EPA 600/3-83-104
                                                                     PB84-120369
      Effect  of Load Simulation  on
      Auto Emissions and Model Performance
       (U.S.)  Environmental Sciences Research Lab.
      Research Triangle Park, NC
      Nov 83
   U.S. Department of Commerce
   Katkmai Technical Information Service

-------
                                   TECHNICAL REPORT DATA
                           (Please read Instructions on :he reverse before completing)
1. REPORT NO.

 EPA-600/3-83-104
                                                           3. RE
4. TITLE ANP SUBTITLE
  EFFECT OF LOAD  SIMULATION ON AUTO EMISSIONS AND
  MODEL PERFORMANCE
                                                           5. REPORT DATE
                                                             November 1983
             6. PERFORMING ORGANIZATION CODE
7. AUTHORIS)
  P. Gabele and  R.  Snow
                                                           8. PERFORMING ORGANIZATION REPORT NO.
9. PERFORMING ORGANIZATION NAME AND ADDRESS
  Environmental  Sciences Research Laboratory
  Office of Research and Development
  U.S. Environmental Protection Agency
  Research Triangle  Park, NC 27711
             10. PROGRAM ELEMENT NO.

                 C9YA1C/04-3080 (FY-83)
             11. CONTRACT/GRANT NO.
12. SPONSORING AGENCY NAME AND ADDRESS
                                                           13. TYPE OF REPORT AND PERIOD COVERED
  Environmental  Sciences Research Laboratory - RTP, NC
  Office of  Research and Development
  U.S. Environmental Protection Agency
  Research Triangle  Park, NC 27711
             14. SPONSORING AGENCY CODE

                  EPA/600/09
15. SUPPLEMENTARY NO "ES
       the overall objective of  this  study was to identify sources which might contri-
   bute to errors in mobile source  emission rate model predictions.  The  effect of road
   load simulation  on  exhaust  emissions  was examined  and an  evaluation  of  the  U.S.
   Environmental  Protection  Agency's  Automobile  Exhaust  Emission  Modal  Model  was
   conducted.  The  Modal  Model is  a component  of the Intersection  Midblock  Model  and
   MOBILE2, two widely  used programs for  predicting emissions from mobile sources.
        Results from tests  on  a Chevrolet  Celebrity  (3000 pounds gross vehicle weight)
   indicated  that  emissions  during  tests  with water  brake  load simulation did  not
   differ significantly from those  during  tests with  actual  road load  simulation.  For
   the Celebrity, the  load  applied by the  water brake with the tire rolling resistance
   losses on  the dynamometer was approximately equal  to  the  actual  road  load measured
   in highway tests.
        Evaluation of  the Modal Model  was  completed  by  comparing actual  emissions data
   with predicted  values.   The Celebrity was  used  to generate emissions data  for  the
   New York City Cycle, the Surveillance Driving  Schedule, and the Federal  Test Proce-
   dure.    Results  indicated  that  the  Modal  Model  was  unable to  accurately  predict
   emission rates for  the Celebrity.
                                        "~
17.
                                KEY WORDS AND DOCUMENT ANALYSIS
                  DESCRIPTORS
                                              b.lDENTIFIERS/OPEN ENDED TERMS
                                                                         c.  COSATI Field/Group
18. DISTRIBUTION STATEMENT

  RELEASE TO PUBLIC
19. SECURITY CLASS ITMi Repartl
  UNCLASSIFIED
21. NO. OF PAGES

    29
                                              20. SECURITY CLASS

                                                UNCLASSIFIED
                                                                         22. PRICE
EPA Form 2220-1 (R«v. 4-77)   PREVIOUS EDITION is OBSOLETE

-------
                                          EPA-600/3-83-104
                                          November 1983
    EFFECT OF LOAD SIMULATION ON AUTO EMISSIONS
               AND MODEL PERFORMANCE
                        by
           Peter Gabele and Richard Snow

Emissions Measurement and Characterization Division
    Environmental Sciences Research Laboratory
       U.S. Environmental Protection Agency
   Research Triangle Park, North Carolina 27711
                                                           BEST

                                                        AVAILABLE

                                                            COPY
    ENVIRONMENTAL SCIENCES RESEARCH LABORATORY
        OFFICE OF RESEARCH AND DEVELOPMENT
       U. S. ENVIRONMENTAL PROTECTION AGENCY
         RESEARCH TRIANGLE PARK, NC 27711

-------
                      NOTICE

This document has been reviewed in accordarce with
U.S. Environmental Protection Agency policv and
approved for publicaticn.  Mention of trade names
or commercial products does not constitute endorse-
ment or recommendation for use.
                        ii

-------
                                    PREFACE

     Ambient, standards for air pollutants  exist  for  the  protection  of humans
and their  environment.   Adequate planning is  necessary to  assure  compliance
with standard  levels.  But  unless communities are able to  predict  in advance
the cause  and  effect  relationships which  take  place  when emission  sources
arrive  on  scene,  planning   for a  clean  and  healthy  environment  becomes
unmanageable.
     Models  for predicting emissions from  automobiles  have  been  developed and
promulgated  by the  EPA.  These model:;  are  used  extensively  by  states  to
develop  scenarios  for   environmental  planning.  The  Environmental  Sciences
Research Laboratory  contributes   to  the formulation  of  predictive models  by
providing  emission factors  from mobile  sources.   Because emission  control
technology progresses with time,  emission  factors and  the methodology used to
obtain them must be updated periodically.
     This  report  evaluates  the effectiveness  in  predicting  current  vehicles'
emissions of an emission rate model  which is  used widely throughout the United
States.   It  also  examines  the  dynamometer test  procedure  which  is  used  to
obtain automobile emission factors.
                                     111

