EPA-460/3-76-011
April 1976
FEDERAL TEST PROCEDURE
AND SHORT TEST
CORRELATION ANALYSES
U.S. ENVIRONMENTAL PROTECTION AGENCY
Office of Air and Waste Management
Office of Mobile Source Air Pollution Control
Emission Control Technology Division
Ann Arbor, Michigan 48105
-------
EPA-460/3-76-011
FEDERAL TEST PROCEDURE
AND SHORT TEST
CORRELATION ANALYSES
by
Mobile Systems Group
Aerospace Corporation
P.O. Box 92957
Los Angeles, California 90009
Contract No. 68-01-0417
EPA Project Officer: F. Peter Hutchins
Prepared for
ENVIRONMENTAL PROTECTION AGENCY
Office of Air and Waste Management
Office of Mobile Source Air Pollution Control
Emission Control Technology Division
Ann Arbor, Michigan 48105
April 1976
-------
This report is issued by the Environmental Protection Agency to report
technical data of interest to a limited number of readers. Copies are
available free of charge to Federal employees, current contractors and
grantees, and nonprofit organizations - as supplies permit - from the
Air Pollution Technical Information Center, Environmental Protection
Agency, Research Triangle Park, North Carolina 27711; or, for a fee,
from the National Technical Information Service, 5285 Port Royal Road,
Springfield, Virginia 22161.
This report was furnished to the Environmental Protection Agency by
Aerospace Corporation, Los Angeles California 90009. in fulfillment
of Contract No. 68-01-0417. The contents of this report are reproduced
herein as received from Aerospace Corpcration. The opinions, findings,
and conclusions expressed are those of the author and not necessarily
those of the Environmental Protection Agency. Mention of company or
product names is not to be considered as an endorsement by the Environmen-
tal Protection Agency.
Publication No. EPA-460/3-76-011
ii
-------
FOREWORD
This report, prepared by The Aerospace Corporation for
the U.S. Environmental Protection Agency, Emission Control Technology
Division, presents the results of a statistical analysis of the degree of
correlation between five short tests and the 1975 Federal Test Procedure.
The correlation analyses were based on experimental test data from 147
1974-model-year vehicles, composed of three inertia test weight groups,
and on 40 catalyst-equipped experimental vehicles.
The results of the study are presented in six sections.
Section 1 contains a summary of the study results. The background, scope,
objectives, and method of approach are given in Section 2. The short tests,
test conditions, and test fleet composition are described and discussed in
Section 3. Section 4 describes the data-screening procedures, the primary
statistical tools used in the correlation analyses, and results of the
statistical analysis in detail for the catalyst-equipped experimental vehicle
fleet. Similar results for the 1974 model year in-use fleet and a five-
vehicle defect test fleet are presented in Sections 5 and 6, respectively.
The conduct of this analysis effort resulted in over 1000
pages of correlation table printouts, regression plots, scattergrams, etc.
This information is summarized in the tables and figures presented in the
report; the voluminous printout material is not included in order to enhance
the readability of the report. However, the printout material is on file at
the Emission Control Technology Division of EPA, Ann Arbor, Michigan,
and may be borrowed for limited periods for reproduction for purposes of
detailed examination.
111
-------
ACKNOWLEDGMENT
During the course of this study, Mr. F. P. Hutchins of the
Environmental Protection Agency's Emission Control Technology Division,
who served as EPA Project Officer for the study, provided valuable guidance
and assistance. His efforts are gratefully acknowledged.
