,cO - q't (h4;
Figure 9 - Cumulative Relative Frequency Distribution of Emission
Constituents, Derived from a Forecast of Unidentified
Manufacturer for Model 1976 Cars
D-19
-------
Table 3
Emission Results - Rebels - EPA
(Reference 5)
I 1972 FTP
Car grams/mile
Number
HC CO NOx
44044 4.23 55.38 8. 04
54378 2. 67 20. 70 7.27
54368 2. 24 13.78 6. 04
5-B69 3.62 28.98 7.59
54380 1. 60 19. 52 3. 59
54373 2.34 9. 09 8. 28
5-B60 1. 89 9.95 6. 19
5-B64 2. 18 18. 48 6.85
54370 3. 28 23. 31 7.82
D-20
-------
Table 4
Emission Results - Falcons - EPA
(Reference 5)
Car 1972 FTP
Number grams / mile
HC CO NOx
54356 3. 17 11.01 9. 73
46982 4. 19 13.07 9.96
49738 3.66 13.05 9. 20
54352 2. 81 16. 82 8.02
49741 4. 21 24. 62 12. 17
46983 3. 33 22.97 8.06
46988 4. 34 13. 67 10.49
46985 3.56 15.77 9. 31
49734 4.04 12. 91 7. 85
D-21
-------
er8
'5
110
fo
10
°/0
50
~o
II
.
I
!O
7
20
30
co- '3"("';
.
I
I
I
I
.
I
.
I
.
'0 16
/./0'(- tel-'
1S"3 4- ,.
H( - gf/r'ft;
Figure 10 - Cumulative Relative Frequency Distribution of Emission
Constituents of Nine Falcons, Model 1970, Tested by
EPA (C VS -C Cycle) - Referenc e 5
, l' l' . .
L I
r,
D-22
-------
/
..
/
.
.
it
S"
NOw
10
~o
20
'to 50
, ' .
10
Ul
q5 +
'to
%
fD
70 .
50
3D
2.
3 4- b
11£- '3~fW4"
Figure 11 - Cumulative Relative F.requency Distribution of Emis s ion
Constituents of Nine Rebels, Model 1970, Tested by EPA
(C YS-C Cycle) - Referenc e 5
'0 - ''!1(M;
D-23
-------
Table 5
Emis sion Data of Two Unidentified Cars
Equipped with Experimental Control Systems
T
HC CO NOx Mileage
Car 11 1 o. 4c 1 2.48 1. 4c9 0
O. 02 2. 76 2. 27 1, 000
0.38 2.37 1. 4c5 2, 000
O. 35 2. 36 2.17 5, 000
O. 32 2. 20 1. 52 10, 000
O. 38 3.90 2. 25 15,000
Car #2 O. 51 2. 80 - 0
O. 21 3. 20 1.90 0
O. 31 2.30 1. 65 1, 000
O. 04c 1. 50 1. 90 2,000
O. 26 6. 80 1. 30 5, 000
o.n 2.00 1. 15 5, 000
---
D-24
-------
C...-I' 4f: I
C CAr ft 1..
.
j
1.5"
2
3
b
it
CD- ~'fl~i
Figure 12
I
~S- ~
qO
0/0
10
50 j
30 1
.1
.~
.5"
~(.- i{fh<;
D-25
2. 3
N0'tc - qt i~;
- Cumulative Relative Frequency Distribution of Emission
Constituents of Two Unidentified Cars (CVS-C Cycle)
0= Car#l
x = Car #2
-------
qq~ HC, co
~
qq
30
lo
30
CD- !"/";
-------
(7)
In a memorandum to EPA~7) CAL described the results of an emission
test performed on a jeep equipped with a low emission FCP engine.
The results of this evaluation together with the computed xSO' x16'
and x84 values are presented in Table 6 and plotted in Figure 14.
Table 7 summarizes the important input information of Figures 3
through 14.
The first columns indicate the sourc es of variability involved
in the data plots.
The test error, discussed earlier, is always present in
all measurements.
The product variation enters if more than one engine
is involved.
Figure 11 gives an example where the emission distribution
of nine Rebels is plotted and where the distribution reflects the test error
of each engine, combined with the variation between the nine engines.
The deterioration factor would apply for measurements taken at
different stages of engine life, and the variation between engines would add
still another portion of the total variation if different engine makes are
involved, such as in Figures S through 7.
No conclusions about these four sources of variation can be made at
this point other than to state their presence or absence.
We would expect
the total variation to increase with the number of error sources involved,
but this conjecture would have to be verified by facts.
The following
evaluation of test results is one first step in this direction.
D-27
-------
t:J
I
N
00
Measured Emission Data Computed from x and s
Exhaust
Constituent Standard
Mean Deviation Median
-
x s xso x84 x16
HC .328 .0578 . 323 .385 . 271
CO .824 .1347 . 813 .956 .691
I NOx .298 .0476 .294 .345 .250
Table 6
Data x50' x84' and x16 of Log-Distribution, Computed from
x and s of Measured Emission Data of FCP Engine
(Referenc e 7)
-------
qqff
Oft>
qq
qS-
qO
80
10
50
30
I
.2.
.3 .4
NOy - q.,/...;
, I r
.S
1.0
CO-I!j'f/_i
.2.
-3
.4
..5"' .,
Ii(- IJ"I (.,.;
Figure 14 - Cumulative Relative Frequency Distribution of Emission
Constitutents of One Jeep Equipped with Low-Emission
FCP Engine (CVS-CH Cycle) - Reference 7
D-29
-------
Figure
4
tJ
I
v.>
o
10
11
12
13
14
3
5
6
7
..
8
9
h
o
h
h
W
J::
"'-' 0
U .....
::J "'-'
"';j .~
o h
h ro
0,>
"'-'
(f)
CJ
E-<
x
x
x
x
x
x
x
x
x
I h
ro 0
h "'-'
o u
..... ro
~ ~
....., J::
Q) 0
0,""
.....,
x
x
-
x
x
x
-
x
-
(f)
....., Q)
J:: ..'G
Q) ro
~ ~
::::: h
..... ro
°u
x
x
-
-
-
-
x
no
info
-
Milcaf.2 '
-
0-20,000
.F:mis sion
Component
HC
co
(f)
'+-< J::
o 0
.....
h .....,
Q) ro
Test E ~
Proc edure ::J (f)
2c)
NOx
FTP-hot
C VS -HC
x
x
CVS-C
Remarks
Unidentified GM cars
~160 (model 1970) in customer
use
54
Unidentified fleet cars
CJ
u
J::
Q)
h
Q)
'+-<
Q)
r:r:
( 1 )
( 2)
( 3)
( 4)
( 5)
Two cars equipped with exper (6)
imental 1975 control systems
I' "1970/71 production engines
large
of same displacement
6/12
~17
12
Two Jeeps (production 1971)
no emis s ion controls
x
4, 000 to
32, 000
x
x
x
x
x
x
x
C VS-HC
00
Anticipated level of
1976 cars
-
0< 60
x
x
x
C VS-C
9
1970 Falcons (Ford)
-
end of
assem-
bly lin e
x
x
x
CVS-c
9
1970 Rebels (J\M)
-
Low
x
x
x
CVS-c
FC P engine in jeep
Summary of Input Data for Figures 3 through 14
-
Low
-
0-12,000
,
-
no info
-
4,000
x
x
x
x
x
x
Table 7
-
FTP-hot
x
C VS-C
( 6)
(7)
-------
DATA EVALUATION
In order to obtain some idea of the relative magnitude of the test error
and its relation to emission differences due to product variation and other
sources, all distributions pictured in Figures 3 through 14 were standardized.
For normal distributions, standardization can be achieved by referring the
standard deviation, s, to the mean, X.
The relative variation, six, would
then serve to compare the various normal distributions on a percentage
bas is.
For logarithmic distributions, as they prevail in emission tests,
comparisons can be made if all cumulative frequencies are referred to the
median value, x50' This is accomplished easily by shifting the averaging
distribution lines transversely to the emission value "unity", as indicated
in Figure 15.
In this fashion, all emission distributions, Figures 3 through 14,
were standardized.
Figure 16 shows the standardized HC distributions,
Figure 17 those of CO, Figure 18 of NOx.
Each distribution is represented
by a straight line.
Note, however, that these lines reflect distributions
of samples only and not of populations.
Table 7 indicates that, in some
instances, the number of observations per sample was rather small, so
that the deviation of the sample distribution from the population distribution
may be large.
Because the distributions are not normal, we cannot readily
compute their confidenc e limits.
We have to keep in mind, however, that
the positions of distribution lines for small samples (Nos. 8, 10, 11, 12,
and 1..J:) are r-ot exact;
they can deviate towards smaller values by, say,
20% or more, and towards larger values by, say, 50% or more.
In general, one would surmise the slope of the cumulative emission
distribution to reflect the number of involved error sources;
emission data
D-31
-------
~
'to. ~ / /
7\ .~/
.... 'b' Ijc
s
~ .. ,
" ~~ I
.. 80 . ~ / tv~
-.:: .......~;'
"-/ ~I
70 / ,'
v - /
~ J/
'.,
- &0
III
~
..
So
~
..;. 40 .
cs
"';
I 30.
"::.
....,
20.
I I
I , I
1 E~;!5IO'" CD'" ~Dn(l<. - q' 1M;
Figure 15 - Standardization of Cumulative Relative Frequency
Distribution
D-32
-------
Figure 16 -
"" .1
0/0
qq
q~
qo
80
70
so
30
13 12. q II ,3
! ,I /ij
..
Sources of variation:
test error
test error + product variation + deterioration
----
test error + product variation +
deterioration + different car makes
, ,
, .
It
~
1
.3
Cumulative Relative Frequency Distributions of HC
Emissions -- Standardized. (The numbers refer to
Figures 3 through 14)
z.
~c
D-33
-------
12// lit 10 Lt ((
/ ' /
qq.«f I
;0
-------
~q.q
qq
qg
(15
. qo
$u
7D
fO
I ~O,
[
I t
co 14 q U
--~ ~ \ \
7
\
g
I
12.
I
f ---
I
I
I
of variation:
-
test error
test error + product variation + deterioration
----
test error + product variation +
deterioration + different car makes
i--- -- ~-- --0-
1-
NO
~i --~
3
-'+
5"
I
-HI
J:2.g,~':~~ 18
Cumulative Relative Frequency Distributions of Nox
Emissions -- Standardized. (The numbers refer to
Figures 3 through 14)
D-35
-------
encompassing the test error, the product variations, the deterioration due
to mileage, and the variation between cars should be larger than those
containing only, say. the test error.
Figures 16 through 18, however, do
not always comply with this idealized picture.
The lowest CO variation
is realized by a car equipped with 1975 experimental control equipment,
curve No. 12/1 in Figure 17, but the same car exhibits the highest NOx
variation, curve No. 12 in Figure 18; whereas the lowest NOx (and HC)
variation is achieved by a group of nine new Falcons, model 1970, curve
No. 10 in Figures 16 and 18, and not by a single engine.
cars,
The largest HC variation is exhibited by approximately 160 GM
model 1970, driven by various customers up to 20,000 miles, curve
No.3 in Figure 16.
This is as expected becaus e the emis sion data of the
GM cars encompass all error sources that can possibly be involved.
The
largest CO variation, on the other hand, is realized by a group of nine
new Rebels, model 1970, curve No. 11 in Figure 17, with only the test
error and the variability between cars involved.
Figures 16 through 18 dis?lay also the distributions of two single
engines.
Curve
No.8 shows the distribution of a standard L-141 engine in
a M-151 military jeep, production 1971, with practically no emission controls,
and curve No. 14 the distribution of a FCP 141 engine in the same jeep with
maximum emission controls.
The CO and the HC variation of both curves
are low (albeit not the lowest) and ess entially equal, Figures 16 and 17.
The NOx'variation of the FCP engine, curve No. 14 in Figure 18, is still
low;-butthat of the lUlcontrolled engine, curve No.8, is high.
We suspect this lack of consistency to be caused primarily by the test
e r ro r.
It seems likely that the test error, which is present in all measurements
D-36
-------
changes widely with the car make, the car service, the measuring equipment,
the test operator, and the like, to the extent as to obscure the contributions
of other sourc es of variation.
( 8)
recently conducted a survey of six emission test labora.tories by having
The Automobile Manufacturers Association
each of them test the same low emission vehicle.
The results are listed in
Table 8.
Although the given data do not allow computatiull of the test error
within labs, we suspect the variation between lacbs to be large.
For instance, the test error of the Ford lab seems to be much larger than
that of the AMC lab as regards HC emissions.
Similar differences of test errors may occur when cars of different
makes are tested.
The Falcons, model 1970, for instance, exhibit a
relatively small variation in HC and NOx, see distribution No. 10 in Figures
16 and 18.
Since this variation encompasses both the test error and the
variation between cars, the reproduction error must be even smaller than
the shown total variation.
The experimental cars, distribution No. 12 in
Figures 16 and 18, on the other hand, show a relatively large test ~rror in
HC and NOx.
The ratio between the reproduction errors of the Falcons
and the experimental cars is at least 1. 8 for HC and 3. 2 for NOx.
It is
impossible to decide to what extent these large ranges are caused by the
cars thems elves and not by errors of the instrumentation, but we suspect
that a large part is due to differences in car performance.
1\11 we can state then is that the test error is by no means a constant.
It depends on the measuring equipment, on the test procedure, on the car
make, on the operator, and the like.
The test error can be small as evidenced
by distribution No.1 0 in Figures 16 and 18, and No. 12/1 in Figure 17.
If we want to as sign numbers, we could form the ratio of x99. 9 and xso
(the median) and state that since the ratio of x99. 9/xSO' encompassing test
D-37
-------
tJ
I
Vc
00
1972 Test Procedure
No. Grams/Mile (Average)
of
Test Laboratory Tests HC CO NOx
American Motors Corporation 2 0.48 22. 93 1. 99
General Motors Proving Groillld 1 O. 58 23. 00 2, 03
Chrysler 1 0.43 33. 26 1. 11
Ford 3 1. 00 22.84 1. 54
International Harvester 2 O. 82 30.60 3.08
Environmental Protection Agency 1 O. 80 24.06 1. 67
Table 8
Emission Data of the Same Vehicle Tested at Various Laboratories
-------
error plus variations between cars, reached a m1n1mum of approximately
1. 5, the test error could well be significantly smaller.
In other words, the
smallest test error encountered in our studies was of a magnitude that would
refer 99.9% or more of all data (measured repeatedly on one single engine under
identically controlled conditions) below the limit of 1. 5 x median value.
Whether this small test error can be realized for all emission measurements
if performed with care and good instruments, or whether some engines
perform erratically to the extent as to surpass this minimum even with ideal
instrumentation, cannot be determined without additional information not
available at this time.
data.
Figures 16 through 18 exhibit also the maX1mum variation of emission
Maximum variation should be expected at a combin<..tion of all pos sible
sources of variation, i. e., the test error, product variation, deterioration
factor, and variation between cars of different makes, which is realized by
distribution No.5 in Figure 16, No.6 in Figure 17, and No.7 in Figure 18.
The data show, however, that distribution No.5, which reflects the distribution
of HC data of 54 fleet cars of different makes, is rather small compared to
distribution No.3 reflecting the HC distribution of 160 GM cars.
This is
perhaps understandable because fleet cars are usually better serviced than
cars in customer use.
The CO data of both groups of car,
however, are
almost equal, see distributions No.6 and 4 in Figure 17.
50 again, no
consistent picture evolves.