-------
                                   ABSTRACT

     The overall objective  of  this study w
-------
                                   CONTENTS
Preface    	   iii
Abstract   	   iv
Figures    	   vi
Tables     	   vii
Abbreviations  	   viii

     1.  Introduction  	   1
     2.  Conclusions   	   3
     3.  Recommendations 	   4
     4.  Experimental Procedures 	   5
          Facility description   	   5
          Test vehicle   	   5
          Experimental design  	   6
          Dynamometer road load simulation   	   6
          Dynamometer inertia simulation 	   7
          Emissions testing  	   8
          Test cycle descriptions  	   8
          Computer model predictions 	   9
     5.  Results and Discussion	   12
          Actual road load determination	   12
          Dynamometer road load simulation   	   12
          Dynamometer water brake simulation 	   14
          Vehicle inertia load simulation  	   14
          Emissions testing  	   14
          Model predictions	   15

References   	   21

-------
                                    FIGURES

Number                                                                Page

  1            Sampling scheme     	    18
  2            Actual road load curves ,	    18
  3            Actual road load curve, .ictual road load
               simulation plus rolling resistance losses on the
               dynamorneter, rolling resistance losses on the
               dynamometer   	    19
  4            Actual road load and simulated actual road load
               curves        	    19
  5            Actual road load versus water brake load
               simulation    	    20
  6            Emission rate function curves for accelerations
               held constant at 1.5 to 2.25 mph/s	    20

-------
                                    TABLES

Number                                                                Page

  1            Test Vehicle Description  	       6
  2            Test Matrix   	       8
  3            Test Cycle Description  	       9
  4            Means, Standard Deviations, and Estimated
               Errors for Emission Rates   	      11
  5            Torque Data Collected During Road Tests 	      13
  6            Vehicle Inertia Measurements for 0 to 60 mph
               WOT Accelerations   	      14
  7            Exhaust Emissions Summary   	      15
  8            Percentage Error of Predicted Values  	      16
  9            Percentage Error of Predicted Values  	      17

-------
                       LIST OF ABBREVIATIONS AND SYMBOLS

ABBREVIATIONS

A/D            analog to digital
CFM            cubic feet per minute
CVS            constant volume sampler
FTP            Federal Test Procedure
HC(s)          Hydrocarbon(s)
I MM            Intersection Midblock Model
NO             nitrogen oxides
NYCC           New York City Cycle
SDS            Surveillance Driving Schedule
SS             steady state
rpm            revolutions per minute
VMT            vehicle miles traveled

-------
                                   SECTION 1

                                 INTRODUCTION
    >
     Ambient air concentrations of carbon monoxide (CO)  within urban  areas  are
often significantly higher than those predicted by dispersion models.   Because
CO  is  emitted  predominantly  by  motor  vehicles,  models  used  to  predict  CO
emission rates from groups or classes of  motor  vehicles could be contributing
substantially   to   the  shortfall   ard   should   be   examined  for   obvious
inaccuracies.   Two  widely  used models  or r,-ograms  for estimating  emissions
concentrations  for  hydrocarbons(HCs),   nitrogen   oxides  (NO  ),  and  CO  are
MOBILE2 (1) and the Intersection Midblock Model(IMM)  (2).      x

     MOG1LE2  was   developed  by  EPA  using  data  acquired  over 12  years  of
emissions testing.  Because the bulk  of  emissions data was  collected  for  the
Federal Test Procedure (FTP), other models are  used within  MOBILE2 to  correct
emission rates  to non-FTP conditions.   For  example,  the  EPA  Modal  Analysis
Model is used  to  estimate  emission rates for vehicles  operating over  driving
cycles with average speeds different  from the  FTP.  These emission  rates  are
used  to develop  speed  correction  factors  for  correcting  FTP data  to  the
specific case being modeled.