Dr. John Thacker was principally responsible for the statis-
tical analysis effort reported herein. The following technical personnel of
The Aerospace Corporation also made valuable contributions to the analyses
performed under this contract:
W. B. Lee
R. F. Janz
A. M. Timmer
M. G. Hinton, Group Director
Mobile Systems
Approved by:
Itzer, General Manager
fronment and Energy
Snservation Division
-------
CONTENTS
1. SUMMARY 1_1
1. 1 Catalyst-Equipped Experimental Vehicle Fleet .... 1-2
1. 1. 1 Direct Relatability Results 1-2
1. 1. 2 Contingency Table Analysis Results 1-9
1.2 In-Use 1974 Model Year Vehicle Fleet 1-26
1. 2. 1 Direct Relatability Results 1-26
1.2.2 Contingency Table Analysis Results 1-32
1. 3 Defect Data From Catalyst-Equipped Experimental
Vehicle Fleet 1-42
1. 3. 1 Nature of Defects and Statistical Impact. . . 1-42
1. 3. 2 Contingency Table Analysis Results 1-42
1.4 General Overview Remarks 1-47
1. 4. 1 Mode vs Bag ST 1-47
1.4. 2 Single Mode vs Weighted Mode Tests 1-47
1. 4. 3 Garage Instrument vs Laboratory
Analyzer 1-47
1. 4. 4 Correlation Coefficient vs Contingency
Table Analysis 1-48
1. 4. 5 Relative Impact on Air Quality 1-49
2. INTRODUCTION 2-1
2. 1 Background and Objectives 2-1
2.2 Study Scope 2-2
2.3 Method of Approach 2-4
2.4 Organization of Report 2-5
3. TEST CHARACTERISTICS AND PROCEDURES 3-1
3. 1 Short Tests 3-1
3. 1. 1 General 3-1
3. 1. 2 ST Definition 3-2
3. 1. 3 Short Test Sequence 3-6
vn
-------
CONTENTS (Continued)
3. 2 Test Fleets 3-7
3. 2. i Catalyst-Equipped Experimental
Vehicle Fleet (CEV) 3-7
3. 2. 2 Defect Test Fleet 3-8
3.2.3 In-Use 1974 Model Year Vehicle Fleet ... 3-9
4. CATALYST-EQUIPPED EXPERIMENTAL VEHICLE
FLEET 4-1
4. 1 Preliminary Statistical Analyses 4-1
4. 1. 1 Data Screening 4-1
4. i. 2 Multivariate Analysis of Variance 4-3
4. 1. 3 Canonical Correlation Analysis 4-5
4. 1. 4 Summary of Preliminary Analysis
Results 4-7
4. 2 Principal Statistical Analysis Techniques and
Results 4-8
4. 2. 1 Correlation Analysis 4-8
4. 2. 2 Contingency Table Analysis 4-17
4. 3 References for Section 4 4-97
5. IN-USE 1974 MODEL YEAR VEHICLE FLEET 5-1
5. 1 Correlation Analysis Results 5-2
5. 2 Contingency Table Analysis Results 5-9
5. 2. 1 Maximum Correlation Method 5-9
5. 2. 2 Bounded Errors of Commission Method . . . 5-9
5. 3 Relative Impact on Air Quality 5-53
5. 3. 1 By Individual Pollutant 5-53
5. 3. 2 Multiple Constituent Tests 5-57
6. DEFECT DATA FROM CATALYST-EQUIPPED
EXPERIMENTAL VEHICLE FLEET 6-1
6. 1 Statistical Analysis of Defect Tests 6-1
vi u
-------
CONTENTS (Continued)
6. 1. 1 Data Selection Procedures 6-5
6. 2 Contingency Table Analysis of Defect Data 6-6
6.3 Conclusions 6-12
APPENDIX DEFECT TEST DESCRIPTIONS A-l
IX
-------
TABLES
1-1. ST/FTP Correlation Summary (CEV Fleet) 1-3
1-2. Correlations for Weighted Mode Tests (CEV Fleet) .... 1-6
1-3. Correlations for Selected Car Deletions: Federal
Short Cycle vs FTP (CEV Fleet) 1-7
1-4. ST Correlation Ratings (CEV Fleet) 1-8
1-5. Contingency Table 1-9
1-6. Assumed FTP Levels for CEV Fleet 1-11
1-7. Comparison of Selected ST Hydrocarbon Results:
CEV Fleet, Bounded Errors of Commission Analysis,
HC FTP Level = 0.41 gm/mile (E = constant = 5%). ... 1-15
1-8. Comparison of Selected ST Hydrocarbon Results: CEV
Fleet, Bounded Errors of Commission Analysis, HC
FTP Level = 0.90 gm/mile (E = constant = 5%) 1-16
1-9. Comparison of Selected ST Carbon Monoxide Results:
CEV Fleet, Bounded Errors of Commission Analysis,
CO FTP Level = 9. 0 gm/mi (E = constant =5%) 1-18
1-10. Comparison of Selected ST Carbon Monoxide Results:
CEV Fleet, Bounded Errors of Commission Analysis,
CO FTP Level = 3. 4 gm/mi (E = constant =5%) 1-19
1-11. Comparison of Selected ST NOX Results: CEV
Fleet, Bounded Errors of Commission Analysis,
NO FTP Level = 3. 1 gm/mi (E = constant =5%) 1-22
jf. C
1-12. Comparison of Three-Constituent Test Results: CEV
Fleet, Bounded Errors of Commission Analysis,
(Predicted EC = constant =5%) 1-24
1-13. Correlation Coefficient Summary: 1974 Model Year