On the whole, it seems that the maximum variation of CO data is
about twice as large as the maximum variation of HC and NOx data.
in terms of the ratio of x99. 9/x50' the maximum range 1S
(I>'(' He
Expressed
X,I.,'I Ix50
~
u,
fo '('
{H
Co
NO"
D-39
-------
As regards the FCP engine, we may conclude that although its test
error is low, see distribution No.
14 in Figures 15 through 17, it is not the
lowest.
There seems no reason to as sume that the FC P engine would
exhibit a lower test error than any ordinary carbureted engine.
D-40
-------
SUMMARY
Emis sion data of various groups of vehicles, all equipped with internal
combustion engines, were analyzed with the objective of extracting information
about the test error of the single car tested repeatedly with the same equip-
ment under identically controlled conditions.
The analysis revealed that:
.
The distribution of emis sion data can be transformed to approximate
normality by using the logarithm of the emis sion values rather
than the values thems elves.
.
It seems that the test error changes widely between cars and
between test labs.
.
The smallest test error encountered was of a magnitude that
would place 99.9% of all measured HC, CO, and NOx data below
the limit of 1. 5 x median value.
Whether this range is typical
for good measuring techniques and well-serviced engines could
not be decided.
.
There seems no reason for believing that the test error of the FCP
engine would yield a lower test error than ordinary carbureted
engines.
D-41
-------
REFERENCES
( 1 )
General Motors Corporation:
"A Progress Report by the General
Motors Corporation to U. S. Environmental Protection Agency" -
Ma r c h 1 2, 1 97 1.
( 2)
Private communication
(3)
Environmental Protection Agency, Offic e of Air Programs:
"A Report
on the Exhaust Emissions of the 1971 Production Version of
the Army M-151-Jeep" - by J. C. Thomson, Rep. No. 71-27,
April 1971.
(4)
Private communication
( 5)
HEW, Nat. Air Pollution Control Administration, Div. of Motor Vehicle
R &: D:
"Exhaust Emissions from Ten GSA Rebels and Ten
GSA Falcons Equipped with LPG Conversion Kits"
by H. L.
Gompf, Rep. No. 71-10, October 1970.
( 6)
Private communication
(7)
Cornell Aeronautical Laboratory:
"Analysis of Emissions Test Data
on Low-Emis sion FC P Engine Installed in M-151 Vehicle" -
prepared for EPA, December 1971.
( 8)
"1975/1976 Light Duty Vehicle Emission Control
Ford Motor Company:
Program" - Status Report to the Environmental Protection
Agency, October 18,1971.
D-42
-------
APPENDIX E
Ei\!ISSIOI\S ANAL YSI~ OF 54 FLEET-TYPE PASSENGER
VEHICLES
D, J, Schuring
Il\'Tl\ODUC nON
Fifty-four
1970-model
fleet cars of different makes and mileages
were tested by £PA using both the CVS cold-start and the CVS hot-start
test conditions.
On !\'ovember 15,
1971, CAL received from £PA the results
of these en1ission tests \\'ith the following remarks: "A two-bag cold start
test was run on each car followed by a two-bag hot start test.
To obtain
three-bag data, it is necessary to weight and assemble the data from the
t\,-o cold start bags with the first bag of the hot start test using the procedure
described in the Federal Register."
The identity of the vehicles could not
be disclosed by £PA in any way other than by range of engine displacement,
number of cylinders, acc urnulated mileage, and car weight.
The emission data listed in the computer print-outs gave occasion to
study the following questions:
.
To \\hat extent does the accumulated mileage influenc e the
emissions level?
.
Is there a significant difference between the emissions levels
for cold start and hot start?
.
What is the numerical relation between the CVS-C and the
CVS-CH testing procedure?
The answers to these questions were felt to provide, by inference,
some
inputs into the planning of Stratified-Charge-£ngine tests.
£ -1
-------
INFLUENCE OF ACCUMULATED MILEAGE ON EMISSIONS LEVEL
--
The accumulated mileage of the cars tested ranged from approximately
4,000 miles to 35,000 miles.
A visual inspection of the emis sions data
from an analysis of each bag plotted versus accumulated mileage strongly
suggested negligible correlation between the two variables, Figures 1
through lZ.
An analysis of variance, described elsewhere, also failed to
reveal a mileage trend (nor did it confirm any other trend associated with
number of cylinders, displacement, and car weight).
Hence, all cars can
be considered as drawn from the same homogeneous population.
As an interesting aside, it may be mentioned that in contras t to the
toxic emission components (CO, HC, and NOx), the non-toxic component,
COZ' did indeed reveal a mileage correlation. Figure 13 shows C02
data in gr/mi for the first cold bag as a function of mileage suggesting an
upward trend of emissions with mileage.
An analys is of varianc e, Table 1,
confirmed the significant contribution of mileage to the overall variability
of COZ.
Table 1
Analysis of Varianc e on COZ Emis sions (gr / mi) of Fifty-
four Cars, Divided into Three Mileage Classes (<.10,000;
10,000 - ZO, 000; > ZO, 000 miles)
Source Sum of Squares I Degree of Mean Squar e
I Freedom
,
Between mileage clas ses 4 x 104
14.9ZxlO 2 7.69
Within mileage class es 4 x 104
65. 80 x 10 51 1. 29
TOTAL 4
81.72 x 10 53 F = 6. 17
F - 5 0
2,51,99%~ .
-_.
E-Z
-------
INFLUENCE OF START CONDITIONS ON EMISSIONS LEVEL
Figures 14 through 17 depict the means and the standard deviations of
the HC, CO, NOx' and C02 constituents for each of the four bags.
The mean
is indicated by a small circle, the standard deviation by the length of the
vertical bar extending from the mean value in both directions.
Obviously,
the HC and the CO contents (in gr/mi) of the first (cold-transient) bag are
significantly higher than thos e of the rest.
In fact, the differenc es between
the second (cold-stabilized), third (hot-transient), and fourth (hot-stabilized)
bag are so slight as to suggest a constant level of CO and HC emissions after
the first bag.
Thus, the HC level of the first bag is approximately 1. 6 times
higher than the average level of subsequent bags, and the CO level of the
first bag approximately 3.4 times higher.
The situation is different for NOx'
Here, the NOx content of the first bag is always higher than that of the
subsequent bag regardless of whether the engine is started after a cold soak
or a hot soak.
The C02 level is of lesser importance.
We note that it
remains approximately unchanged.
Another descriptor of interpst is the standard deviation.
In HC and
CO emis sions, the standard deviation seems to change proportionally with
the mean, an obs ervation confirmed by analys es of airc raft emis sion data
(Reference 1).
The standard deviations of both NO and C02' however,
x
indicate no dependence on the mean.
For comparison,
the emis sion data of the 54 cars were contrasted
with the emission data of an FCP engine installed in a jeep (Reference 2).
The FCP data are listed in Table 2 and indicated in Figures 14 through 17
by black bars;
the bar c enter is identical with the mean, the bar length
with twic e the standard deviation.
The general trends of both plots are
E-3
-------
similar although the HC, CO, and NO values of the FCP engine are
x
strikingly lower than those of the 54 cars (due to effective emission control
and lack of variance between engines).
Table 2
Constant Volume Sampler Results of FCP Engine
EPA, October 7, 1971 (Reference 2)
Standard Rel.
Bag Size Gas Mean Deviation Variation
gr/mi gr/mi 0/0
HC - FID 0.97 O. 16 16
Cold CO - IR 1. 30 0.43 33
Start 14 C02 - IR 480. 45. 9
NOx - C I 0.40 O. 13 33
HC - FID O. 18 0.05 3
CO - IR 0.78 O. 28 36
Stabiliz ed 14 C02 - IR 509. 37. 7
NO - CI O. 28 0.06 22
x
HC - FID O. 27 O. 1 11
Hot CO - IR O. 76 O. 26 34
14
Start C02 - IR 451. 53. 12
NOx - CI 0.39 O. 12 30
E-4
-------
CYS-C VERSUS CYS-CH TEST PROCEDURE
The Constant Volume Sampling Procedure for 1972 prescribes a 1369
second, 7.5 mile, non-repe~itive driving cycle with a 12-hour cold soak
i
before testing and a cold start (Notation: CYS-C).
The Constant Volume Sampling Technique for 1975-76 prescribes the
same 7.5 mile driving pattern as the 1972 procedure.
The e mis s ions of the
first 505 seconds are collected in a "cold transient" bag, thos e of the next
864 seconds in a "stabilized" bag.
After ten minutes hot soak with engine off,
the driving cycle is repeated with the emissions of the first 505 seconds
collected in a "hot transient" bag (Notation: C YS-CH).
The first (cold transient)
bag is weighted by a factor of 0.43, the second (cold stabilized) bag by a factor
of 1. and the third (hot transient) bag by O. 57.
All weighted emiss ions are
then added and the sum divided by 7.5 to give the emissionf in gr/mi.
In order to compare the CYS-C and the CYS-CH procedures, the four-
bag data of the 54 cars (a cold-transient bag, a cold stabilized bag, a hot-
transient bag, and a hot-stabilized bag) were ass embled in two ways.
CYS-C
CYS-CH
(Y + Y )/7.5
ct cs
(0.43 Y + Y + 0.57 Yht)/7. 5
ct cs
= Y
m
=
Y
wm
in gr/mi
in gr/mi
Y
- ct
\
= me 5S emissions of pollutant (CO, HC, NO ) as calculated
x
from the cold-transient phase (bag 1) of the cold-start test,
in grams
Y = mass emissions of pollutant (CO, HC, NO ) as calculated
cs x
from the cold-stabilized phas e (bag 2) of the ~old-start test,
in grams
E-5
-------
Y =
ht
mass emissions of pollutant (CO, HC, NO ) as calculated
x
from the hot-transient phase (bag 3) of the hot-start test, in
grams
Y
m
=
mass emissions of pollutant in gr/mi;
CYS-C test procedure
Y = weighted mass emissions of pollutant in gr/mi;
wm
CYS-CH
procedure
These computations were performed for each of the 5-1 cars (actually only
53 because one car's data were quite erratic and had to be omitted).
For
each car, the ratio of Y /Y was then calculated, and the mean and the
wm m
standard deviation of the distribution of thes e ratios computed. The results
are listed in Table 3 together with the standard deviation of the mean
(s
mean
= s/ (53).
Table 3
Comparison of Emis sion Levels Computed
According to CYS-C and CYS-CH Method
Y
wm
- CYS-CH Mass Emissions in gr/mi
= CYS-C Mass Emissions in gr/mi
Y
m
/Y /Y Ywm/Y m :
Y Y
Pollutant wm m wm m Numb e r Standard
Mean Standard of Deviation
Deviation Cars of Mean
I-IC O. tl8 0.07 53 0.01
CO O. 71 O. 13 53 O. 02
N0L, 1. 01 0.05 53 O. 01
Table 3 demonstrates clearly the attenuating effect of bag weighting as
regards HC
and CO.
The HC level is reduced by 12"" and the CO le\"l'l by
L <);:J.
The NOx level, Oil the other hand,
reJnains wlal1ectE'd.
F-6
-------
SUMMARY
Fifty-four fleet cars of different makes and mileages, all 1970 models,
were tested by EPA using both the 1972 CVS one-bag and the 1976 CVS
three-bag test conditions.
Although the accumulated mileage of the cars
tested ranged from approximately 4,000 to 35,000 miles, a correlation
between mileage and level of CO, HC, NOx emissions was not detectable.
(There was a slight but significant increase of C02 with mileage, however.)
The mean and the standard deviations of the CO and HC emissions sampled
in the cold-start bag were significantly higher than those of subsequent bags.
The HC mean of the cold-transient bag was approximately 1. 6 times higher
than the average of the subsequent bags; the CO mean, 3.4 times.
Also,
the standard deviation of both emissions seemed to change approximately
proportional '.'lith the mean.
The situation was different for NOx'
Here,
the mean of the transient bag was always higher than that of the stabilized
bag regardless of starting conditions.
The C02 level remained approxi-
mately constant for all bags.
The results were qualitatively corroborated
(except for mileage effects) by results of emission tests performed by EPA
on one FC P engine installed in a jeep.
An investigation of the effect of bag weighting on the measure.d emission
level revealed that the C VS-CH technique produces a 12% lower HC level
than the CVS-C procedure and a 29% lower CO level.
unaffec ted.
The NOx level remains
E-7
-------
REFERENCES
1.
Cornell Aeronautical Laboratory, Inc.:
II Analys is of Airc raft
Exhaust Emis sion Measurements" - CAL No. NA-5007 -K,
November 1971.
2.
Private Communication:
Computer Printout of EPA.
E-8
-------
c..00. ..
j--
!
I
:0
I
I
o
I
M
I
--0
, I I :
: I I I
. I '
I II ; !
I I'
, I I I
I, '
: : :0 0
80' _,.L! r".1 ~. O,C{)(-l'>") . ~
:-1. }\ IS!
- (.--1, 0 J$---'O - 6) 'CO-;
1:0; !--;!;i@ OO()~ 0 . - C~ i
b i I; I n Ud0
,.; ;-I iDc 0 0, .
'/ t , J > .
:~ \':11'1 ~ .!: ; . .0
. I I "" I 1 JI ; .
. i v I '
~ 0 --' i - r: - r
I I
: c=- !J-i
'l' I
--':_Hi~----I ~ H"'.
5 -dLi
o
o
100;-
i
: - - ~-i ~
so. .).
YJ -
:0
. I. L---.: - l
1.-:-~----
-.1- j ~1=~~-~ -- -~ - n- . ~ -_.
1;1 --_1.0
-I -1.1-
1.1:
i I ! ! i
~-1 -- --. . . ,--. .
i 11.1- - - - L
I : ;
I I
I ~ .
I I I '
.: I :
,- ~.. :
. , I I
! ~ - ;
! I .
I I
I ; t
Ii:
l _I.
; i
I
I .
I I
. -..
I - ~ .-1--
i ~-~r' "j
I
I
- . . 1--.
I
!.
3
I
i
--I
I
!
I
.1.
. ---
'i! I
I ! : ,
. :MI/..';::'AG..E:
5 10
: 3
!<~O
2.2.
J-
.3
2..0
30
I :
, . i
. - .1...
J : ~
. j
, ,
I - - ~
.J
!
I
I ;
1 J
I' 1 I
I '
I ,
. -"i
; :
! i ;
50
80 Joe,
Figure 1 -
CO Emis sions of C old- Trans ient Bag 1
Versus Accumulated Mileage
! .
. - . .'- I - -, . 1
j Ct;- - .-4 . ; - :-: ~-
~ -~ I - i: ( ! - --' - - - .. ~ !
I H- I ! pi -- - ; --. I q , , i
I .. 1- [T i ~._- ;--~ poJ ! j i
30 - :4~~/MI.' i ,'--I-! i _u- J7-'! :V,O; i i r
.. 'I . Cf- . . U ,-- , ,
! . ! t I!' , 10 i ;: : ~
'O;! i ..I-#I:~: ~~o r TI. II
10 ~._~: --':,-~--?ot~r'l r~ j~ -~~.--._.::--' -~- H~'~~J ~:~~! ~--
8 . ..: -'1'---'-- -:.m I t- r I q -, --0 . 0-=rh' - - L- -+ -; '1-- !U~t
-~.. Ju~i~~rtru~~~~Q--~~.?:' L~~J'~~~'T~1-J-T
i ! I ! I : I ~ I J . i ~ !
5 .m: - i---. ,Hi-fi' r-'O' ~u i .0--,-- ;--1'-1' T
1---, 't!':' r--1-- T----r'-- -~--'il r-:
- - : -. ~- _J .. 1 ~! I: i I
~-----;- I i I : I I ---i---' -'[U'- '- T- '-1-~
1 ; 1 ~ - ~ - . . . 1
, HILEN-,e X:/DJ ; I :
e 10 20 ~a Su 80 100
2CO;
;
for---. .