     The Intersection Midblock Model  ([MM) was  also developed by EPA  to aid in
the identification  and analysis of CO  hot  spot  locations.   It uses  the  EPA
Modal  Analysis Model  to  calculate   CO  emissions  due  to  vehicle  cruising,
acceleration-deceleration  (accel-decel),  and  assigns  these  emissions   to
traffic links based upon calculated  intersection  parameters.   After  emissions
have been distributed among individual lanes  of each link,  the EPA  HIWAY Model
is used to predict CO ambient concentrations  at the desired  locations.
     One questionable component of both  models  is  the  EPA  Modal Analysis Model
which  is  designed  to  predict emission  rates  for specific vehicles  being
operated over  any  defined  driving schedule.   The Modal  Analysis  Model  was
developed in 1973  using  data  obtained on pre catalyst cars.   Altnough  it  was
later refined and updated to 1977 model-year cars,  the  model  remains outdated
in  the  context of automotive  pollution  control  advances  which have occurred
since that time.
     Additionally, the emissions used to develop the Modal  Model were obtained
from tests  on water  brake  rather than  electric  dynamometers.  Although  the
test  procedures using  water  brake  dynamometers   is  an adequate  method  for
emissions  certification,  it  is  unabla   to  simulate  vehicle  road  loads  as
accurately as an electric dynamometer (3,4).
     Both the Modal Model and dynamometer load  simulation technique  should be
examined  for  obvious  inaccuracies.    If  inaccuracies  exist, compensatory
methods or techniques  can, hopefully, be  applied  to reduce  errors  and  improve
model quality.
     The  objectives   of  this   experimental   program   are   to   investigate
contentions that water brake dynamometers fail  to accurately simulate  vehicle
road  loads  during  tests  and  to  evaluate  the  Modal  Model's   ability  to
effectively predict emission rates for new cars. The effect  of water  brake

-------
dynamometer  loading on  exhaust  emissions  was  measured  for  three  driving
cycles:  the  FTP,  the  Surveillance Driving  Schedule  (SDS),  and the New  York
City Cycle  (NYCC).   For each  cycle,  emission rates  measured  at actual  road
load were compared  with  those  measured for water brake  load.   Examination  of
the Modal Model involved comparing emission  rates predicted by  the  model  with
those actually measured.   Predictions made for Test  Phases  2  and 3  of the  FTP,
the SDS, and  NYCC were  compared  with measured emissions  data  that had  been
obtained during the load study  phase of the program.

-------
                                   SECTION 2

                                  CONCLUSIONS

     Investigation  of the  effect  of  dynamometer  load  characteristics  upon
regulated emission  rates  and an  evaluation of the  EPA  Modal Analysis  Model
were completed.  Based upon the study's findings the following is concluded:

     1.  The load applied  by  the  water brake and the tire  rolling  resistance
     losses on the dynamometer was approximately equal  to the actual  road load
     measured in highway tests.

     2.   Regulated  emission  rates  for  the Celebrity  are   not  significantly
     different, when tested using the wati;r brake simulation  versus actual road
     load simulation on an electric dynamometer.

     3.  The EPA  Exhaust  Emissions  Moda'i  Model  is an inaccurate predictor  of
     regulated emissions from the Celebrity.

     Ihe conclusions suggest that water orake dynamometers adequately simulate
actual  road  loads  for emissions  test purposes.  This   should  hold  true  for
vehicles  such  as  the  Celebrity  which  have  large inertia load  components
relative to aerodynamic load  components.   When  the  aerodynamic  load  component
becomes a  significant portion  of the total  road load,  dynamometer  absorbed
power  theoretically  deviates  with  speed  from the  actual   road load.   Tne
tendency for this occurrence, which makes  simulation of  road loads with water
brake dynamometers more difficult, increases for extremely lightweight cars.
     Because most data collected  for  use  in  MOBILE2  have been  from  vehicles
roughly equal  in  size to or  larger than  the Celebrity,  inaccuracies  in load
simulation do  not  have  any significant  effect on the  accuracy of  MOBILE2.
However, should minicars  (<2000  Ib) ever  occupy  a significant  percentage  of
the  vehicle  miles  traveled  (VMT),  a  re-evaluation   of  dynamometer  load
simulation will become necessary.
     With regard  to the Modal  Model evaluation, results in  tests on  only one
vehicle cannot  in themselves disprove the model.  This is true because  the
model  was  recommended for  prediction of vehicle  group  emissions  and  not
individual vehicle emissions (5).   However, because high tech emission control
systems have  changed  the relationship  between vehicle  speed   and  emissions
since the model's development, the Modal Model should be updated.

-------
                                   SECTION 3

                                RECOMMENDATIONS

     It  is  suspected  that the  Modal  Model   is  an  inaccurate  predictor  of
regulated emissions  f®)m late model  cars which are  equipped with high  tech
emission control  systems.   Because this  model  plays  an active  role  in  both
MOBILE2 and the  IMM,  it  should be updated or, if necessary,  replaced with  an
acceptable alternative method.

     In the case  of  MOBILE2,  speed correction factors are  now being  obtained
using  actual  emission test results  rather than  results  predicted using  the
Modal Model.  This requires testing  a rather  large cross section  of  vehicles
over test cycles having different average speeds.  These data will  be used  in
developing realistic speed correction factors  for use  in MOBiLE3,  an  upcoming
revision of the current MOBILE2.