Fleet 1-27
1-14. ST Ratings: 1974 Model Year Fleet 1-33
XI
-------
TABLES (Continued)
1-15. Comparison of ST Hydrocarbon Results: 1974 Model
Year Fleet, Bounded Errors of Commission Analysis
(E = constant = 5%) I-37
c
1-16. Comparison of ST Carbon Monoxide Results: 1974
Model Year Fleet, Bounded Errors of Commission
Analysis (E = constant =5%) 1-39
1-17. Comparison of ST NOX Results: 1974 Model Year
Fleet, Bounded Errors of Commission Analysis
(E = constant =5%) 1-40
1-18. ST Comparison: 1974 Model Year Fleet, Multiple
Constituent Tests (Actual E < 2%) 1-41
C
1-19. Defect Analysis Comparison Summary: Predicted
Population [% EC = 5, FTP Level I] 1-44
1-20. Key Mode Composite Test (Laboratory Data) 1-45
1-21. Short Test Effectiveness; EC = 5%; 1974 Model Year
Fleet 1-50
1-22. Short Test Effectiveness Values for Multiple Con-
stituent Tests; 1974 Model Year Fleet 1-54
4-1. Number of Cases Available for Statistical Analysis
(CEV Fleet) 4-2
4-2. Summary of Variance Components (CEV Fleet) 4-4
4-3. Canonical Correlation Coefficients Between the FTP and
ST for the CEV Fleet (first good data set) 4-6
4-4. FTP Composite vs Bag Correlation Summary
(CEV Fleet) 4-12
4-5. ST/FTP Correlation Summary (CEV Fleet) 4-13
4-6. ST/FTP Correlations for Weighted Mode Tests
(CEV Fleet) (first good data only) 4-14
xn
-------
TABLES (Continued)
4-7. Correlation Coefficients for Selected Car Deletions;
Federal Short Cycle vs FTP (CEV Fleet) 4-15
4-8. ST Correlation Ratings 4-18
4-9. Contingency Table 4-19
4-10. Assumed FTP Levels (CEV Fleet) 4-27
4-11. Maximum Correlation Summary, FTP Level I
(CEV Fleet) 4-32
4-12. Maximum Correlation Summary, FTP Level II
(CEV Fleet) 4-33
4-13. Maximum Correlation Summary, FTP Level III
(CEV Fleet) 4-34
4-14. Maximum Correlation Summary, FTP Level IV
(CEV Fleet) 4-35
4-15. Comparison of Selected ST Hydrocarbon Results:
CEV Fleet, Bounded Errors of Commission Analysis,
HC FTP Level = 0.90 gm/mile (E = constant = 5%) 4-59
4-16. Comparison of Selected ST Hydrocarbon Results:
CEV Fleet, Bounded Errors of Commission Analysis,
HC FTP Level = 0.41 gm/mile (E = constant =5%) 4-60
4-17. Comparison of Selected ST Carbon Monoxide Results:
CEV Fleet, Bounded Errors of Commission Analysis,
CO FTP Level = 9. 0 gm/mi (E = constant = 5%) 4-71
4-18. Comparison of Selected ST Carbon Monoxide Results:
CEV Fleet, Bounded Errors of Commission Analysis,
CO FTP Level = 3.4 gm/mi (E = constant = 5%) 4-72
4-19. Comparison of Selected ST NOX Results: CEV Fleet,
Bounded Errors of Commission Analysis, NOX FTP
Level = 3. 1 gm/mi (E = constant = 5%) 4-78
c
4-20. Key Mode Weighting Factors 4-79
Xlll
-------
TABLES (Continued)
4-21. Approximate Standard Deviation for Three-Constituent
Tests - CEV Fleet, N = 40 4-95
5-1. Correlation Coefficient Summary: 1974 Model
Year Fleet 5~3
5-2. FTP Composite Versus Bag 2 + 3 Correlation
Coefficients: 1974 Model Year Fleet 5-6
5-3. ST Ratings: 1974 Model Year Fleet 5-7
5-4. Maximum Correlation Summary; 1974 Model Year Fleet,
Predicted Population 5-10
5-5. Comparison of ST Hydrocarbon Results: 1974 Model
Year Fleet, Bounded Errors of Commission Analysis
(E = constant = 5%) 5-18
c
5-6. Comparison of ST Carbon Monoxide Results: 1974
Model Year Fleet, Bounded Errors of Commission
Analysis (E = constant = 5%) 5-28
5-7. Comparison of ST NOX Results: 1974 Model Year
Fleet, Bounded Errors of Commission Analysis
(E = constant = 5%) 5-31
5-8. Standard Deviation for Three-Constituent Tests:
1974 Model Year Fleet, N = 147 5-31
5-9. ST Comparison: 1974 Model Year Fleet; Multiple
Constituent Tests (E < 2%) 5-52
c
5-10. Short Test Effectiveness; E = 5%; 1974 Model
Year Fleet c. 5-54
5-11. Short Test Effectiveness Values for Multiple'
Constituent Tests; 1974 Model Year Fleet 5-58
6-1. ST/FTP Correlation Coefficient Comparison: Defect
Test Vehicles vs Original CEV Fleet (laboratory
instruments) 6-2
6-2. Elementary FTP Statistics: Defect Test Vehicles vs
Original CEV Fleet (gm/mi) 6-3
xiv
-------
TABLES (Continued)
6-3.