E;Q
50
I
, - -- - 1
I i
I
I 1
, ' ,
! ! .
I : !
ii,
I I '
II i ;
! '
, '
, . I
;
i
I '
I !
i
. .J
..- ,J .;
,
I i
f- ; 1':-
.' -, '1' 1'1
1 -. -I c'
L I.: ;---
I i I
- -1' ~ - - ; .
- i J :.
I - r
! un -.j
I
-~
3
2
2
.3
s
Figure 2 -
CO Emissions of Cold-Stable Bag 2
Versus Accumulated Mileage
-------
, I
! ; I I
M
I
>---
I' "
: j : i I J I
ILl:. ; j , ;
200, ' I ' : 1 -11- I' '
i - . I :! I
! : Iii! i
'oof~ -' - ,:-;:-!! j ,
, :--,--- '-: -it - -- - -'-j --,- : - i: -t i~
AO~n~-::__; --~+~-'~-;,'_~J--ji -lU-:~= -. ---;- -, , : '-I'~ i
. i - --. t 'I!: - I I i -:
50!- - ;--fi -jj-iTi 0 - l_q --1- r ;-1-11
30&Pif!-i; !o I , :: :
, ,~ :5 5' B' G 2.0 30 $0 So 100
Figure 4 - CO Emissions of Hot-Stable Bag 4
Versus Accumulated Mileage
-------
zo
M
i
>-'
>-'
! ! 1 ;
ii' I
,[ I I
: Ii' , I
':'.u.J-)utJrJ.'I~~ll ,--~o . :01 .u.,.:.: " !,
.-1-1-:-:+1:- ,Q --: i i I
s~=::)-l~i,-r;I:-~ti~~Lii: ii,
3 ',_~M/;~I_V--C-'G'- ,-L-H' -- a<%-; @: . u1 u ',Ll '
; I L \ I ' I I , ' ! 'i 1 ' : ,
I "~ '0 I I 111 j 0' i. Ii : i i i i
2 i-i-I-- -lT1lT -~T u r'r.1 iTii
I L.LJ _L]IJJ.-'----LJ__--- .-.1-. LtJ.!
, 'I f +'" , ;, , L" I
"--~:L---- --t---T---\--t-J--r-----;--- r- - --"1 ----i~~- d', -i- r-r":
Q'81-'~'[:-- _+--1' r-' ,-1--c---i ... '1- 1" I' " r-t-1':
oS '-=-=+ r:ftj7,:~jt- rr: ....T '--f:Ti,T
'.---t~ :" I'"
o ;'!'-In 1--I--lTl ~~EA(!~-~:'/~;- r-' 'rf --j'ri':
,32. 3 S 810 2030 50 BOlDa
i I
Figure 5 - HC Emissions of Cold- Transient Bag 1
Versus Accumulated Mileage
20, :; " i :! :' - i ; ii' , . '1'j; I i
: :: ! f ;, i I - ; - 'I . ~, ; - j - "I In I . ' 'I! i
I I I I I I " l '
'; :-:.:A~d !tJ:~=!= 1~!: .. fC:f-:tL:f!
i--' .t- ut- -/'''1 -; :'n-_n_~_.-t---"T- -', --'r-j- rj-:
,'-'-r-...!-----t-..-j.!,;.t...----,___L--,- [---:--,'U' +Ti
: '; i ! ; 1 i : :0 : i I ; I , 1 ,
5 -~, ~~- i--~-'r~. !~--['~!'~~Jd.~__~~~~.~. ~~~~~-r.~- 'l~-' ['.~ r .~t[-D
3;' ~~(~~J-JiJ1JJ-Q~-O~i u~c0~'- -' ---~_! _I Li
" ~ . ,I ~OVi . l ,I l
2 'r:-~J----'.';: y'ttlJO ~~- . --~--o._L'Tr~Jj~,. u
I ' I I I .V I ' I. ;
h IJJ~r!]1 1-r0?O:: r .u. .....1
I :--.- ,1_,+,-1 :-_u: : . , :-rl. :
~-;-t=-T--!T;r-r:'---._'-j--:-+- "-[-:1, I
0.8 ;~::':~~t-'.~.--=j=cr:-[~~:~-=~'~-=t=--i~-r+-=- j
I. -". - ,--J.-L-I-Ll-L.i.--"":"_....1.~L-- . -- H
os i----;'-~-"u'I--1-~t-tli.--~J-rl-u"- ,. 1 .
03 .r -j-H!~IIUOAL",lj~~T-~f I ...
2. ::5 ~- e, 10 2 c:> 3 a 50 Bo I DO
Figure 6 - HC Emissions of Cold-Stable Bag 2
Versus Accumulated Mileage
-------
I ; !
[-
I
(- - -- u -
. . -i ,~-L ".; ,: f'f ' ;
'; I I " I ' '
S ~~_.,.' !-~=-j: . .= [=j': r' f. ~_.~ -,' r
-.t- l-f'-':- r l' :'1--' : - .
5: ---:-'1 U,] i' '....: i jiO' 6~ .
-,-L'-i--j-l-liti--~. ,~-
:, -~t'jO It! 11~~;rtr
'i 1m - I i ' 1- I
, ; ~ L ~~~~~~!-- . ,c r. ~.~' [ Ii; u
01'>'- H-r--- t---+-il f I 'i' ~ .+-- , , 'f~-
--;--r=--= - '~~ L-' ~jJ ti- - ~-- ~-.l. : . r'=. '.: 'r:t:
0,5 ~__.Ll- I-~,.IJ-) j !- ,.1.- !i, _.- n .1 !-
, , l 1 I i I I I I : ,I I Iii
Cd !-I-iTljUt~A:~ ,j,O> .. - -1!lr
2. 3 ::r B 10 2.0 30 50 bC- 160
20
. ii'
I : i
J t"
i
! '
10 '
I
,J-- -
M
I
>-'
N
J"
I
I
I, ,
.--'J
;-[ , ;
I'i [ J
n i: . !
~-r i-
I i- -i --
i I i
1 I'
, '!
,
,
I I
: I I
i: 'r
I I-
I
L
I
Figure 7 - HC Emissions of Hot-Transient Bag 3
Versus Accumulated Mileage
2°1 . : I I : I ,I ill I
,I " I ,I ' , I 'L ~
8 ~ ..==f== f~~r=r-t~FfF====-~:='T=~~ '1.-.=1 nn_f-'Y ='l=
u~ 'r--Irr--,-"-i'icy-Lul --"; --t -l-'I-', t-
5, £f:,u!P~ f,-,=-~l:I.-I=II+IL
J u-'T IUI~~ --~;I-'tll:
, -I +;--,- ,-- ,_u-l-C U It =1~
05 uu t~-:11-lr~-=~;n -Fe. LF-F -. if
0S .'-+-J=f,lf ,,-,fl--- u- :-:-L~Jf-~fl- - tf1'f
r---+-- -- .'~._nl r-!' H- ~ 1 '\ 'l
CJ! J, i ;.1 H,~~~-J~~j 'OJ--.!' - -, 1
2. 3, S 8 10 2.0,30 So 80 100
Figure 8 - HC Emissions of Hot-Stable Bag 4
Versus Accumulated Mileage
-------
:20: ",' Ii l! " I I' ' I ! i I ! : I ! l' i
i . : !: . ; I : I
t I ......,. -.., I ; .. . -. I ,.. I .1. I "/... .
1 [i : ii' ii' I ! ! . .1 i. II
IO:-=:~:_J~~J~-:-J:~J.-..L1 : f:l-~_- ~:"..' :', .,J.. ?, : r:::'~': -~-:-I ::t t:
i I"'" iJtt " I ~ "~,
8i--r-f-'[---I-" I-T'-~"--OO--&' --,- -----1'---- -I-;-i-tl
I 7?t' -£)D "
'''- -'1"'\",""'" .--.--- t''-'Y-''()t--,''- ----'-r-l
i I IV 'j I 0" : I
,=:l 1 _1--- _-I r+,'j --- 1 -- .f1' t?l..hl -- --l". L-- ,
.-- -'-- - ,,- - ,"t", 'iG) , I
: : I -L : ' "9Dq: 0 j i I I j !
5:'--j"" b-'!'-!---lbf'='B-: - -:-; 60 /60
Figure 9 - NOZ Emissions of Cold- Tran sient Bag 1
Versus Accumulated Mileage
'0 . 1- r i I: [---T :- ,--[ I tTf~
'0 !_--. _i~~I_~~__!lf{~!;-~; 1]: -~l-lJ ,~h ~
8 : --_:."~-t----i'-H--j~Lc.__._a-_-~--,, -E~. '---+-t-- ~-i-1
: 1~:l'lij=h{T _T:d]?JI~!
in;. !I j -t'llin-,.'-!,--:r~tn ,[+ 11'1.1
,L.LJ- _ULac-L- -1-,- ~!
~--_.; -f-- j. . -1"':"'+-.1..' i +--"";---1--<-- / t --,- - - il
o.e, : --r- !--f--t+-+-11~--;--L-.;--:- ---r- ,--f- .
:~~---~:! ---1-'---:-=F[j-rJ+=:' ~=-t=-'=~i=r=--~titfl-!~
0.$--' : .~-+--- r- -_:--~-"--._----_Lm_'---'+-+'i-.-+;
:. _.~_"L_~ :--.l-U--t-J -:------L---.!--.!--l++lU
03; !-- i i i--i_H1iL~Aq-_e l~.!~:{!1 ! 1111 i
2..3 ~- b 10 20 30 50 80 100
Figure 10 - NOZ Emissions of Cold-Stable Bag Z
Versus Accumulated Mileage
-------
2..0
I ! ~ : I I Iii i i:
~ ,,-, , - - ,,- ! - f L l' - -- - - -: I '" I : - -I :
10 ,1:~-~r==j~!dJ I!--::---C:J=JI!: Jtl
'I I 'I"'t ' , i" I"
o ~~-~~T-=-~S_--_T-:-;k---l,- .J_- ,- °6-g~o~~L~rFf-~IIR
I'J I ~ r ~ ~ I-f- --e--, ~~ q:-_J ---., n, l ,-.-;
5 ~---~- ---r-i--t--lcPP:~'-~ t&~-+-t- '- f-~-~
3 L~:t.ft:~l~----?ao!--;rtt~ In:
~-T ,---t -~- ,;-jTr'---FJ-T-i--'j'f Lr i rl
M <- i--; --- " ., ,:- , -'1-1 -:--r- --i--rr-1
I 1 . -,-. .. I i I .- .
i lJIIL :~I -
o'BFf --= =-= - -ti=t---==-~E~ - _:r=ttuTll
~_~-~r - rt--r-+-- ~-I-- ii-~J-+til
D5 ~- : Pi- --- - --h£~;-FH=--r+=jjllj
0,-'1 ,I ~'~"'A~L<---=I'OrT-1 tlllTl
'2 -3 ~- 5 10 2.0 -30 So 60 100
Figure 11 - NOZ Emissions of Hot-Transient Bag 3
Versus Accumulated Mileage
2 o!U i-- I i 1- i ! i i _U_- i'-- i ': i j' I, ,! " - "
1-TJuJji[_-L1Jn+~. -I
f-:- -'_._----;-+ t--f-d+-~--'---;- t' -fl~ '- ~
8 F-: =-I-=t-=i-i:t1-'-=1 0--- ~_n~~'i-~L~-- r -f-- I
5'-~TT1ql :;-"oiP i- - -1 j
iI.I; ),=t_11 0° I ..-9 1@-1--.- ~nF . I: J
-3 :-i-JF-r-I i!" 0 -~ -
1--':---1--:-1"-1- -y: j-:::~, -1---------
~ ~~J~I!_-~~c:.~tTi_ri
'.1 ' , : . - li~,.: > .:1 -..- -
oI8ri-=- -- __f~ -- -j- ~l--j , i--+---t-~.' ~ - , .
~ ~ _t~- -=- - ~~--=t3~-r= '.l. ~ - 1
:J-dil-, .,jLJ~-+tuL.~.;l..
2..3 S 8 Ie> 2.0 3D SO 80 100
Figure 12 - NOZ Emissions of Hot-Stable Bag 4
Versus Accumulated Mileage
-------
2, 000 ~
I ,i, .\ II i I 1[;.' i"
. I ! I ; i i . I ' Ii; I' !
,-- _P i -. II . - j-tl:j - -I ,1m-II' .. r - --- + - -- ~ -- -r T T -1-;--[-; -11--;
,: ;. : II ;: I : ) :
I, <.)60 .,-_. - - ----- - --' -,- -.! --I! -I-t -~--- ------,___-':"_nl__~~~-f- .:.._J- -
, i:: : I ': i;! : I : i : I I : ;
_un -i-;- i -i-I---[--I-:- ;---o--o--_nt-__-o~-r---~---!----'--
Soo.u -- --. -l---- - '-_u' ~-- _L-_, --II -Q~- --;-_. -~}d~-~~:_~_:_~u':"'_L_l..-
_dUU- - J - ~ - i --1-- _1_- L - I J ~ i <£ ~-1 i i 1-
soo ..-: M/ r'....-! 0..1. .:J.. .~. r;~.~o~~bOJ .~J:....[~~..~--!......
- - +- . - b ~n 1-' '-+ - !--~-+-i-- nu +-- __+_m~- -f---~-t---+--+-+-:
un' - L_L_l -~-L-~--LJJJ n_~~__t~~--L nn:-j '!-L~-J-' rl~_:
I ! I I ' r ' , ,I I',! I
' I ,I ,t I ,; I
, ! : - J -! i- - n---!- ,-1 _n -'
'I ::: 3 :' I : I
, iMIL..E:AGIoe>l 110 I' I ' '
5 2:> 1.0 2. 0 30 50
!
300 i-u-
M
I
~
lJ1
;:>00
~
3
60
Figure 13 -
C02 Emissions of Cold-Transient Bag 1 Versus Accumulated Mileage
-------
M
I
(j'-
G::,
---
"+
0.8
f !
."t
t
1
0
'BA(,#I BA c., 14 L BAG. .3 BAG. b. If
CO L1) COki) HOT HOT
TRA.NS. STA.BLE TRANS. STA8LE
G, M/M I
2.
o
Figure 14 -
Mean and Standard Deviation of HC
f 54 Fleet Cars
I Jeep with FCP Engine (EPA)
-------
M
I
I-'
-J
60
I
+- 05'-1 FLEj;T r;,...r<~ F <:"'P. EI'tG.1N Eo...
, -
--~.
<4 MIMi. ~M/MI
(~ -
-I -
! f r t
1
~ - .-
, I I
o
'2.0
3
2..
-'+0
o
8Ac.. b I
CC:>LD
TR.-.NS.
13-"'G,i>ll..
COLD
STAe,L E
!3Ac.,,,, 3
HOT
TRAN S.
13AG,~4-
/-JOT
STABLE
Figure 15 - Mean and Standard Deviation of CO
i
I
54 Fleet Cars
Jeep with FCP Engine (EPA)
-------
8 0.8
.:::JL./ PLEL:"T CAR::, Fc,P ~Nc..INE:-
. I
Go 1 1 0.(;,
C;; MIMI t
'+ 0
1 t 0 "f
M
I
>-'
CD
2. 0.2
o
f~;::.,(, 1:1 I
c..GL!)
TR AN $.
8,A~ '" 2-
C. GL D
STP--BL~
BAG.. 101:3
H 01
TRANS.