-------
                                   SECTION 4

                            EXPERIMENTAL PROCEDURES
FACILITY DESCRIPTION
     All emissions tests were conducted with a chassis dynamometer for vehicle
road load simulation.   The  Horiba  model  CD6-800/DMA-915  dynamometer  simulated
road load by  means of  a DC  electric  motor-generator directly coupled to  the
front rolls of  the dynamometer.  A control  system  was used to vary  armature
current to achieve the desired motor torque.  In addition to vehicle  road  load
simulation,  the  electric  drive was capable  of simulating vehicle inertia  as
1-lb increments.   Rolls of the dynamometer  are  22  cm (8.65 in.)  in  diameter
and are coupled during the automatic calibration mode.
     Exhaust gases from the  test  vehicle were directed via a flexible 7.6  cm
(3-in.) stainless  steel line to ;i  45.7-cm (18-in.)  diameter dilution, tunnel
(Figure 1).   A  Constant Volume Sampling  System  (CVS) located  at  the  rear  of
the dilution tunnel drew diluent a.nd exhaust gas at  a rate of about 700 CFM.
     Regulated  gaseous  emissions  and carbon  dioxide  (C0~)  were determined
using standard  bag sampling  and  analysis procedures  in accordance  with  the
Federal   Register  (6).   In  selected   test  runs, these   same  emission:,   were
measured using  a  real-time  computer  system to  obtain modal emissions data.
The real-time system, which has been previously described (7),  centered around
operation of a Texas  Instruments 960B minicomputer.   Analyzer  response times,
which vary with exhaust gas flow rate, were  determined with  the aid of a  flow
measuring  device  at  the  engine  air inlet.   Analog  outputs  from  the   gas
analyzers were   directed  to  the   computer  through   analog  to  digital (A/D)
converters.    In  addition  to  gas  data, modal  calculations  of  CVS flow rates
corrected to standard atmospheric conditions were also determined.

TEST VEHICLE

     The  test vehicle  used  in this study,  a  1982 model  year,  Chevrolet
Celebrity with  a 2.5-1, in-line,  4-cylinder engine is described  in  Tajle  1.
The engine  was  fitted  with  throttle-body fuel  injection, and engine  exhaust
gases were treated in a three-way single bed catalytic converter.   The vehicle
was equipped with cruise control, which was used during steady  speed  testing.

     In order to  measure  torques  required to operate  the  vehicle  during  road
and dynamometer  testing, wheel  torque sensors  were  instrumented on both front
drive wheels.  Signals  from  each   Censor  were transmitted  to  a  strain gauge
conditioner which  provided  an analog output  signal   as well as a  calibration
feature.  Torque signals were  stored  on  tape using  a four channel, frequency-
modulated instrumentation  recorder which was  powered off  the  vehicle's  DC
system.

-------
                     TABLE 1.   TEST VEHICLE DESCRIPTION
             Vehicle                            Chevrolet Celebrity
             Model year                         1982
             Engine; type                        L4
             Displexement (1)                   2.5
             Carburetion                        EFI (Throttle-body)
             Emission control                   3-Way catalyst
             Inertia Weight, (Ibs)              3500
             Compressions ratio                 8.2
             Net HP @ RPM                       112/4800
             Transmission                       Auto
             Number of Doors                    Four
             Odometer (mi)                      10,000
     During the  road  tests a fifth wheel  was  used to measure  vehicle  speed.
The speed signal was  transmitted  inl.o  a  separate  channel  on  the tape recorder
to enable calculation of load or power since power is a function of torque and
speed.

     All testing was done using the Goodyear Viva  II steel belted radials with
which the car  came  equipped.   Tire inflation pressures were held  at about 35
psi  during  road tests  and 45  psi  on the dynamometer.   All  tires  had  been
driven about 10,000 mi before the test program began.


EXPERIMENTAL DESIGN

     The experimental program involved three stages:  (1)  electric dynamometer
simulation  of  actual  road  and  water  brake  dynamometer  leads,  (2)  emissions
testing, and  (3)  computer model predictions.  Each  stage required completion
before thf following stage could proceed.

Dynamometer Road Load Simulations

     In  the initial  stage,  the  actual  road  load  for the  test  vehicle  was
determined  through  road  testing.   Wheel  torques  and   vehicle   speeds  were
recorded  in both directions  on a  level  stretch  of  highway  located on  U.S.
Route 64 at the  Lake  Jordan  Dam Project.  The  test section  extended for about
2000 ft with a .02 % grade.  On the day of testing the wind speeds, which were
measured by hand-held anemometer, at no time exceeded 1 knot.   Low wind  speeds
and dry weather created nearly ideal conditions for testing road loads.

     In  order  to develop  the  required speed-load relationship, the  loads on
the vehicle were measured  using cruise control at  steady speeds  ranging from
70 to  30 mph  in  10-mph  increments.  An additional  load  point at  15 mph  was
taken without  cruise  control.  All  tests were  run in both directions and some
of the tests were repeated.  Points at 70, 30,  and 15 mph  were rerun after the
entire test sequence  in order to determine test repeatability.

-------
     For each of the speeds examined, load (horsepower) values were calculated
as a  function  of the average  torque  multiplied  by the wheel  revolutions  per
minute  (rpm).   Speed-load  points were  fed  into a  computer  which provided  a
best  fit  quadradic  equation  using  a  nonlinear  least  squares  method.   The
resultant speed-load equation  or curve  represented  the actual  road load case.
A previously described  procedure  for determining road  load with wheel  torque
meters (4) was generally adhered to in this  study.