6-4.
6-5.
Groups Distinguishable from Baseline
Defect Test Fleet
Defect Analysis Comparison Summary
Population [% E = 5, FTP Level I] .
Kev Mode Composite Test (laboratory
Operation:
: Predicted
data)
6-4
6-7
6-12
XV
-------
FIGURES
1-1. Contingency Table Representation 1-10
1-2. Bounded Errors of Commission Method 1-12
1-3. Variation of EC, Eo, and FF with HC Cut-Point;
CEV Fleet; Federal Short Cycle; Bounded Errors
of Commission Method 1-13
1-4. Variation of Ec, Eo, and FF with CO Cut-Point;
CEV Fleet; Federal Short Cycle; Bounded Errors of
Commission Method 1-17
1-5. Variation of Ec, Eo, and FF with NOX Cut-Point; CEV
Fleet; Federal Short Cycle; Bounded Errors of
Commission Method 1-20
1-6. Variability of Predicted Population Results 1-23
1-7. Variation of Ec, Eo, and FF with HC and NOX Cut-
Points; 1974 Model Year Fleet; Federal Short
Cycle; Bounded Errors of Commission Method 1-36
1-8. Variation of Ec, Eo, and FF with CO Cut-Point;
1974 Model Year Fleet; Federal Short Cycle; Bounded
Errors of Commission Method 1-38
1-9. Impact of Percent Population Sampled on CO
Removed (Illustrative Example Only) 1-52
3-1. Federal Short Cycle and Composite NY/NJ Short
Cycle Test Driving Schedules 3-5
4-1. Correlation Analysis Scattergram 4-10
4-2. Contingency Table Representation 4-20
4-3. Maximum. Correlation Method 4-22
4-4. Bounded Errors of Commission Method 4-22
4-5. Weighted Errors Method 4-23
4-6. Percent Rejection Method 4-23
XVII
-------
FIGURES (Continued)
4-7. Parametric Model 4~25
4-8. Probability Equations 4-25
4-9. Expected Values 4-26
4-10. Equations for Parametric Techniques 4-26
4-11. Variation of Ec, EQ, and FF with HC FTP Level;
Maximum Correlation Method; Data Analytic
Technique; CEV Fleet 4-29
4-12. Variation of Ec, Eo, and FF with HC FTP Level;
Maximum Correlation Method; Parametric Tech-
nique; CEV Fleet 4-30
4-13. Variation of EC, Eo, and FF with HC FTP Level;
Maximum Correlation Method; Predicted Popula-
tion Technique; CEV Fleet 4-31
4-14. Variation of EC, Eo, and FF with HC FTP Level;
Federal Three-Mode Test; Maximum Correlation
Method; Predicted Population of CEV Fleet 4-36
4- 1 5. Variation of Ec, Eo, and FF with HC FTP Level;
Key Mode Test; Maximum Correlation Method;
Predicted Population of CEV Fleet 4-37
4-16. Variation of Ec, EQ, and FF with CO FTP Level;
Maximum Correlation Method; Predicted Population
of CEV Fleet 4-38
4-17. Variation of Ec, EQ, and FF with CO FTP Level;
Federal Three-Mode Test; Maximum Correlation
Method; Predicted Population of CEV Fleet 4-39
4-18. Variation of Ec, Eo, and FF with CO FTP Level;
Key Mode Test; Maximum Correlation Method;
Predicted Population of CEV Fleet 4-40
4-19. Variation of Ec, EQ, and FF with HC FTP Level;
Unloaded 2500 rpm Test; Garage Instruments; Maxi-
mum Correlation Method; Predicted Population of
CEV Fleet 4-41
XVlll
-------
FIGURES (Continued)
4-20. Variation of Ec, Eo, and FF with HC FTP Level;
Federal Three-Mode Test; Garage Instruments;
Maximum Correlation Method; Predicted Population
of CEV Fleet 4-42
4-21. Variation of Ec, Eo, and FF with HC FTP Level;