B A C, "I 'r
HOT
ST Ae L E
o
Figure 16 -
Mean and Standard Deviation of NO
x
f 54 Fleet Cars
~ Jeep with FCP Engine (EPA)
-------
800
~oo
G.M/MI t I
400
M
I
.....
-.D
200
o
BAG-. 1>1 I
c..oLD
TRAN'iI.
Figure 17 -
bAG. ""2-
Cu L"D
5TA ~ LEO
13A G-. "13
H6T
rRANs,.
Mean and Standard Deviation of C02
f
I
54 Fleet Cars
Jeep with FCP Engine (EPA)
B A L"\." 'f
HOT
5TA 8 L E
-------
APPENDIX F
EXHAUST EMISSIONS FOR CHEVROLET AND FORD
AUTOMOBILES
H. T. McAdams
Data on five Chevrolet and five Ford automobiles have been made
available for analysis of the effect of mileage accumulation on exhaust emissions.
The Chevrolets were 1969 Impala sedans and the Fords were 1969 c'alaxie sedans.
The data were reported by Automotive Research Associates, Inc. under Contract
CPA-22-69-140 for the Environmental Protection Agency.* Though these tests failed
to show a linear reI ationship between mi leage and emission of HC and CO, they can
be used to formulate an analysis of variance in which "wi thin-automohi Ie"
sources of variance can be assessed in reI ation to "hetween-automobile" sources
of variance.
The term "within-automobile variance" here is interpreted to mean
those differences in emissions which are observed when the same automobile is
tested for emissions at different times.
These successive tests renresent
different mileage accumulations but, inasmuch as re~ression analysis failed to
show a significant trend with mileage, it is permissihle to conclude that the
successive tests exhibit differences attributable to random causes.
These
differences are due, in part, to testing errors but may reflect actual changes
in vehicle performance from test to test. The term "hetween-automobile
variance" is interpreted to mean those di fferances in emissions which are associated
with the individual idiosyncrasies of the several presumably identical cars tested.
By comparing the magnitudes of the "between" and "wi thin" variance,
one can estimate the effect which individual differences among cars might be
expected to contribute to emission assessment.
These differences presumably
arise, at least in part, from manufacturine tolerances.
Thus it was felt
that by separating "between" from "within" variability, one might be able to
speculate on the effect of manufacturing tolerances on stratified charge emissions.
Analysis of variance was performed by considering variation within
cars as being nested within car-to-car variation. Thys, in a certain sense,
the several mileage checks on a particular car were considered as replicates
or repeat tests of that car. It was naturally reasoned that this variation
* Relationship of En~ine Deterioration to
Final Report. May 28, 1971. Automotive
San Antonio, Texas.
Exhaust Emissions.
Research Associates, Inc.
F-l
-------
would be less than if one looked at the several mileage points without notinr
the identity of the vehicle. Only that subset of mileage accumulat~ons were
used which were common to all five vehicles of a particular make (i.e., either
Chevy or Ford).
Each group of vehicles tested exhibited seven mileage points which
were common to all vehicles in the group.
Though the seven mileare points
for the Chevy vehicles were not the same as the seven mileage points for the
Ford vehicles, it was felt that both provided a fair assessment of within-
vehicle variation.
For each make of automobile, therefore, there were 4 degrees
of freedom for assessing between-vehicle variance and 6 x 5
freedom for assessinr within-vehicle variance.
30 der,rees of
A convenient measure of the relative importance of these two sources
of variability is nrovided by a statistic called the intraclass correlation
coefficient. 1 t can be shown that the usual "mean squares wi thin", as
computed from analysis of variance, is a measure of:?, the "within" variance.
The'between mean squares", however, is an estimate of {j2 + 702 where
a '
ua2 is the "between-automobile" component of variance. One can thus formulate
two equat ions
ty2 + 7ts.2
If 2.
mean squares among
mean squares within
which can be solved for t?
and
ua2.
Then ()2 + 0-: 2 is an estimate of what
a
It is the variance which would he
mi ght be termed the "comhined variance."
ohserved if each automohile were tested once only and the standard deviation
of resul ts was computed from these tests. The ratio
tJ a:2-
tJ'2r- ~c:
is the fraction of this combined variance which is accounted for by vehicle-to-
vehicle differences.
The data to be analyzed were regarded as consisting of 12 measures.
The first six denote HC emissions, the last six CO emissions.
In each groun
of six measures, there were hot cycle, cold cycle and comhined tests, each
test heinr evaluated in terms of constant volume samnlinr (CVS) and concen-
tration (cone) measurements.
F-2
-------
Resul ts of analysis of variance on the Chevy and Ford automobiles are
provided in Tables I and II respectively. TWo sets of F-ratios were computed.
Those tabulated in the rightmost column of each table are intended to answer
the question of whether the between-variance is significant for the particular
make of vehicle - i.e., Chevrolet or Ford. If the F-ratio is significant at
the 0.05 level, it is labeled with an asterisk. The F-ratios tabulated in the
"
Mean Square between" and "Mean Square within" columns provide a measure of
whether significant differences exist between the variability of Ford and Chev-
rolet cars. If it is noted that the F-ratio for "MS within" is sirnificant, this
fact implies that one make of automobile shows si~nificantly more variability in
successi ve tests of the same automobile than does the other make. On the other hand
if the F-ratio for "~1S between" is signi ficant, it is indi cated that one make
did not reproduce as well as the other in the vehicle-to-vehicle sense.
This fact could imply that manufacturing variability for one make is greater
than for the other. If one of the F-ratios discussed here is significant, it
is followed by either a (C) or an (F).
The presence of the letter denotes
significances at the 0.05 level;
C or F denote~, respectively. whether the
Chevrolets or Fords exhibited the higher mean squares.
In general, it appears that the Chevrolets were more variahlp than the
Fords. This fact is further evidenced by the magnitude of the intraclass
correlation coefficients. Note that these range from a few percent up to more
than 50%.
Tables III and IV were constructed by first transforminr. the emission
measurement values to their natural logarithms.
These logarithms were then
subjected to analysis of variance as before. Differences between the CheVY
and Ford vehicles, especially as to their CO emissions, are now more nronounced.
Justification for performing analysis of variance on logarithmically
transformed data resides in the fact that variability in emissions tends to be
proportional to the mean level of emissions. It can be shown that the taking
of logarithms provides a "variance-stabilizing transformation" so that variability
of the transformed data is substantially independent of its mean magnitude.
F-3
-------
On the basis of these test5 it is concluded th~t appreciable variation in
emissions can exist among vehicles of the same make.
Thus it can not he
concluded that engine-to-engine variability is negligible. On the other hand,
by virtue of the fact that one make of automobile in these tests was less
variable than the other, the prospect for good reproducibility of engines is
offered. Which of these circumstances is most applicable to stratified-charge
engines, or whether either is, is unknown.
F-4
-------
TABLE 1
/lC UlI ,c;c; r '~S FOR C/IEVY AND FORD Al1TOMlB I LES
-_.----- --
ANALYSIS OF VARIANCE
~1ean Std.Dev. Coef.Var. ~1S Between ~1S With in Intra Cor F
--~-----~--
'1easure 1 /lC-Co1d-CVS
Chevy 13.138 17.745 1 .351 430.551 295.593 0.061 1.457
Ford 15.070 10.088 0.669 135.715 96.110 0.056 1 .412
R-Ratio 2.951 3.076
~1eas ure 2 IIC-Cold-Cone
Chevy 4.816 1.273 0.264 3.559 1.299 0.199 2.741'*
Ford 5.690 1 .710 0.300 5.302 2.527 0.136 2.098
F-Ratio 1.490 1.945
~1easure 3 IIC-/lot-CVS
Chevy 10,486 4.339 0.414 21.21.) 18.429 0.021 1 .151
'"J Ford 9.620 2.822 0.293 10.432 7.552 0.052 1 .381
I F-Rat io 2.033 2.440
\Jl
Measure 4 IIC-liot-Cone.
Chevy 4.911 1 . 5 76 0.321 3.888 2.252 0.094 1 . 727
Ford 4.462 141\5 0.333 4 . I 78 1.877 0.149 2.225
F-Ratio 1.075 1.200
~1easure 5 IIC-CVS-Composi te
Chevy 10.315 5.162 0.500 61.042 20.910 0.215 2.919*
Ford 11. 830 4.424 0.374 19 .245 19.624 -0.003 0.981
F-Ratio 3. 172 I .066
"easure 6 IIC -Cone- Compos i te
Chevy 4.757 1 .261 0.265 1.859 1.544 0.028 I .204
Fon.l 5.010 I .380 0.'276 3.932 1 .568 0.177 :.509
F-Ratio 2.115 1 .016
, ])c:v)te~ sirni hcailce at II. () 5 level
-------
TAGLE II
CO L'lISe;WNe; FOR ClII.VY A~D FOR!> AUTO\fJHILES
-- ---- ---._- -- ------ -----_.
~fean
Std.[)ev.
--_._--------- ------
Coef.Var.
\IS Retween
---~-- ------------------------------ -----------
\feasure 7 CO-Cold-CVS
Chevy
Ford
F-Ratio
~leasure 8 CO-Cold-Cone
Chevy
Ford
F-Ratio
Measure 9 CO-Jlot-cve;
I-rJ
I
J'
Chevy
Ford
F-Ratio
\feasure 10 OJ-Jlot-Cone
Chevy
Ford
F-Ratio
Measure 11 CO-CVS-Composite
Chevy
Ford
F-Ratio
\feasure 12 CO-Cone- Composi te
Chevy
Ford
F-Ratio
41 435
64.558
35.825
42.247
20.917
12.889
23.909
9.880
37. 239
38.441
28.195
20.441
26.392
22.034
27.982
16.622
13.562
7.665
19.227
9.282
39 . 677
13.081
20. 116
5.733
* Denotes sirnifieanee at o.ns level
0.637
0.341
2902.709
1128.609
2.572
0.781
0.393
3038.193
190.921
15.913(C)
0.648
0.595
8 06 . 74 2
93.230
8.653(C)
0.804
0.939
2157.386
144 . 393
14.941(C)
1.065
0.340
4518.391
263.245
17.164(C)
0.713
0.280
2141.301
69.427
30.842(C)
~IS Wi thin
~~ALYSIS OF VARIANCE
Intra Cor
F
328.871
378.287
1 .150
407.120
290.528
1.401
80 . 112
53.001
1 .512
71.728
76.449
1.066
1083.540
155.750
6.957(C)
115.194
26.779
4.302(C)
0.528
0.221
0.480
-0.052
0.564
0.098
0.806
0.113
0.312
0.090
0.715
0.185
8.826*
2.983*
7.463*
0.657
10.070*
1.759
30.077*
1.889
4.170*
1.690
18.589*
2.593
-------
TABLE III
LOGARI 11IMI CALLY TRANS FORMED IIr E'-lISSIONS FOR CIIEVY AND FORD
ALJroMJB I LES
ANALYSIS OF VARIANCE
~lean Std.Dev. Coef. Var. MS Between ~IS With i n Intra Cor F
Measure 1 HC-Cold-CVS
Chevy 2.265 0.653 0.388 1.081 0.317 0.256 3.414
Ford 2.578 0.478 0.185 0.250 0.225 0.015 l.11C
F-Ratio 4.324 1.409
Measure 2 IIC-Cold-Cone
Chevy 1.542 0.247 0.160 0.116 0.052 0.151 2.248
Ford 1.669 0.454 0.272 0.258 0.198 0.041 1 .301
F-Ratio 2.224 3.808
Measure 3 HC-Hot-CVS
~ Chevy 2.289 0.339 0.148 0.109 0.116 -0.008 0.944
I
--.J Ford 2.218 0.322 0.145 0.141 0.098 0.60 1.445
F-Ratio 1.294 1.184
Measure 4 IIC-lIot-Cone
Chevy 1.547 0.298 0.193 0.139 0.080 0.094 1.726
Ford 1 .441 0.342 0.237 0.154 0.111 0.052 1.387
F-Ratio 1.108 1.388
Measure 5 IIC-CVS-Composite
Chevy 2.256 0.382 0.169 0.367 0.109 0.254 3.383
Ford 2.415 0.330 0.137 0.105 0.109 -0.005 0.962
F-Ratio 3.495 1.000
Measure 6 IIC-Cone-Composite
Chevy 1.528 0.252 0.165 0.072 0.062 0.022 1.154
Ford 1.574 0.286 0.182 0.112 0.077 0.063 1.467
F-Ratio 1.556 1.242
* Denotes si~ifieanee at 0.05 level
-------
TABLE IV
LOGARI Tllm CALLY TRA\iSFOHJIEIJ CO E'fISSIO:-JS FOR CIIEVY A'JIJ FORO
AUTOI>UilILES
--------
A'JALYSIS OF VARIANCE
~Iean Std.Oev. Coef. Var. MS BE1WEEN "15 Within Intra Cor F
~Ieasure 7 CQ-Co1d-CYS
Chevy 3.520 0.744 0.211 2.466 0.235 0.576 10.504*
Ford 4.089 0.464 0.114 0.389 0.187 0.134 2.084
F-Ratio 6.339 1.257
Measure 8 CO-Co1d-Cone
Chevy 3.386 0.632 0.187 1.571 0.205 0.488 7.666*
Ford 3.672 0.393 0.107 1.102 0.163 -0.056 0.627
F-Ratio 1. 540 1.258
"'easure 9 CO-II T-Cus
'TJ Chevy 2.848 0.681 0.239 1.924 0.220 0.525 8.747*
I
00 Ford 2.424 0.510 0.210 0.345 0.246 0.054 1.401
F-Ratio 5.577 1.118
Measure 10 CO-llot-Cone
Chevy 2.870 0.921 0.321 4.762 0.196 0.769 24.297*
Ford 2.071 0.625 0.302 0.893 0.306 0.215 2.915.
F-Ratio 5.333 1.561
Measure 11 CO-CVS-Composite
Chevy 3.351 0.729 0.217 2.118 0.267 0.498 7.945.
Ford 3.579 0.416 0.116 0.225 0.164 0.050 1.368
F-Ratio 9.413(C) 1.628
Measure 12 CO-Cone-Composite
Chevy 3.093 0.886 0.286 2.431 0.510 0.350 4.768*
Ford 2.980 0.285 0.096 0.167 0.067 0.176 2.500
F-Ratio 14.557(C) 7.612(C)
* Denotes significance at 0.05 level
-------
APPENDIX G - THE LOG NORMAL DISTRIBUTION AS A STATISTICAL
MODEL FOR EXHACST E/'I'!ISSIONS
H. T. McAdams
1.
INTRODUCTION
The log-normal distribution has been indicated, empirically. to be
applicable to the statistical distribution of exhaust emissions. A foundation
for this observation is sought in order to substantiate the use of this model
on physical grounds.
First, the distribution and its properties are examined
and, second, assumptions which might give rise to the distribution in an
emissions context are explored.
2.
PROPERTIES OF TIJE LOG-NORMAL DISTRIBUTION
A random variable X is said to have a log-normal distribution if
log X has a normal distribution. Let Y = log X. Then X = eY and we want
to find the expected value and variance of eY.
Consider
"'" -.h ( ~"""'r'"
J e?- cr. e ~ J.}
-1>0
(1)
-L-
E [X] -= E [e.Yj:: trJ;:Tf
where E denotes expectation, and note that the moment-generatinp, function of
the random variable Y is
or:> k ( :J. ... .M\.) '1.. ).. ":l-
E fet';] ~ -L \ e - ~ 7 t} - "tM\f{t (j
() G"rr J... ~ . e - e ( 2)
Now when t = 1, (2) is formally equivalent to (1), so that
E [X] -
e h'fI + Js: a- ')..