     After  the  actual   road  load  equation  had   been  determined,   it  was
programmed  into  the electric  dynamometer and  the  vehicle was tested.   Wheel
torque and  rpm measurements were made so  that  a  new speed-load equation could
be  obtained.   This  represented  the  actual   road  load  simulated   by  the
dynamometer plus tire rolling  resistance  losses  on the dynamometer rolls.  To
compensate for added rolling resistance losses, a new equation was  obtained by
subtracting the  tire losses (power measured at  the wheels minus  power being
absorbed  by  the  dynamometer)  from   the  actual  road  load  equation.   The
resulting equation  was  then used to  simulate  loads  which when added  to tire
rolling resistance  losses on the dynamometer closely  approximated  actual  road
loads.   Some  slight  adjustments  were  made   to   the  coefficients   of  the
aerodynamic and  rolling  resistance  terms  to more closely  simulate  the actual
case.

     Once the  actual road  load  curve could be  accurately reproduced  on  the
dynamometer,  an  equation   for  programming  water  brake  loads  was   sought.
Coast-down  data  from a  Clayton  water brake dynamometer  located  at  EPA,  Ann
Arbor, was  used  to  develop  the necessary  water brake  speed-load  relationship.
Horiba,  Inc.,  the  electric dynamometer  manufacturer, had  also  provided  an
equation which could be used to simulate water  brake loads.  This  curve, while
found to be almost  identical to the one derived from data supplied by  EPA, Ann
Arbor, was not used  in the program except as a  verification device.

Dynamometer Inertia  Simulation

     Most chassis  dynamometers employ  flywheels  to  simulate  vehicle inertia
loads.   More recently,  however, electric  dynamometers  have dispensed  with
flywheels and instead use  electric  simulation.   Many  versions provide inertia
selections  in  1-lb  increments, a  feature not  practical with  flywheels.   But
the  principal   argument in favor  of  flywheel  elimination   is  the  obvious
space-saving advantage.

     The  inertia setting for the Chevrolet  Celebrity  was  3000 Ib.   While this
value was used in emissions certification, it was somewhat less than the total
effective mass (gravitational  plus rotating  component)  as tested  on the road.
The  rotating component,  estimated  from data obtained  previously on a similar
car  (8),  plus  the  weight of  the vehicle, test equipment, and test  personnel
was about 3500 Ib.

     To measure  the accuracy of dynamometer inertia  simulation, a  group of 0
to 60 mph wide-open  throttle accelerations were run on both the level  road and
the  dynamometer.   Integrated torque  values  measured  during  the  accelerations
were  compared  in   each  case  to determine  accuracy  of   dynamometer  inertia
simulation.  Because the actual  vehicle  weight accelerated  on  the  road  was
about 3500  Ib., dynamometer inertia simulation was  set at this value.

-------
Emissions Testing

     In the second stage of the experimental  program, emissions tests  were  run
on the dynamometer.  The  test  matrix  which was followed is shown in  Table  2.
Exhaust  emission rates  were  measured  for  three test  cycles and  two load
conditions.  The  test  cycles were the  FTP,  NYCC, and  the  SDS,  and  the load
conditions were  actual road  load  and  water-brake  load.   Real-time  emission
testing was used with the SDS because  modal data from that  cycle were  required
by the model  to predict emission rates for other cycles.


                        TABLE 2.  TEST MATRIX*

                          Actual road      Water brake    Real  time
 Test Cycle	simulation      simulation	system	

      1st Day
              FTP             X
              NYCC            X
              NYCC                            X
              SDS                             X               X
              SDS             X                               X
              SDS             X                               X
              SDS                             X               X

     2nd Day
              FTP                             X
              NYCC                            X
              NYCC            X
              SDS             X                         •      X
              SDS                             X               X
              SDS                             X               X
              SDS             X                               X
* Sequence repeated six times

Test Cycle Uescriptions--

     Of the three test cycles or driving sequences examined in this study,  the
FTP is most  familiar  to  those in the automotive emissions control  field.   It
represents a  typical  urban  driving schedule which has been adopted by  EPA in
its certification procedure.  Total distance of the FTP  is 7.5  mi  and  average
speed is  19.6 mph.  The  cycle contains  three  distinct phases—cold transient,
hot  stabilized,  and  hot transient-- and  each phase  has its  characteristic
emissions.   A  more detailed  description  of the FTP  is  given  in  the  Federal
Register.

     The  NYCC (sometimes  referred to  as   the  New York  City  Driving  Cycle)
represents a typical Manhattan driving experience.   The cycle  is characterized
by low speeds,  very high accelerations, frequent stops., and  a  40% idle time.
Total distance of the NYCC is 1.2 mi with an average speed of  7.1 mph.

-------
     The SDS, unlike the  FTP and NYCC, does not represent routine driving but
is designed  to  measure  vehicle emissions over  a  variety of steady  state and
transient driving conditions.  To accomplish this  the SDS contains 37 distinct
modes:  32 at differing accel/decel rates which originate at  different speeds
and 5 at steady  state speeds  of  0, l!i, 30,  45, and  60  mph.   Acceleration and
deceleration  rates  covered  within the  driving sequence  represent  the  full
range of rates observed  in the CAPE-10 car-chase study (9).

     Except for  the FTP  in  which the  car was started cold  following an over-
night soak period, eacli of  the cycles was run following  a  hot  soak  period of
10 min.  A brief  summary  of  the  three driving cycles discussed  above is shown
in Table 3.

                       TABLE 3.  TEST  CYCLE  DESCRIPTIONS

Test         Cold         Avg.      ."Stops     Total     Duration   %  Time
Cycle	Start	Speed	i:er Mi    Distance    (min)	Idle	

FTP          Yes          19.6       '2.40       7.5       22.9      19.0

NYCC         No            7.1       9.32       1.2       10.0      35.2

SDS          No           33.5       0.82       9.8       17.6      11.7
Computer Model Predictions—

     Following the emissions  testing  stage,  data from the  SDS  were  available
for use in the Modal Model  to arrive  at  emission rate predictions.   The Modal
Model  formulated  an  instantaneous  emission  rate function  for the  vehicle,
which was  used  to calculate  second-by-second  emissions  over any given  speed
versus time driving sequence.   Integration of these emission rates  resulted in
predicted  values  for the  emissions over  the  entire  test  cycle or  portions
thereof.

     The emission rate function which was developed within the model  was based
on assumptions that  steady state emission rates  are a quadratic function of
speed,  that acceleration  is a perturbation to the steady  state emission rate
function, and that quadratic functions of acceleration represent good approxi-
mations to  the  perturbation  (5).   Two mathematical  expressions are used to
define the emission rate  function,  one representing  the  steady  state function
and  one  representing  the  non-steady  state  or  transient  function.   Taken
together, the two functions require specification  of 12  coefficients  -- 3 for
the steady state function and 9 for the transient functions.

     Coefficiont specification is accomplished through processing data obtain-
ed in 37 SDS modes.  Data from the 5 steady state modes and from the remaining
32 accel/decel modes  are  used to define  the  transient function.  Predictions
of instantaneous emissions are carried out by joining the two functions  with a
weighting function.  The  weighting  function,  which  is a  function of acceler-
ation, allows for a smooth  transition  between  the steady state  and  the  accel/
decel emission rate functions.

-------
     Because modal data, repeatabilities are poor, a sample of  at  least  25 SDS
tests was  obtained.   For example, the  relative  standard deviation for  27  CO
samples obtained  in Mode  11  during  replicate SDS tests was about  38%.   Given
27 replicate  results  with a  standard  deviation  of 38%,  there is a  95% cer-
tainty that the mean  emission rate  calculated for Mode 11 will be within 15%
of the true value.  A complete  listing  of  the means,  standard  deviations, and
estimated errors  for  emission rates  in each  of  the  37 SDS modes  is  shown  in
Table 4.   It  is  noted that modal data exceeding  two  standard  deviations were
eliminated before the above statistics were performed.

     Using mean emission  rate values  for eacli mode,  the  Modal Model  was used
to predict emission  rates for Test Phases 2  and  3  of the FTP, the NYCC, and
the SDS.   Emissions predictions  for each of the 38 SDS modes  were also made.
Predicted  values  were compcired  with  measured values  obtained  in 12 FTP,  16
NYCC, and 27 SDS tests.
                                       10

-------
   TABLE 4.   MEANS,  STANDARD  DEVIATIONS,  AND  ESTIMATED  ERRORS  FOR
        EMISSION RATES.   MEASURED IN  27  REPLICATE  SOS TESTS.
Mode
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27*
28
29
30
31
32
33
34
35
36
37
Mean
g/mi
10.5
0.7
8.8
4.3
6.8
0.6
173.2
3.7
33.5
0.6
43.6
0.6
33.0
0.6
0.1
0.9
4.1
0.1
22.6
0.1
137.0
0.8
2.1
14.2
0.7
0.1
6.3
0.6
1.9
73.5
1.1
0.1
1.7
1.3
1.2
1.0
3.7
Std. Dev.
g/mi
5.2
0.6
5.2
2.1
3.7
0.5
31.1
1.4
22.0
0.2
16.6
0.2
17.2
0.3
0.1
0.7
2.7
0.1
15.7
0.1
29.8
0.3
1.2
12.1
0.4
0.2
7.3
0.3
1.6
21.5
0.5
0.1
0.6
0.7
0.5
0.3
1.6
Estimated
error
18
32
22
18
20
33
6
14
24
14
14
13
19
21
43
30
24
34
26
24
8
15
22
32
22
57
44
20
31
11
18
36
13
20
17
11
17
*Estimated error = (Error/mean) x 100%.
                                       11

-------
                              SECTION 5

                       RESULTS AND DISCUSSIONS


ACTUAL ROAD LOAD DETERMINATION

     Wheel torque data collected during road tests are shown in Table  5.   The
numbers  represent average  torque  measured  during  the  test  run.   Slightly
higher  values  shown  in  runs  headed north  are  related  to a  small  (0.02%)
positive  grade  in that direction.   The  torque  data  were collected  at  steady
speeds ranging from about  15  mph  to 70 mph.   When the data were  reduced,  two
speed-load  curves were  drawn  up:   on;  for  speeds  <  70  mph   and  one  for
speeds <  50 mph.   Figure 2  shows  the two curves, which are similar  in  shape,
alongside each other.  Equations  for Curves A  (<70 mph data) and  B (<50 mph),
respectively, are:

                         T  = 37.2  - 0.17 v +  0.024 v2                 (A)

                         T  = 26 +  0.33 v + 0.008 v2                   (B)


               where:     v  = velocity in mph

                         T  = torque in ft-lb

At 30 mph there  is only a  2  ft-lb difference  in torque (3.6%)  between  the  two
curves and at 50  mph  a 3  ft-lb. difference (3.4%).  Because most test  cycles
run  in  this  program were at  speeds  <50  mph,  Curve 8 was selected  to  provide
actual road loads required  for simulation.