( 3)
G-l
-------
where m is what is sometimes called (incorrectly) the "lo? population mean"
and ~ is what is similarly called "log population standard deviation."
Similarly, by definition of the variance of a random variable
VO-([x]-:::-
E fj'"J - i E [x]} Z
But
-L f -~ ( ~ -2) ~ .. d
E [;< '-1 ~ E [e ~ Y] ~ e ~,e cJ 4.
6"' Virr -00 cf
(4)
which is formally equivalent to (2) when t = 2.
Therefore,
E [x. 'J. J
=-
). """ + ~ () 'J-
e
and
[ED<]}1-=
'1-
e ~I'W"\. + U
so that
VO-y [X] =-
e
7L """' .,. ~ 0'"" l...
').1YI'\f-o-
e
( 5)
Manipulation of the results of (3) and (5) gives rise to some
interest in? relationships between the mean and standard deviation of the loe-
normal distribution and its percentile points.
rewrite (3) and (5) as
Let us, for convenience,
}C
e
"YY\ +- k. u '-
(6 )
G-2
-------
and
./.l..~::
)...
").. Nt'"\. + )... iJ
e
.l-
e ?............ + iJ
( 7)
where
x
denotes the mean and
/-1 denotes the standard deviation of
the random variable X.
Then
X~f-4'J....
e
~ "",,+
6"'2.-
t-
~IY"'\ + ;J...() '1-
e.
- e
~"""'-rcJ'-
e ').. """ t- ~cr ~
or
:f 1- 1"" /,l.- 1....
(j'-
e
.....
e ?. """- r 0
-~
X
0"),...
e
x)...
of<..
h()'- \I~~+~'L
t?:::. VA-
Then
d'-"-
e
X ? +- /l. :I.-
( 8)
x
Also, from (6),
el'>"'\..
x
e y~ {) .....
(9)
Combining (8) and (9), one obtains
ell""\. '"
X
V;i1-+4;l..
=-
i?--
FJz;..+-a....
(10)
x
But
m
is the mean logarithm and, since the logarithms are normally distributed
m is also the median logarithm. If 50% of the logarithms are less than m,
m . h m
then 50% of the antilogarithms are less than e -- In sort, e is the
median
X50
of the log-normal distribution.
Therefore,
XSO
~':I--
V j':J- -f-- ~ ~
(11 )
G- )
-------
Let us now consider the 84th Percentile
x84 .
space, this is the value m + cr: or, in antilog space,
e h'\+ Cf'
In the log-transform
e'-'"
eO
etr'
'tso
(11)
From (8) we have, taking lo~arithms of both sides,
()
u
or
e()
)..
( f~;:-~)
.-k
~~(P;:LJ
/ ~ ( X ~; ~ <-)
e
( 12)
Then, coml>ininp (11) and (12), one has
Y8'f
/YY\ 1- ~
e -=. X,fa
e. 0 £.- l P:~ L)
Kv similar reasoning it can be shown that
/'h-.,-O ~So
fife, -::. e - eff..- (X {~~)
(13)
OT l xSCJ) 1
111~ ~ -' SD
,x ilor-
This supgests that Dercentiles can be computed as sinmle multinles of the
rned i an ,
G-4
-------
Further generalization is, in fact, straightforward.
For, consider
a real number
k
and the quantity
m + kc:r.
From tables of the cumulative
distribution function for a normal distribution, a value of k can be
determined for any desired percentile.
Then, in antilog space,
e /W\ t-h()
=
e IY"'-. e.h. 0"'
defines exactly the same percentile
or
E M)-f~cr
"J So .
~(j'
e
1-50 . l t ~) ~
Thus it is seen that the antiloQ of (), the log standard deviation, provides
a multiplying factor applicable to the median in exactly the same way that an
additive factor is applied to the mean of a normal distribution. In practice,
computations are simpler if the quantities
m
and () are computed by trans-
forming observations to their logarithmic equivalents.
3.
ORIGIN OF TIlE LOG-NORMAL DISTRIBUTION
If emissions measurements tend to follow a log-normal distribution,
it is of interest to ask why this relationship holds. If a logical basis for
the log-normal distribution can be established, considerable light may be shed
on the nature of error pronagation, the requirements to be satisfied in complying
wi th emission standards, hardware requirements needed to answer questions of
technical feasibility, and the form that test plans should assume.
G-5
-------
A statistical process may tend toward producing a log-normal
distribution if the followinp, conditions are met:
al
Several sources of error or variability combine in
generating the random variable under study.
b)
Thes~ sources combine multiplicativelY rather than additively.
wJ.t,ur, :rc,'~I:,f]able lill.its, the form of the statistical distribution of the
source variables is iTrelevant.
To justi fy the above hypothesis, at least heuristically, consider
the central limit theorem of mathematical statistics.
Though this theorem can
he exnrf-ssed in various ways, the following statement suffices for present
purposes, If an arhitrary population distribution has mean fJ'- and finite variance
2
.() , then '[he distrihution of the sample mean approaches the normal distribution
with mean fA. and variance {/2/n as the sample size n increases. Sometimes
1 t is saId Hlat the distribution of the sample means tends asy~totically to a
normal distribution,
Ilow larre
n
has to be in order to annroach closely to
a normal di~lrihution depends on the form of the distribution of the variahles
bein\" comhin'2d.
As has been shown in experience with statistical quality control
concepts, normality can be closely approached if
n
is as small as 3 or 4, even
if the distributions of the original distributions depart markedly from a normal
dist rinut ion.
Now consider the case of n sources of error--that is, consider n random
variables
,\ I' X~,
.. , X .
n
PE:rform the logari thmic transformation
log X
log Xl
+
or' X2
+ . . . + log Xn
or
Y
Yl
+ Y"
+...+ Y
n
where
{.
1
J.o,:: :...
1
G-6
-------
Then, if the central limit theorem applies, Y will tend asymntotically toward a
normal distribution, and this fact implies that in antilog space X will tend
toward a log-normal distribut ion.
Is there any reason to helieve that the sources of variability in
automotive exhaust emissions should combine multiplicatively? Perhaps
there is. Consider, first of all, errors of measurement in determining the
concentration of a particular pollutant in the exhaust stream.
Let the
error involved in measuring the concentration C be measured by its standard
deviation
~
There is reason to believe that
6"'c.
is pronortional to C--
cr'ic
that is, that the relative standard deviation or coefficient of variation
is constant for a particular effluent and a particular measurement situation.
Now, consider a particular engine and let its emissions be measured
repeatedly by the process under consideration. The actual concentration C will
vary from time to time, due to differences in ambient environmental conditions,
condition of linkages and engine adjustments, type of fuel, operator-induced
variations and the like. It will follow, therefore, that ""'c. will also vary
proportionally to C and that the net result will be analogous to the multiplication
of two sources of error.
Let us further consider the relationship which obtains between
repeated tests of the same automobile and variation among many automobiles in a
fleet or in a production output. It might, perhaps, be argued that if a part-
icular automobile has a high mean level of emissions, that same automobile may
tend to have highly-variable emissions. Conversely, an automobile having a
relatively low mean level of emissions may tend to have relatively constant
emissions.
A moment's reflection will show that these statements must, to some
extent at least, be true.
Since emissions are bounded on the low side of the
scale (they can not be less than zero), it is impossible for individual emission
measurements to range upward beyond a certain level and yet maintain a low
mean, unless the high measurements are offset by a large number of measurements
close to zero.
This bounding process will tend to limit the magnitude of the
standard deviation and, at the same time, may tend to produce a skewed dis-
tribution which, itself, may resemble a log-normal distribution.
G-7
-------
Now consider the combination of vehicle-to-vehicle variability, time-
to-time variability within a given vehicle, and test-to-test variability at a
given time. If test-to-test variability is proportional to the mean level of the
particular automobile at a given time, and time-to-time variability is proportional
to the mean level for the particular vehicle, then the three sources of
variability would combine in a multiplicatively way. The result, by virtue of the
central limit theorem, would be a log-normal distribution.
4.
IMPLICATIONS OF STATISTICAL TREATMENT OF EMISSION ~IEASlJREMENTS
is clear.
The implications for statistical analysis of emission measurements
Whereas in normally distributed data our concern is with two quantities,
the mean and standard deviation of the measurements, }lere we are concerned with
m and ()
, the mean and standard deviation of the lo~arithms of the measure-
ments.
If one reasons in antilog space, the correspondinp quantities are
m
e
the median of the measurements
and
(J'
e
the "ratio standard deviation".
Th . 0"
e quantIty e is a fundamental quantity which defines the dispersion of the
results and which, together with the median, can be used to define any per-
centile of the distribution.
Thus the median and the ratio standard deviation
play the role that the mean ar.d simple standard deviation usually play. Note
that the use of a ratio standard deviation applied to the median does not differ
radically from the notion of a constant coefficient of variation applied to
the mean.
If is of interest, in fact, to compare the ratio standard deviation
with the coefficient of variation and to provide a means for convertin~ from
one to the other.
The coefficient of variation is
G-8
-------
~
.--:::=
"
Le_:J.'-""-/- ~ cr:. e;" """,of cr ~
e... """"- ~ .v~ !5"' ......
e
=-.""'" +;1.. cr"'"
?.. """1 .., () .......
e
e ~,.,..,.....+ 6""'''''''
v o-~
L -
(14 )
A more usable approximation is provided by expanding
accentable number of terms:
0"""
e to an
'1...-
eu
It {)'J- r-
cf f/
/ -;.... 1-
For small
.....
0- , it suffices to retain only the first two terms, so that
A..-
-::
.]-
'""'-
"'-
I ~
V )1-0 -/
-u
Then
e U
~-
t ~
-----
........,
I-+-
/.Lli
or
Ratio standard deviation
"-
"-
1 + coefficient of variation
For example, if the coefficient of variation is 0.2, the ratio standard
deviation is approximately 1.2.
yield
An exact computation according to (14) would
G-9
-------
cr?- (0.2)2
e 1 =
e 6-:1- 1.04
U~ £.n, j. r/f = 0. () 3' 2.1.-
-j--' .192
G
or
ecJ 1.212
Thus, for approximation, the ratio standard deviation can be estimated by
adding 1 to the coefficient of variation.
Further implication of the analysis is th3t, in analysis of variance
aimed at estimating sources of variahility, the emissions data can he subjected
first to logarithmic transformation. Analysis of variance performed in log
snace would yield components of variance combinin~ multiplicatively, according
to the postulated basis of the log-normal distribution of emissions.
G-IO
-------
APPENDIX H
EXPERIMEh'T DESIGN CONSIDERATIONS FOR EMIS-
SIONS TESTING OF STRATIFIED-CHARGE ENGINES
H. T McAdams
1.
INTRODUCTION
A test program to evaluate emissions for stratified charge engines
',iust take into consideration Ii number of variables which can affect the
results.
Variables which have been recognized by EPA as being important to
are listed in Table I.
,-mi $ S ions
TABLE r
Test Variables
<>
"
Engine Supplier (Ford/Texaco)
Transmission/Drive (4 Whee] Stick/2 Wheel Auto)
Engine Tuning (Max. Fuel Economy/~lin,Emissions)
Catalyst (None/Platinum/Palladium)
EGR (None/A Little/A Lot)
Aspiration (Natural/Turbo)
Fuel (Gasoline/CITE/Diesel)
o Mileage
o Inertia
" Road Load
" No. Repl ications
o No. Operators
"
"
"
"
t,C!n<:\Lcd":n C~' the38 variables, as well as the postulation and discu~sion of
c' ;:~~F~:r:C ~;,'!' ,~::: of variabil it;', is cc'nside-red to be in order before .'oceeding
'c'; "- iormu12pdon of tee::' programs for either available hardware or for hardwlFe
to bot pn':.lcul',::d a.nd test"d in a latter time frame,
In previous prescntations &nd docu~ents, it has been shown that two
1!\lljOT sources of variability &fft-ct (~mi5:sions as detemined from an emissions
test"
In one source are all thos~ f:ft.c 1:');1'5 which affect the measurement process
itself: the test driving cyc10, the opera(oI', the instrumentation accuracy
@.nd precision, the ambient 1.!!'IviroY1ment'll conditions (temperatur.;::, humidity,
t'ltlOOspheric pressure). In the other sou!'c.~ 91"e mIl those factors whi ch affect
the engine and/or its p'i:!l"f",rmmHe: it:> basic design (e.g.. whether Ford SCE,
T0)'{tiCO SCE. conventional, etc.), the tr&nsmissioTi or drive I'Ilechanism to which
H.-I
-------
it is coupled, the load to which it is subjected, the age and condition of
the engine, the type of fuel used, the ,manner in which the engine is tuned,
and the number and type of emission-control devices employed. In the light
of the many sources of variabi Ii ty, which the above catalog by no means
exhausts, it becomes very difficult if not impossible to answer the general
question, "What numbers, in terms of - say - grams pollutant per mile,
characterize the stratified charge engine?" The question clearly has little meaning
until it has been adequately qualified by specification of such key items as
type of test employed, fuel used, and so on. On the other hand, complete
particularization of all the factors which can affect emission measurements is
neither possible nor desirable. Rather, certain of these effects are more
appropriately addressed by randomization and statistical aggregatio~, An
important decision prior to the formulation of a testing program is to decide
which variables are to be controlled and which are to be randomized.
The distinction between fixed and random effects can be better
appreciated by considering the following two mathematical formulations or models.
MODEL I:
XLJ
=
fA-
-f d
J
t [,j
cJ,.
J
. th
fixed effect for the J column
t . rv
" J
N (0) c)~)
MODEL II:
.x ( J
~ +- a.. j + t~j
a.. j IiJ N (0') ~ ~ )
t.,ii AI N(O) cr?-)
In these models, it is assumed that data subject to two-way classification is
available and that this data has been arranged in rows (indexed by i) and columns
(indexed by j). In Model I, the "fixed-effects" model, it is assumed that all
d '1 to the J.th column 'b f' d dd' .
ata 'u e onging contn utes a lxe a 1 tl ve quantity 0( ,
, th f h . th J
to the value Ji'J listed in the 1 row 0 t e J column. Furthermore,
,th
it is assumed that it is inappropriate to consider the J
constraint, whatever
.th
it might be, as belonging to a statistical population. In short, the J
constraint is not imposed by chance but by intelligence and control and is
H-2
-------
1X'%1p~fitable with negligible error. In Model II, the "random-effects" model,
i% is assumed that the differences which exist among columns are induced by
Tnndom rather than by conscious choice and that there is no direct way by
which the contribution a. can be exactly reproduced in a subsequent
J
~xperiment. In short, a. is an outcome or realization of a random variable
J ~
A assumed to be normally distributed with mean 0 and variance CTA . It is
meaningless to attempt to reproduce the results for say column 3, because
either there is no available basis for choice or the basis for choice has
been relinquished in the interest of simplifying the problem under consideration
This being the case, a new choice for column 3 will differ from the original
choice in the same way as do the other columns in the array.
An example of ~ach of these models in the context of emissions testing
~ill h~lp to clarify their implications.
Suppose, for example, that a particular vehicle has been tested several
times ~t each of three inertial weights: 2000 pounds, 2500 pounds and 3000
pounds.
These three sets of tests constitute the columns, and the entries
within the columns constitute replicate test runs. Variation within the
columns result from random causes affecting the testing and measurement
operation and can be considered to be normally distributed with mean 0 and
variance ();).. The contribution 01-, of the first column, however, is
peculiar to an inertial weight of 2000 pounds. If the test is abandoned
and later returned to, it should be possible to obtain virtually the same ~I
simply by setting up the experiment so as to provide an inertial weight of 2000
pounds. Furthermore, it makes no sense to consider 2000 pounds as the outcome
of a random variable so long as it is practicable to treat it as an assignable
cause under the control of the experimenter.
Consider, now, a situation in which inertial weight is fixed at a
particular value- say 2500 pounds but three vehicles are obtained from the
manufacturer for test. There is nothing about the three vehicles that dis-
tinguishes one from the other or from other vehicles which might still be in the
manufacturer's inventory. Nevertheless, upon being tested, the vehicles
exhibit distinct differences in emissions, as observed by differences among
H-3
-------
the column means. Though a value &.
J
three vehicles. it is meaningless to at tempt to select a fourth veh lcle
duplicating either aI' a2' or Bl3 because there is no basis on which such a
selection can be made. Nevertheless, we can assume that the variation
is uniquely ascribahle to each of c]-ie
among the a.(j = 1,2,3) reflects a
J
of vehicles, in that it provides an
statistical distribution associated
sta!JJe property of subsequent choices
~
estimate of ~ ,the variance; of the
with the vehicle population.
It is hurriedly pointed out that the above examples are oversimplified
and that the distinction between fixed and random effects is somewhat
chimerical.
In the real world, vehicles may 'ell exhibit a spectrum of
inertial weights and it may be appropriate to characterize these YI(,ight
variations as a statistical distribution.
1ne wei~ht variations could accord-
ingly induce a statistical distribution of emis:' ions, as sho'...n in Figure 1.
Weight~ I m
Distribution 4~' /
Input ~ ~. ~
We i gh t
-~;;:;.~,
~ Emission
~,,--~~, I ~' .t .
~, lstrl1utlo~
./ Output --
.j
FI r,URE 1
RANDOM EFFECTS AS AN INDUCED DISTRIBUTION
Testing procedures and the attendant data rrocessinp, will differ appreciablY
depending on whether it is desired to qualify vehicles at a fixed weight or to
qualify vehicles collectively according to the prevailinp, frequency distrihutio0
of weights. Similar questions arise in tb:; structuring of other variables
which affect emissions measurements, either th~ougr the ph/sical attributes of
the vehicle and its performance capabilities 0, through variables affecting
the test operation. For example, vehicle operat:ors or inter-loooratoyy
differences in instrumentation may \veU c)ntrihutc di fferences in emissions
measurements.
Is it better to aH.,,"mpt to "standardize" these influences OT to
aggregate them statistically &mJ J.Hol'I ',:h'~ir jnnuence to he felt as stati~;ti'~ai
"orrors"?
H~4
-------
Though it is a point of some degree of subtlety, it will be evident that
the emissions distribution arising from Figure I differs in an important
respect from the distribution of errors [,J in models I and I I. It is
axiomatic that the E'J are induced in a manner analogous to Figure I,
but in the case of the C:'J the underlying distribution of causes is
unknown and not determinable. Consequently, if we are to estimate the
distribution of the
E:, j
, we must measure the distribution at the output.
On the other hand, there are variables such as inertial weight wh ich can be
varied parametrically so as to determine the input-output relationship.
Once this relationship is known, it can be combined with the input distribution
to compute the output distribution. One form which such a "computation" can
take is to employ a set of inputs having weights or frequencies which approxi-
mate the input distribution and actually to "phys ically compute" the response
distribution by performing an "aggregated experiment". This approach, in
effect, is the one followed in the use of the California driving cycle as a
basis for emissions tests. An alternative approach might be to isolate various
operating "modes" which make up the schedule and to determine the corresponding
times in mode. If the emission rates for each mode were determined by test,
the emissions for the entire cycle could be approximat('d by weighting each
mode according
to its corresponding time in mode.
The advantage of such an
approach is that it would not be constrained to a single driving schedule;
rather, one could compute the effect of vehicle emissions on the atmospheric
h!. ~ utant burden according to a variety of assumptions about "typical"
driving experience.
In summary, two general approaches to aggregated assessment of automotive
emissions have been delineated, each implying a different approach to the layout
of emissions experiments associated with development of the stratified-charge
engine. In one, we endeavor to put together a "representative" set of input
conditions and determine the corresponding output set of emissions without
attempting to define cause-effect relations on a one-to-one basis.
In th i s
approach, the "transfer function" of the "black box" in Figure I is ignored
on the basis that it is unnecessary to know the effect of specific parameters on
emissions so long as their effects are aggregated in accordance with prevailing
H-5
-------
or realistic aggregation of causes. In the other approach, we determine the
input-output relationships. Then, given any input spectrum we can estimate
the output spectrum. For some aspects of the development program, the first may be
desirable or necessary; for other aspects of the program the second approach
may be preferred. The appropriate tradeoff between the two approaches will be
borne in mind in subsequent sections of this discussion as they pertain to
evolving a plan for engine emissions assessment. A general guideline is that
relatively small effects can afford to be statistically aggregated and the
factors which induced them relegated to the !irrbo of "unassignable causes."
Relatively large effects, on the other hand, should be traced to their causes,
in the interest of both better repeatability of test results and engineering
exploitation of the cause-effect relation to minimize emissions.
2.
RANDOM FACTORS AFFECTING EMISSIONS
Factors which can appropriately be treated as random contributions in a
statistical model often occur as "nested" or "hierarchical" components. For
example, in describing an experiment aimed at determining calcium concentration
in turnip greens, Snedecor examined replicate chemical analysis within a single
leaf, variation among leaves of the same plant, and variation among plants.
In the same way that determinations are nested in leaves and leaves are nested
in plants, we can consider replicate emission tests conducted by the same operator,
differences among operators in the same laboratory, and differences between
laboratories. At each level of the hierarchy it is possible for errors to be
introduced, so that a general model might be
x,'j~
fA- + d.., -t- .-t,j +- [,i.~~
{.
oJ. , -::. fixed effect f. th laboratory
, 0 1
l..r" N N ( 0) 68 c)
'J
f Ai N ( 0 ") (J :I.)
where
H-6
-------
It is postulated that f.. 1- ,the test-to-test variability within a
&J-a
single operator and laboratory is a random variable and that b. ., the
1)
operator component, can be considered a random variable. On the other hand,
each laboratory might bring a particular bias to the determination (as a
result of instrumentation differences or other causes), so that the laboratory
contribution
0/.'
~
can be regarded as a fixed effect.
This model is
accordingly a "mixed model" in the sense that some components of the model are
fixed while others are regarded as outcomes of random variables.
How can variation attributable to laboratory-to-Iaboratory and operator-
to-operator sources be controlled? Laboratories are often brought into agreement
by a system of interlaboratory comparisons, and operator differences can be
"averaged out" by replicating the test a number of times, each time with a
different operator. Obviously these techniques are expensive and time con-
suming, and other, more economical approaches are desired.
An apparently straightforward approach to this problem is by the use of
a control sample. This control might take the form of a conventional or
modified engine the emissions of which have been well established. In any
test series, this "standard engine" could be included as a reference point. If
a particular operator or laboratory tends to "run high" or "run low" on this
standard, test resul ts can be adjusted accordingly.
Though the use of a standard or control engine appears attractive, the
approach is not without problems. Engines, even conventional ones, differ
among themselves and will operate differently from time to time for reasons
other than those associated with operator response and laboratory-dependent
causes. For that reason, the concept of a standard or control engine provides
something less than a "stable platform" on which to base results. An approach
which should be investigated is the use of an "internal standard" slaved
to the system under test, as shown in Figure 2.
H-7
-------
TEST
ENGINE
TEST
PLAN
CONTROL
AND
TEST
SYSTEM
COMPARE
ADJUSTED
EMISSIONS
INTERNAL
STANDARD
DE VI CE
FI GURE 2
STfu~DARDIZATION OF EMISSION TESTS
The internal standard device could be an engine subjected to the same controls
as the test engine, but more appropriately it might be an 'engine simulator'
designed to 'emit' pollutants such as CO and hydrocarhons under known and
precise control in accordance with operator demand. These effluents could be
precisely metered, so that an engine simulator in one laboratory could be
expected to perform very closely like one in another laboratory. Differences
from one operator to another would be registered as differences in output of the
standard device and, by comparison of the test engine output with the simulator
output, an emission measurement adjusted for operator idiosyncrasies could be
achieved. It is even feasible that the simulator could he entirely mathematical
a device which integrates operator demands in such a way that the standard
output could be computed rather than being physically generated.
H-8
-------
3.
NJN-RANOOM FACroRS AFFECTING EMISSIONS
Uncontrollable variables which inadvertently enter an emissions test and
make exact repeatability impossible are clearly to be regarded as random effects.
Operator variables, it has been seen, are of such a nature that they might be
treated as either fixed or random effects, depending on whether it is more
practical to isolate operator idiosyncrasies and adjust for them or to average
them out by statistical aggregation via replication.
Laboratory differences,
on the other hand, are often rather clearly of a fixed nature, constituting a bias
in resul ts.
Similarly, it has been seen that influences which might seem to
be clearly of a fixed variety. such as the effect of inertial weight, can be
couched in a statistical framework as an induced statistical distribution.
Thus it is apparent that any discussion of non-random factors affecting emissions
must be interpreted in a non-absolute way.
Though there are a numher of variahles
which "usually" are to he regarded deterministically. practically all of them are
capable of statistical aggregation under suitably defined ground rules.
Perhaps the most clear-cut case of a fixed or non-random factor is the
set of design elements which differentiates the Ford from the Texaco stratified-
charge engine.
This observation is not meant to imply a large engineering
difference in the two concepts but rather to emphasize that we are dealing with a
dichotomy
and that if we are to consider the two concepts statistically they
constitute two elements which exhaust the entire population.
It is clear,
therefore, that a variable labeled "make of engine" and having two "levels",
Ford .and Texaco, provides a starting point for an exp
-------
Next, how shall the engines be tested? In a military or civilian
vehicle, with standard or automatic transmissions, or by means of an engine
dynamometer? What catalyst shall be used, what fuel(s), and what basis for
EGR? These and many other variables have been recogni zed by EPA as candidates
for parametric variation to determine their corresponding "fixed effects" on
emissions. These major variables are listed in Table II.
TABLE II
MAJOR VARIABLES
:::;upplier: Ford Texl&c@
q~ransmij;
-------
3.1
Parametric Variables
First, let us consider that set of variables in Table II, which can
be considered as variables which we wish to explore parametrically to determine
their effect on emissions. These variables will be referred to as "treatment"
variables, and, if a level of each variable is specified, the resulting set of
constraints will be called a "treatment". For example, one treatment might be
the Ford engine with automatic drive, without catalyst, weight 3000, high EGR,
gasoline fuel and natural aspiration.
Let xl
Denote
Consider an experiment in n treatment variables xl' x2"" ,xn'
assume kl levels, x2 assume k2 levels, .,., and xn assume kn levels,
these sets of levels as:
S',
{ ;:,,) r-/~? 1/".;,
. . . , ~/-hl }
5 =
~
[ X".../) ;tl.~, .x;>..3:>
... )t~~-l
) .... ,)
5",-
[ 'X/f\/, X,.".?-, ;i"",- 3.)
""XMJv",~
Then the Cartesian product 51 x 52 x ... x 5n constitutes a factorial experiment
consisting of K = kl k2" .kn points in treatment space. That is, a factorial
experiment is made up of all combinations of levels in which one level is taken
from each factor, It can be considered as a startinp point for attempts to develop
an "econo'J1ized" version of an experimental program to adequately explore the
treatment space.
noted that
The desirability of such economization becomes evident when it is
2 suppliers x 2 transmissions x 2 catalysts
x 2 weights x 3 EGR options =
24 t reatmen ts
and that each treatment would require replication in order to assess random
effects or errors.
H-ll
-------
In order to appreciate the logic by which the above experiment can
be abbreviated, it is instructive to consider the matter of interaction between
variables. Suppose, for example, that it is possihle to express emissions in
terms of the equation
~ = ~ 0 1-0 c 1, ) 1--;.)
),x/V\) +-.-t,+I(X1)~"J")Jl_)+'" +- .J,-/.-f.!L"/I)(,..," '.JJ"\J
The function fo' fl' ,.. fk can be of any form but they are often considered as
polynomials of increasing degree in the x.. For example, consider a 2 x 2
1
factorial experiment (two factors each at two levels), often called a 22 factorial,
in the two variables xl and x2' Then it is customary to take the model as
';j -:::. 40 t ~I 'tl
1- ..{,. ~ x. '>- -r ~,... X I ;X; ::t-
where the subscripts of the regression coefficients are assigned to conform to the
variables involved.
The main effect of a factor is defined to be the averape effect of that
factor on the yield or response variable, where the average is taken over all
levels of the other
variables. In quantitative experiments, the main effect of
..c. =- ~ , in the event that there is
, () ;t.G
for the vanable
only one degree of
levels for the
.-tl, " '= 0 2. ~ / 0 X/-
x. corresponds to
1
freedom (2 levels)
X.. I f there are three
1
<",>,cror, we can compute an average quadratic main effect as
afc':: simi 1 arly for higher order terms.
A two-factor interaction (or first-order interaction) tells how the
effect of one factor depends on the levels of one or more other factors. For
example, in
~~
.-to + -t-, ;t,
1- .t- j.. 1- '- t- ,,(, I}. X:, ~ z.-
~Ie have
~Jhich clearly
~ 'i- -=
() I,
shows t ha t
-t,t- );I;''J~
the effect produced by xl on ';f may be modified by the
value assumed by x2' TItUS,
3- (~) ::: t,'J-
OX~
H-12
-------
measures the interaction between the treatment variables xl and x2' If
there are more than two fact~; in the experiment, the (xl,x2)-interaction
is the average val ue of %~I o,l~ , where the average is taken over
all levels of the other variables.
For example, in a 23 factorial experiment, with the model
~:: .-t,,-+.t,it, + --t~~1--+1r3:X3 rh/~X,J.1- t-,tdr.')(1+./r~3Y-;X3-fk,,")~,.1~)(3
we have
~I ~
-t I +.t,;. X ~ 1- -£1) X 3 + -(.,,.. J X,. X J
and
~ -
7J"f, O)C)- -
,t I ~ r -tl '" J r- 3
Then the first-order interaction of
xl and x2 is the value obtained when
Z> ').. ~ / ?; X, ~ I 'J..
is evaluated at all levels of x3
and averaged.
Upon further
differentiation, one obtains
L ( 2)?-'J. \
t)X~ L '(;1, '01:w) :: ~ I). 3
which, in this case, measures the three-factor interaction (second-order inter-
action) of xl,x2' and x3'
action of "1-1 and )t ....
It shows the extent to which the first order inter-
depends on the level of Xj .
I f there are more than
two levels in the treatment variables, we can have, in addition to the linear x
linear interaction above, such interactions as
quadratic (and so on).
linear x quadratic, and quadratic x
A factorial experiment is designed to estimate all interactions, up to
the maximum degree possible within the scope of the design. It may often be
evident, however, from engineering analysis and experience that certain inter-
actions are unlikely to be of any significant magnitude. In such cases, the
experiment can be redesigned on the assumption that the interactions in question
are non-existent. This approach allows simplification of the experiment and the
H-13
-------
elimination of certain treatments, the result being a more economical design.
A typical example is provided by the class of so-called "fractional factorial"
designs, in which only a fraction-- say 1/2 or 1/4-- of all the treatments
in the full factorial are employed.
An alternative approach is provided by
designs of the type developed by G.f:.r. Rox and his colleagues, called response-
surface designs. These designs are developed so that interactions of little
importance are ignored, but those likely to be important are not biased by the
simplification of the design.