DYNAMOMETER ROAD LOAD SIMULATION

     Curve B was  programmed  into  the dynamometer and the vehicle was  readied
for  testing.   With wheel  torque  meters  in  place,  torque  measurements were
taken over the same steady speed  points examined  on  the  road.  The speed-load
relationship obtained is shown in  Figure 3 as Curve C.  The  difference between
Curve  C  and Curve   B  is  due  to  tire   rolling resistance  losses  on  the
dynamometer and is plotted  as Curve D in Figure 3. When Curve D was subtracted
from  Curve  C  and   the   resulting   relationship was   programmed  into   the
dynamometer, wheel  torque  values  were again  obtained  and,  after some  slight
adjustments to  the dynamometer load equations,  these  values were  plotted  as
Curve E  in  Figure 4.   This  load  curve very  closely simulates the  load curve
(Curve B) obtained in actual road  testing.
                                       12

-------
TABLE 5.  TORQUE DATA COLLECTED DURING  ROAD  TESTS
Speed
(mph)
68.9
67.6
70.9
67.9
67.1
59.4
58.9
49.5
49.0
40.3
39.4
31.7
30.1
30.9
28.9
14.4
15.0
14.7
14.3
«,^, — ,»«•-• — -..- — — — ^-.w.^ — — -sa-» — — — —
Direction

N
N
S
S
S
H
S
N
S
N
S
N
N
S
S
N
N
S
S
— — — •» — — .-•. — — .» — ••-.-. — — —
Torque
(ft-lb)
150
143
145
135
128
114
106
88
82
76
67
61
65
53
52
41
41
40
36
                            13

-------
DYNAMOMETER WATER BRAKE SIMULATION

     The dynamometer manufacturer had furnished a load equation which  could  be
used to simulate a water brake dynamometer.   In addition,  coast down data were
available from one of  the  Clayton water brake dynamometers at the EPA  in Ann
Arbor.  Since both curves were similar,  it was  arbitrarily decided to use the
relationship developed from the EPA, Ann Arbor coast down  data.

     Wheel  torque measurements  obtained  while  the  vehicle  was undergoing
water-brake-simulated  loads  accounted  for tire  rolling  resistance losses  on
the  dynamometer  rolls  in  addition  to  dynamometer  load.   The  load  curve
representing this  condition  is shown  in  figure 5 with the actual road  load
curve  previously  obtained.   While: differences  do  appear  at the low  and  high
speed ends, overall the curves an; not significantly different.

VEHICLE INERTIA LOAD SIMULATION

     As a means of insuring accurate dynamometer simulation of vehicle inertia
weight,  integrated  inertias  were  measured  during wide-open throttle  (WOT)
accelerations on  both  the  dynamometer  and level road.  The results are shown
in Table 6.                                      '

               •   TABLE 6.   VEHICLE INERTIA MEASUREMENTS  FOR
                            0 to 60 MPH WOT ACCELERATIONS
Inertia
Weight
(lb)
Time
(sec)

Integrated
torque
(ft-lb-sec)
Dynamometer              3000                16.4            8522
                         3000                16.6            8569
                         3500                19.1            10003
                         3500                19.0            10043
Road
(6 runs)                 3500        *   x = 19.4            10909
                                    **   s =  2.2%             1.7%
*  x =  the mean of 6 runs.
** s = relative standard deviation.


At simulated  inertia  of  3000 lb.  the integrated torque values  fall  about  15%
below those at 3500 lb.  The values obtained in dynamometer simulations of  the
loaded test  car (3500 lb)  are  within 10% of  those  obtained on  the roadway.
Acceleration  times are also shown  in Table 6.   While a decrease  of  about  15%
again  noted  in going  from 3500 lb  to 3000 lb,  a difference  of only 2%  is
observed between dynamometer  and  roadway  acceleration  times.   Indications  are
that the dynamometer is accurately simulating vehicle inertia load.

EMISSIONS TESTING

     Emissions of HC,  CO,  NO ,  and tO~ for the Celebrity  were  determined  for
three  different tests  cycles using  actual  road  load ,and  water brake  load


                                       14

-------
simulations.  Because both  load  curves were similar,  significant  differences
in emissions rates were not anticipated.  Emissions data  presented  in  Table 7
supported the expectation  that no nignificant differences  in  emissions  occur
as a result of load parameter.