It is proposed to employ these principles, although not necessarily
formally, in the formulation of a testing program for strati fied charge engines.
For example, consider the transmission/drive variable and such variables as
catalyst and EGR.
If it can be postulated that the transmission/drive variahle
does not interact with the catalyst or EGR variables, then it would be unnecessary
to run through the gamut of EGR and catalyst in both automatic and non-automatic
transmission modes. Rather, it might suffIce to run through the gamut of these
variables in either automatic or non-automatic mode and obtain only a single or
"bench-mark" point in the other mode.
Note that the assumed absence of inter-
action implies parallel responses in the two transmission modes but allows for
the possibility that the response in one mode might be displaced from the response
in the other mode, as shown in Figure 3.
Gm. /'1'
1.
o Points eliminated hy the
assumption of no inter-
~~~- .,'ion
~=-8--- Tran'Oi"ion Mode]
I p-- Transmission ~iode 2
None Design High
EGR
Emissions
FI GURE 3
ILLUSTRATION OF LACK OF INTERACTION
H-l..Jo
-------
Thus two of the required six test points could be eliminated without seriously
jeopardizing the validity of conclusions which mi~ht be drawn from the testing
program. If one pushes this approach even further, it will be apparent that
many variations in engine operating variables could be explored by tests on an
engine dynamometer, especially if it could be shown that the vehicle configuration
environment tends to introduce only differences of the "bench-mark" variety.
3.2
Variables Amenable to Statistical Aggregation
Those variables to be considered for statistical aggregation are those
which are associated with engine "personality." These include, for example,
minor deviations in dimensional tolerances, day-to-day vagaries of engine perform-
ance, and the like. Admittedly, the effects of many of these variables could
be studied by parametric variation.
For example, one could deliberately
introduce dimensional variations and observe their consequences on emissions.
It is considered more practical, however, to undertake statistical aggregation
by testing a large enough sample to allow statistical variation to come into
play.
Consider a collection of N
engines of a particular class and let these
engines be tested repeatedly. It will be assumed that the test laboratory is
fixed, that the same operator is used, and that all other testing variables
which might introduce variation have been eliminated. Nevertheless, it will
be observed that successive tests of a particular engine do not give identical
results. It has been said that no one can step in the same river twice. In
the same vein, it can be said that no one can test the same engine twice. I t has
"aged" by some finite increment, if nothing else, and therefore may not perform
in exactly the same way as it did originally. As the tests continue, other
influences will occur which will also cause some degree of variation in test
results--quite apart, please note, from testing errors introduced hy instru-
mentation and operator effects. Some of these influences, such as the aging
effect noted above, may be capable of being treated deterministically. others
will need to be treated as random variables.
H-15
-------
In previous attempts to separate engine-to-engine variability from
test-to-test variability, it has been assumed that repeat tests performed on
the same vehicle provided a measure of such test-associated variation as that
arising from instrumentation sources, deviations from the driving schedule, etc.
It must be recognized, however, that there are within-engine sources of
variation which are associated with the engine itself: wear of parts,
idiosyncrasies of linkages, and the like. It is also quite conceivable that the
day-to-day or time-to-time differences of this nature within a particular
vehicle may be more pronounced than differences from one vehicle to another.
It is for this reason that repeated testing of a given vehicle over a considerable
period of time and with the vehicle in a range of normally encQuntered operatinp,
conditions is in order. A more realistic variance model is therefore
'to, c
, ~1\.
-
p.
+
a. +-~.t. + 6" ~-f.-
where
a~' is a contribution associated with the kth vehicle, bjk is a vehicle-
. th
associated contribution of the J test on that vehicle, and €. Ok is a test-
IJ
The question naturally arises as to whether the
associated error contribution.
error ~ ijk can be dissociated from the vehicle-associated contribution bjk or
whether these two sources of variability are confounded in such a way that they
can not be separated. We believe, however, that if replicate tests are run on
the same vehicle at as nearly the same time as possible, this procedure will
tend to eliminate time-to-time variations in the engine itself. On the other
hand, if the engine is tested over a period of time and conditions, again each
time in replicate, the two sources of variability can be, in effect, separated.
In assessing the bjk' one must consider whether these contributions
to the variability of Xij-h. are random or fixed effects. In some cases the
effects may be considered to be of the fixed variety. For example, in the case
of engine age or accumulated mileage, it is possible that a time trend in emISSIons
might be observed, especially if the vehicle is operating with a catalyst subject
to deterioration with time. Such trends could be extracted by means of regression
analysis and the remaining or residual variance treated as arising from random
causes.
H-16
-------
4.
TESTS ON AVAILABLE HARDWARE
The discussion of the previous sections has been reparded as necessary
considerations prerequisite to the formulation of a testing nrorram for
stratified charge engines. Most important is a careful weighinp of what
types of testing are appropriate to existing hardware and what types of
testing should be reserved for new hardware arising from the development
program .
In this connection, it must be acknowledged that with only a few
vehicles available, it is virtually impossible to do much in assessing the
engine-to-engine variability which arises from manufacturinp variations or
associated causes.
It is, however, quite logical to employ these few
vehicles in a program aimed at defining the ran~e of emissions which can
occur from variations within the enpine itself. For example, a program
aimed at mileare accumulation coupled with routine maintenace and adjustment
could do much to define the range within which. specific enpine mipht he
expected to operate, as far as emissions are concerned.
It is also con-
ceivable that parametric variation of major variables could be undertaken
to define emission response to these variables. If the latter course is
followed, however, it might be well to ask whether that type of study is
premature, in view of the fact that design concepts are still in a state of
flux.
We recommend, therefore, that parametric tests be undertaken with a
view toward defining Drimarily the direction of trends rather than attempting
to measure with high quantitative accuracy the actual magnitude of these
trends. With this thought in mind, we consider it adequate to assess the
effect of such parameters as inertial weight, EGR and catalyst only at the
beginning and end of the mileage accumulation program. Intermediate tests
could be performed on some single'bench mark treatment', such as 2000 lbs.
weight, design EGR, and standard catalyst. This single confip,uration would
make it possible to follow time trends in emissions which mir.ht be caused
by engine deterioration.
Though it is possible that there is interaction
between treatment configuration and mileage, an assessment of these inter-
actions would be available from the parametric tests to be performed at the
H -17
-------
beginning and end of the mileage-accumulation program. The approach would
h~ve the advanta?e that intermediate tests could be expedited, since no
variations in test setup would be required. In this way, the time taken
away from mileage accumulation would be minimized and yet the loss in
parametric information would be minimal. Though intermediate tests might
serve to strenp,then confidence in the effect of treatment parameters on
emissions, a similar strengthening of confidence could be attained by runninp,
additional replicates at the end of the mileage-accumulation program.
The testing
existing hardware,
program outlined in Table III is proposed for evaluating
specifically one 4-wheel drive Jeep equipped with Ford
(FCP) engine, one 4-wheel drive Jeep equipped with
Combustion Process
Texaco Combustion Process (TCP) engine, and one 2-wheel drive postal van,
equipped with FCP engine and automatic transmission. Bench-mark tests
of a standard Jeep equipped with a conventional engine are included as
points of reference against which the stratified-charge engines can be
compared. It is proposed that the standard Jeep tested in the initial
characterization be in substantially new condition and that the Jeep tested
in the final characterization be one with substantial (approximately 50,000)
mileage accumulation. These two tests need not be performed on the same
Jeep, since the intent is only to provide a basis for comparison of the
stratified charge engines with "typical" conventional vehicles. It is
proposed that each test be conducted in duplicate so as to provide a basis
for assessment of errors. Preferably the replicate tests should be per-
formed by di fferent operators drawn randomly from a pool of several operators
taken as representative of operators in general, so that operator variabilit)
will be reflected in the repeatability assessment.
repeatability might be unrealistically optimistic.
Otherwise, the measured
The test array of Table III is regarded as being "adaptive" in the
sense that it may be modified as the program progresses. For example, it
is indicated that mileage accumulation to 50,000 miles is rrojected with
emissions tests scheduled at approximately 10,000 mile intervals. In the
H-18
-------
TARLE II I
EMISSIONS TESTING PROGRAM
R:>R STRATI FlED CIIARGE ENGINES
:r:
I
......
-.D
JEEP FCP - JEEP TCP - POSTAL VAN TCP
CATALYST ~O CATALYST STANDARD
2000# 3000# 2000# 3000# JEEP
I nerti al I~ei gh t Inertial IVeight Inertial Weight Inert i al IVei gh t
No Desip:n II i gh No Des i gn Iligh No Design Iligh No Des i gn Iligh
EGR EGR EGR EGR EGR EGR EGR EGR EGR EGR EGR EGR
Initial II C II II C II II II II C fI II C
10,000 Mi. C
20,000 Mi. C
30, 000 ~1i. C C C
40,000 Mi. C
50,000 ~1i. C
Final II C I II II C II II II II C fI Ii C
Rejuvenated C C
II
C
= flot start only
= Complete cold-start test
-------
the event that changes either in performance or emissions are observed to
occur rapidly with increasing mileage, it may be desirable to alter either
the extent of the milea~e accumulation or the frequency of testing. In
short, it is contemplated that as information accumulates from the testing
progral1l, this information will be used as feedback to the experiment design.
The tc3t
array is not structured accordin~ to any particular maint-
In view of the lack of experience with stratified-charge
en ance s ciledu 1 e ,
EGgines, it is believed that to do so would be unrealistic.
Rather, it
is contemplated that whatever maintenance is required would be conducted
as r1<'Oci'ssary and that any effects introduced by such maintenance would be
treated either as fixed or random effects.
For example, if major or at
least definitive repairs were necessary, such as the changing of spark
plugs or replenishment of catalyst, these would be noted in the testing
log and possihly adjusted for as fixed effects in the data analysis. On
the other hand, such routine maintenance as tuning, adjustment of linkages,
etc., hfine less definitive, would be aggregated as additional statistical
variation or error 1n the test results.
A word of explanation is in order concerning the tests scheduled as
"Re juvenated." The intent here is to see how much of the observed degra-
dation in emission characteristics is of a reversible nature. It is proposed
to replace catalyst and spark plugs, retune the en~ines, and perform any
other simple maintenance which would "give the engine every benefit of
the doubt" as far as emissions are concerned. In short, one would endeavor
tc. see to what extent the engine could be IIhroupht back" to its initial
emIssions hy means which could be readily and simply implemented.
In prol)0sing the s .edule of complete (C) and hot-start-only (H)
emissions tests, 'de were governed by considerations of economy and the
exp;;rirnent-design rdncipl~s previously discussed.
In effect, the two
extensive series of tests desi~nated as Initial and Final can be considered
as t,,>u levels of a factorial experiment in which the v1'iriable in question
H-ZO
-------
is mileage and the levels are nominally 0 and 50,000.
Since these two levels
represent extremes, it is considered that the effects of other variahles,
such as inertial weight, EC,R and catalyst, may be different before and after
the mileage accumulation. In short, it is considered that these variahles
may interac t with the mil eage- accumulat ion vari ab 1 e.
Also, thou~h it is
realized that intermediate mileages may show similar effects, it is considered
adequate to monitor the development of these effects by a single "treatment",
3000# weight, design EGR wi th catalyst. This treatment is regard
-------
\PPENDIX I
NEW HARDWARE REQUIREMENTS FOR ASSESSING EXHAUST EMISSIONS PERFORMANCE
OF THE STRATIFIED CHARGE ENGINE A STATISTICAL CONSIDERATION
H. T. McAdams
One of the most important considerations in any hardware development
program is the question of how many units must be built and tested in order
to have "reasonable assurance" that the developmental item will perform
satisfac,torily.
The question is especially serious if fabrication costs
and time schedules limit the availability of units for testing purposes.
In
such instances, it is necessary to draw conclusions from a limited amount
of data but yet have some measure of the validity of those conclusions.
The nature of the question is at least twofold.
First, one must
define in quantitative terms what is meant by reasonable assurance.
Second,
there must be some inferential process by which the resul ts obtained from
a limited number of units can he projected to ,,,hat might be expected when
the experimental hardware is produced in larger quantities.
1.
CONFIDENCE INTERVALS
The above question is only a special case of one of the most persistent
prohlems in statistical inference: how much data is enough?
Because of the
statistical nature of the question, its answer is often expressed in terms of a
confidence interval.
Properly interpreted, the concept of confidence interval
is a useful one in answering the question of "lIow many observations are
needed to establish the mean value of a random variable to within certain
prescribed 1 imi ts and to a certain level of confidence?"
1-1
-------
However, since the implications of a confidence interval can be misinterpreted,
it is in order to discuss the concept briefly in the context of stratified
charge engine development.
Let us imagine that a very large supply of engines is available.
Suppose that N of th~se engines are selected at random, where ~ is a small
number relative to the total number availi:1ble, and that each engine in this small
sample is tested once for its emissions per mile.
From the N tests, the mean
or average value can be computed, as can also the standard deviation of the
individual measurements about this mean.
Since the N engines measured are
only a "sample" of the entire population of engines, a second set of N engInes
will not necessarily yield the same sample mean or sample standard deviation.
Moreover, neither of the sample means are likely to coincide with the population
mean, which could he determined by measuring emissions for "all" engines in the
population.
Similarly, the sample standard deviations would be unlikely to
coincide with the "true" or population standard deviation similarly ohtained hv
exhaustively measuring all engines in the population.
The concept of
"confidence inter"al" represents an attempt to invest a sample result with some
measure of "nearness" to the popul ation resul t.
For large values of N, for example,
it can be shown that the sample mean wi 11 tend to he "closer" to the popul ation
mean that! would be the case for small values of N.
By the reverse of this
reasoning pTOcess, one can postulate how large N should he in order to achieve
a certain deRree of "nearness" to the true or population mean. It is evident,
h';:J\'i\i;;VIBf, U1,n the que~L(jn of "HeM many tests are needed?" must be preceded hy
& question wh:j en ask:>" in essence p "How close to the truth is close enough?"
1-,.:':
-------
The mechanics of computing a confidence interval is strai~~t-forward.
For example, suppose that 10 engines are tested and that for each engine the
hydrocarbon emissions per mile is determined. Their mean is
x =
~D ('t, + ~~ + "
+ ~/o)
l;Llch£lr sample standard deviation* is
I/O
4.. " ~ (~, - i,)'
10
As a random variable,
;t
is an unbiased estimator of fL
, the
rOi'\:l at ion mean.
I'- ~
U
Similarly, II
.~ (XL -J(
L :: I
N - /
15
an unbiased estimator of the population variance.
Inasmuch as
1-
is a random variable, it has its own variance; the
Il1a;::,"litude of this variance is inversely proportional to N, the sample size.
N
for the variance of ~
N
> (X,-X)'"
~-
N{N-I)
is
Thus an unhiased estimator
,... 2.
U
--
and the square root of this quantity is the standard error of the sample
mean. By making certain assumptions regarding the statistical distrihution of );
one can compute the required confidence interval.
..
Th~ sample variance, defined as ..4.. 'R. -::
unbiased estimate of cr 2., the population
eliminated by using N-l in place of N.
N
II ~ ( _;1"-
/ N-, y,( 1") does not provide
L ~ I
standard 0cviation. The hias is
a 1'1
1- 3
-------
Confidence intervals rely on the assumption that the sample mean, as
a random
variable, oheys 3 normal or Gaussian distribution with some population
rY/tJ
where
~
cJ is the population variance of X
mean t- and variance
and N is the sample size..
If repeated samples are taken from the population,
and if
(J"''j..
is known, approximately 68% of the resulting values of
x
will
fall wi thin an interval hounded on the low side by /'A-. a-
and on the high
side by
?tu
Similarly, approximately 95% of the values of
x
will
lie in the intervalft- ;Uj to
ItA f ~ () and other percentares of the cases can
be bracketed hy appropriate choice of k in the quantities ft -...h 6" and ~ +-htf"'
which bound the interval.
Suppose, however, that only the result from a
particular sample is available. The sample quantities
x
""
and
tJ
provi de
estimates of ~ and
~ , and it is presumed that these values can he used to
construct a confidence interval
X-~6-
-------
A word of caution is in order, however.
The population mean is
a constant, not a random variable.
In reality. therefore, fA either is or is not
contained in the above interval.
It is the interval, as computed from the
sample data, which statistically varies;
both its midpoint and its Io'idth
depends on the particular values obtained in the sample. If repeated samples of
9 engines were tested and a confidence interval computed for each of these
samples, approximately 95% of the intervals so constructed would contain the
population mean ~ .
It is in this sense that the confidence interval should
be interpreted.
Thus, in the above example, if one asserts that fA falls in
the interval 0.27 to 0.37. such an assertion is "more likely" to he true than
false; in the sense indicated, the assertion has probabil ity 0.95 of beinr, a
true statement.
It will be noted that the width of the confideflce interval depends
on several factors: the level of confidence, the standard deviation of the
measurements, and N, the numlt.er of engines tested.
The level of confidence
is often set arbitrarily;
however, its choice c~ b~ mA0e rationally by
examining the consequences of exceeding the interval ho.Uft~s.
'The standard
deviation is determined in part by the variation which exists in,. one engine
to another and in part by the inability to repeat $ucc~sive emission measure-
ments exactly because of experimental errors.
Th e s -p I e size, N , is
determined by availability of engines and, more importantly perhaps, hy
procurement costs.
1-5
-------
'")
VARIANCE COMPO~F.NTS
As pointed out above, there are at least two sources of v~Tiance
which must be reckoned wi th in the assessment of automotive emissions:
b)
vari rtnc:\~
variance induced by differences among engines
induced hy testing errot'.
a)
Thus a model for emissions might be*:
~..
(J
where
fA--
fi.t
fl. ) =
J{ (
fA-
1- rx i f fl,Hd
contribution common to ~lll measurements
h .th
contribution peculiar to tel engine
. th f h . th
contribution peculiar to the J test 0 tel
engine
cJ'<
In any single test of an engine, the variance
.(
(J' can he expressed as
2.-
where ~
u:Z
t
cJ~
~
a:-""
f- t
component of variance attributable to engine-to-engine
variability
component of variance attrihutable to test-to-test
variahi 1 i ty.
A more r.eneral model,however, wouJd further delineate the identifiable
sources of testinr error, such as operator error; and the sources of
differences among enpines, such as production tolerances, adjustment3,
*
etc.
I~ 6
-------
The distinction is an important one because it has direct hearinr on the
strategy of any testing program aimed at predictinr the performance of enrines
as a population.
If the engine component
o-;?- is a small fraction of
(j2..
there is little need to sample extensively the engine population.
Con-
sequently, in this case, statistical needs could be satisfied by testing
only a few randomly selected engines.
It would be necessary to perform
a large numher of repeat, or replicate, tests on each engine, however,
Suppose, on the other hand, that
the testin~ variability would constitute
~
ere
~
the major fraction of ().
u'-:
Then
because
is a larr,e fraction of
a relatively larr,e sample of engines would be necessary, hut a few tests
on each would suffice hecause test-to-test variation would make only a small
O~.
contribution to the total variance
It must also he noted that if
(y~
~
is larr,e, no amount of replicate testing on a few enrines will
strenp,then our confidence suhstantially hecause it provides no ~echanism
for averaginr. di fferences among enrines, the fT1ain source of vari ah i 1 i ty.
1-7
-------
3.
SAMPLE SIZE RELATIONSHIPS
In many testinp, situations, the magnitude of the measurement errors
tends to he proportional to the mean value of the quantity heing measured.
Experience with both automotive and aircraft enpine emissions sur~ests that
this relationship applies, at least to good approximation, in the case of
gaseous emissions measurements.
It is convenient and pertinent, therefore,
to express the standard deviation of the measurements as a fraction of their
mean.
This ratio is often referred to as the coefficient of variation, or
relative standard deviation.
Let
;;(
denote the sample mean for a sample of N engines selected
at random, each enp,ine being tested once.
Suppose that the population mean
IS
f"v
and that the population standard deviation is
() =: "f~
where
p
is the coefficient of variation.
The standard error of the random
variahle
~
is accordingly
2:-
{N
PIjJN
If it is presumed that ()
is known, then a 95% confidence interval for
p.-
is p,iven hy
X-
- /. 90
f~/{N
~ F ~
~
-+--
/. ~ (.
f~/fN
This result anpears circuitous in that ~
estahlish the confidence interval, hut if it were known there would he no
would need to he known in order to
need for a confidence interval.
Suppose, however, that one considers only
the width of the interval, which is
1-8
-------
X-I-lf"
fljJN
(i - /. ';b 1>1fN)
31~ ?YViV
Then the width of the interval, expressed in terms of jU , is just
:3 9z
to }A/Vii
-~
3. 712 P
'iN
or the interval half-width is 1.96
r/fN .
This simple result can afford broad guidance in postulating the
number of tests required to establish a certain degree of confidence in
emissions measurements.
It should be noted that a similar arp,ument can be
developed for other levels of confidence, say ()9%.
The 95% confidence level
was selected, however. as the basis for further analysis.
Confidence
intervals based on the 95% level of probability provide a hip,h der.ree of
assurance that a particular interval will contain the population mean.
Beyond
95%, the width of the interval (in terms of multiples of the standard
deviation) increases rapidly for each additional 1% increase in confidence.
In Tahle 1 is tabulated the quantity 1.96 f>/fN
for various
values of p and N.
The parameter p is varied in 15% increments from 15%
to 75%;
the parameter N is varied from 1 to 100 in intervals which are
perfect squares.
According to the table, for data with precision such that the
coefficient of variation is 15%, approximately 9 tests would be required to
reduce the half-width of the 95% confidence interval to 10% of the population
me an, f- .
Similarly, if the coefficient of variation is 30%, 36 tests would
be required for the same purpose.
I f the coefficient of variation is as
large as 45%, 81 tests would be necessary to determi~e the mean value to
within 10% of its magnitude.
1-9
-------
r:.'\nORC; FOR rO"irTnrNrr TNTrp\,,,I.~
9St Confid~nce Rounds
- --- -- ~-------- -- - - - --- - - -
--~-------..
--- - - --- ..---!
I
I
I
I
I
Coeffici~nt of variation, p (\)
I -1:---1-3~ - -1 .. --4:--1-- '-6--: '-r--~;
:~ -:---~.~~4-11'.- -::8 --- i --8:. ~_u -1--;;;.~ - +- '1-4-;;:- :
I I '
I I '
t1 14.7 20.4 44.1 S8.8 B.S
S3fT1T'If' ') f).8 1~.6 2C) ..i 3C).2 4C).n
S i zp. ~ 1 h 7 4 14.7 22.n 2C1.4 36.8
, -,r S.!) 11.8 17.6 n.s 2Q.4
I ..,
, 'h 4.<1 0 .R 14.7 1°.6 ~.1 .5
,
I
I -1') 4.:' 7.7 1~.6 16.R ::'1.n
,
, (,.t 3. '7 7 t1 11.0 11. 7 1 R..1
I RI 3.3 ('. S C) .R 13.1 1(,.3
! I
: 1 O() ::>. () S.q R.R t ) 1 . R 1.1 - I
I i I
I I I
I I I I I I
I I I I I
I I j .u - --- J - - -_..- - J I
--- - --1--- -1-- -- --- J I
- --- -- - - - - - ---- --
TARLE 1
1-10
-------
4.
IMPLICATIONS FOR COMPLIANCE WInl EMISSION STANDARDS
The actual engineering requirements imposed on an enrine in
order to meet emissions standards is determined, in part, by the manner in
which compliance with the standard is to be interpreted.
To make this
point clear, we consider two cases representing different but reasonahle
interpretations of the standard.
4.1
C as e 1.
Compliance Based on a Single Test
Suppose that compliance is based on a single test of a single
vehicle.
Now obviously if the vehicle is to be passed on the basis of a
single test, the expected value or mean for that vehicle must he lower
than the nominal standard.
One can calculate how nuch lower hy assuminp
---
some confidence level
say 95%
at which to work.
Suppose, for a particular engine, the expected value for all tests
is fA- and the standard deviation is C).
Then, for that engine, if it is to
pass with 95% confidence on the basis of a single test, we must have
~ .. 1. 645
0-
.::-
0.41
~y
fJ-. 0-1-/
....... I. l::,'fJg---
5%
If
CJ'
p% of the mean,
f'c + 1.645 pt--
£
0.41
1-11
-------
and the upper bound is given by
fA-(1 +j. b4Sf) - 0, 4- /
~ 0.41
or, =
1 + 1. 64 5 P
where p is
the coefficient of variation.
One can assume various values for
p
and see the consequences.
(See Table 2).
4.2
Compliance Based on Expected Value
CASE 2
On the other hand, suppose that the expected value for the enr,iRe is not
to exceed the standard value 0.41.
Then the tolerance for an individual test
should be aU,gmented by 1.645 0-.
An estimate of CI would he required, but,
once known would prescribe that
Individual test
.,;..
0.41...l.645C>
Again, suppose that CI
PI-'-
Then,
0.41 t 1.645 r5' = 0.41 + 1.645
t?-
Hut we want ~ not to exceed 0.41.
Then,
o . 41 + 1. 645 1 fJ-
0.41 + 1.645p (0.41)
0.41 + 0.67445 P
where p
coefficient of variation.
This would lead to Table 3.
1-12
-------
TABLE 2
VALUES FOR
;u-
FOR INDIVIDUAL ENGINE RE0lJIRED m
PASS STA.."IDARt1 0.41 @ 95% CON FI DENCE
Coefficient of
Variation
Expected Value rOT
Engine
p
fA-
0.00 0.41
0.05 0.38
0.10 0..15
0.15 0.33
0.30 0.28
0.45 0.24
0.60 0.21
0.75 () . 1 8
.50
..
/
o. 4- /
1+ /. 64S?
.'fo
~ .~o
'~
lo
-........
~............... -
./()
l
,/5" .30 .ifS
P
,bo
.75
1-13
-------
TABLE 3
ALLOWABLE TEST CRITERION TO ASSURE 111AT P. ~
FOR VARIOUS VALUES OF r
0.41
Coefficient of ~1aximum A11owah1e
Variation Sinp:1e Test
r (ri terion
0.00 0.41
0.05 0.44
0.10 0.48
O. ] 5 0.5]
0.30 0.6]
0.45 0.71
0.60 0.1-\2
0.75 0.92
9Q
/
//'
/
/
.~o
.70
Test
Cri terion
. bar
.5°1
.1-) ~
1
o
/
'5 .30 .'!-)
.00
7.:-
. -'
p--
~ U.S. Government Printing Office: 1973--747-787/322 Region No.4
1-14
-------
It should be noted that the desired emissions level (say 0.41
grams per mile) and the basis on which an engine should pass are not
and need not he the same. An acceptance plan must reflect the desired
statistical risks and could be sinp,le sample, multiple sample, etc.
For example, if a single test does not pass, ~ retest could he allowed
under conditions adjusted to give a prescribed statistical risk. The
procedural hasis for various sampling plans is well established and could
provide a sound statistical basis for compliance testing.
5.
I~PLICATIONS FOR TECHNICAL FEASIBILITY OF A STRATIFIED CHARGE
LOW EMISSIONS ENGINE
It is seen that, whatever the basis of compliance with standards,
the expected values
of the emissions for a particular engine must he
known if one is to assess the technical feasihility of producing a low
emissions stratified charge enp,ine.
The term feasibility is internreted
to mean that it is not sufficient to produce a single engine capahle of
meeting emissions standards but that it should he possible to produce such
engines in quantity without incurring excessive failure rates.
component of variance,
One way to approach this prohlem is to assume that the testing
Z
Ot ' can he substantially eliminated from con-
sideration hy performing a sufficient numher of replicate tests on each
enrine.
This assumption presumes, therefore, that it will he possible to
determine the expected value of emission for each pollutant for each enp,ine
tested.
One must he concerned, therefore, only with engine-to-engine
variahility.
The prohlem is to determine what the expected emissions
values must he when expectation applies to th~ entire engine population, if
it is desired to reduce rejection rates to any prescrihed value.
1-15
-------
Suppose, for example, that we do not want more than 5~ of all
engines to exceed 0.41 rms/mile.
Then, if
oe
is the standard deviation
of enrine expected val ues about the population expected val ue, we must have
fA- + j, (,1-5 oe
~ CJ.4-1
or, if
~
p;4-
f-t-
0.41
---
I + 1.645 cr
and Table ~ applies (but this time to expected values for enrines, not to
test values for a particular enrine).
Th us i f p
0.15, the popu'ation
mean for all enrines must he ~ 0.33 to assure that expected values for
individual en~ines will no~ exceed 0.41.
\DII,
how many eneines must be tested to "predict" that the
population mean will not exceed a certain value -- say 0.33.
If we are to work at a value of 1'J to insure a confidence interval
f"
+ 109of-, we must decrease 0.33 by 10% to approximately 0.30.
The better we know the population mean, i.e. the greater N, the less
Jemandinr will the requirement on the estimated mean he.
It should further
he noted that the demands placed on engineerinr. also strongly depends
on
oe
or
more specifically. the coefficient of variation p obtained when
ere
is expressed as a fraction of the population mean for the emission
measurement under consideration.
If P is taken as 15%, and if it is
desired to hold to a confidence interval fA ~ 10% fA. , then the numher of
en~ines to be tested is approximately N = 9.
IfN
4, the half-width
of the interval roes to approximately 15%, and we would have to depress the
0.33 value not to 0.30 but to
0.33
(0.15) (0.33) = 0.2~.
1-16
-------
What constitutes a rood estimate of coefficient of variation is
problematical, since there is no direct statistical basis for assessment.
There is evidence to show that testinr, variation is such that
a; is
approximately 15% of the mean for a particular en~ine.
:\lso, in tests on
aircraft engines, it was observed that engine-to-engine variability often was
at least comparahle to test-to-test variability.
It is on the basis of
these very tenuous arruments that the above analysis was made assuminr a 15°0
coefficient of variation.
In conclusion, it appears that if one were to speak in terms of
orders of macrnitude for the number of engines to he tested, this number
should he at least 10.
Though it is suspected that this number is
conservative, it would at least provide a hasis for a revised estimate of
testine requirements.
If it should turn out that a larrer sample appears
necessary;
additional enpines could he procured and tested sequential to
the first sample lot.
On the other hand, if a larrer numher of enrines
were procured unnecessarily, much expense would have been incurred which
could not he written off to any substantial rain in information.
1-17
-------
ENVIRONMENTAL PROTECTION AGENCY
Technical Publications Branch
Office of Administration
Research Triangle Park. N C 27711
OFFICIAL BUSINESS
AN EQUAL OPPORTUNITY EMPLOYER
POSTAGE AND FEES PAID
ENVIRONMENTAL PROTECTION AGENCY
EPA - 335
SPECIAL FOURTH-CLASS RATE
BOOK
Return this sheet if you do NOT wish to receive this material CL,
or if change of address is needed uJZl. (Indicate change, including
ZIP code.)
------- |