                 TABLE 7.   EXHAUST MISSIONS SUMMARY
Emission
HC
CO
N0y
n
co2
=================
Load
simulation
Actual road
Water brake
Actual road
Water brake
Actual road
Water brake
Actual road
Water brake
FTP
(g/mi)
0.24(±.04)
0..23(±.04)
6.33(±.89)
6.83(±.26)
1..27(±.07)
1.25(±.09)
341,2(43)
348,,9(±5)
NYCC
(g/mi)
0.77(±.23)
0.76(±.26)
10.91(tl.41)
11.13(±1.77)
2.06(4.35)
2.11(4.35)
656.8(12.8)
557.7(±10.27)
SOS
(g/mi)
0.18(4.03)
0.17(4.03)
15.84(±1.97)
16.62(42.3)
0.91(4.05)
0.93(4.06)
308.5(±2.6)
307.8(±3.9)
     Modal emissions data revealed no  significant  differences  betv/een  the two
imposed  load  conditions.   However,  large  scatter,  characteristic  of  data
obtained  in  replicate  modal analysis  runs,  completely masks  any  differences
which  might  have existed modally (see  Table  4).   The  purpose of  obtaining
modal  data  during the SDS  was not  so  much  to examine  emissions  differences
because of load  parameter changes as to evaluate the accuracy  of  Modal  Model
predictions.

MODEL PREDICTIONS

     Emission  rates for HC, CO, and  NO  were predicted  for Test Phase  2 and 3
of the FTP,  the  NYCC,  the SDS, and eacfh of  33 modes of  the SDS.   Predictions
were not made  for Test Phase  1 of the  FTP because  cold  start emissions cannot
be successfully modeled by the Modal  Analysis Model.

     Table 8 shows the percentage error of predicted values compared to those
actually measured.  The negative values indicate that in all cases  predictions
are  lower than measured  values.   For CO there is a trend  of increasing error
as  average  cycle speed  increases.  Generally,  the error in  predicting  NO
emission rates is lower than those for predicting CO and HC.

     CO emission rate  predictions for  each  SDS mode were compared with  the
calculated  emission  rates  for those same modes.   A summary  of  the  results
showing the error of the prediction  is  shown in  Table 9.   The  negative values
again  indicate  that  with  only  one  exception  predictions   are  lower  than
measured  values.  For  acceleration  modes  with average   speeds over  30  mph
predictions appear to be unreasonably  low.   Measured emission  rates  for these
modes were the highest for the entire driving cycle while the predicted values


                                        15

-------
were usually less than those for deceleration modes.   Predictions for idle and
15 mph SS were very close to the measured values; however, those for 30 and 45
mph SS were rather low and high, respectively.

     In  general,  predictions  made  by  the:  EPA Modal  Analysis  Model  were
considerably lower than measured values.  This  was particularly apparent with
CO predictions  for driving cycles and  SDS  accel modes having  average speeds
over  30  mph.   To  further  investigate  this  trend,  a family  of curves  (see
Figure  6)  describing  the  emission  rate  function  were  drawn  showing  CO
emissions as a  function  of velocity for accelerations ranging  from 1.5 mph/s
to 2.25  mph/s.   For the Celebrity  negative  emission rates occur at frequent
points within the  model.   Emissions,  regardless of acceleration, appear about
the  same at 30  mph.   Since these  are  physically impossible  trends  for  the
vehicle tested, the curves illustrate an obvious discrepancy in the  model.

     The  percent  errors  as  given  in  Tables 6 and  7  include  test  to  test
variability as well as the  error  in  model predictions.  Table  3 shows that in
most cases test variability was rather high.  It is  also  noted  that the error
percentages  in  Tables 6  and  7  do not  translate  directly  to  MOBILE2  which
merely uses emission rate prediction to  arrive at speed correction factors.

   TABLE 8.  PERCENTAGE EKROR OF PREDICTED VALUES TO  ACTUAL MEASURED VALUES
                                             HC          CO        NOx
         Cycle               Avg. Speed    % error*   % error    % error
        	(mph)	
NYCC
FTP Phase 2
FTP Phase 3
SDS
7.1
16.
25.6
33.5
-83
-47
-63
-64
-36
-46
-65
-90
-63
0
-27
-39
* % err/r = '(predicted-measured) r measured] x 100%
                                       16

-------
                T/IBLE 9.  PERCENTAGE ERROR OR PREDICTED  VALUES
                  Modes                               CO, %  error*
                  Accel s >30 mph  (10 rnocL>s)                -53
                  Accels <30 mph  (7 modes)                 -77
                  Decels (15 modes)                        -54
                  Idle                                     -2
                  15 mph SS**                              -2
                  30 mph SS                                -70
                  45 mph bi                                +50
                  60 mph SS                                -15
*  % error = ((predicted-measured)  f measured) x  ".00%.
** SS = Steady speed.
                                        17

-------
                                Ftgurt 1.  Simpltng
figure
       2.  Actual rojd '>oh *nd below. B for  ttO Epn  *r,.

-------
160
140
I2B
100
 •to
                                VELOCITY
rtfurt ].
          Actual  road load curva (I), actual  ru«l load llnulltlon plut rolling redltlncc  totfttt
          On the  djMrOMUf (c) > r«IMMQ rtltitarxt loll«« on tfjrnjfiocvttr  (C-tl*l>).
  i ear
    flgwre «.
                          Iw! tni i-wli:-< ;-.;,.•! r>
-------
   100
                               VELOCITY


               Floura S.  Actiill rcil lM-J (0) veriui t.*ter bnlt "io*
-------

-------