Key Mode Test; Garage Instruments; Maximum
Correlation Method; Predicted Population of CEV
Fleet 4-43
4-22. Variation of Ec, Eo, and FF with CO FTP Level;
Unloaded 2500 rpm Test; Garage Instruments;
Maximum Correlation Method; Predicted Population
of CEV Fleet 4-44
4-23. Variation of Ec, Eo, and FF with CO FTP Level;
Federal Three-Mode Test; Garage Instruments;
Maximum Correlation Method; Predicted Population
of CEV Fleet 4-44
4-24. Variation of Ec, EQ, and FF with CO FTP Level; Key
Mode Test; Garage Instruments; Maximum Correlation
Method; Predicted Population of CEV Fleet 4-45
4-25. Variation of E0 and Ec at NOX Level of 3. 1 gm/mi;
CEV Fleet; Maximum Correlation Method; Pre-
dicted Population of CEV Fleet 4-46
4-26. Variation of EC, EQ, and FF with HC Cut-Point;
CEV Fleet; Federal Short Cycle; Bounded Errors
of Commission Method 4-49
4-27. Variation of EC, Eo, and FF with HC Cut-Point;
CEV Fleet; NY/NJ Composite Test; Bounded
Errors of Commission Method 4-50
4-28. Variation of Ec, Eo, and FF with HC Cut-Point;
CEV Fleet, Key Mode Test; Bounded Errors of
Commission Method 4-51
4-29. Variation of Ec, EQ, and FF with HC Cut-Point;
CEV Fleet; Key Mode Test; Garage Instruments;
Bounded Errors of Commission Method 4-52
xix
-------
FIGURES (Continued)
4-30. Variation of Ec, Eo, and FF with HC Cut-Point;
CEV Fleet; 1975 FTP HC Standard = 0.41 gm/mi;
Federal Three-Mode Test; Bounded Errors of
Commission Method 4-53
4-31. Variation of EC, Eo, and FF with HC Cut-Point;
CEV Fleet; 1975 FTP HC Standard = 0. 9 gm/mi;
Federal Three-Mode Test; Bounded Errors of
Commission Method 4-54
4-32. Variation of Ec, EQ, and FF with HC Cut-Point;
CEV Fleet; 1975 FTP Standard = 0.41 gm/mi;
Federal Three-Mode Test; Garage Instruments;
Bounded Errors of Commission Method 4-55
4-33. Variation of Ec, EQ, and FF with HC Cut-Point;
CEV Fleet; 1975 FTP Standard = 0. 9 gm/mi; Federal
Three-Mode Test; Garage Instruments; Bounded
Errors of Commission Method 4-56
4-34. Variation of Ec, Eo, and FF with HC Cut-Point;
CEV Fleet; Unloaded 2500 rpm Test; Bounded
Errors of Commission Method 4-57
4-35. Variation of Ec, Eo, and FF with HC Cut-Point;
CEV Fleet; Unloaded 2500 rpm Test; Garage
Instruments; Bounded Errors of Commission
Method 4-58
4-36. Variation of EC, E0, and FF with CO Cut-Point;
CEV Fleet; Federal Short Cycle; Bounded Errors
of Commission Method 4-62
4-37. Variation of Ec, Eo, and FF with CO Cut-Point;
CEV Fleet; 1975 FTP CO Level =3.4 gm/mi; NY/NJ
Composite Test; Bounded Errors of Commission
Method 4-63
4-38. Variation of EC, E0, and FF with CO Cut-Point;
CEV Fleet; Key Mode Test; Bounded Errors of
Commission Method 4-64
xx
-------
FIGURES (Continued)
4-39. Variation of EC, EQ, and FF with CO Cut-Point;
CEV Fleet; Key Mode Test; Garage Instruments;
Bounded Errors of Commission Method 4-65
4-40. Variation of Ec, E , and FF with CO Cut-Point;
CEV Fleet; 1975 FTP CO Level =3.4 gm/mi;
Federal Three-Mode Test; Bounded Errors of
Commission Method 4-66
4-41. Variation of EC, EQ, and FF with CO Cut-Point;
CEV Fleet; 1975 FTP Level =3.4 gm/mi; Federal
Three-Mode Test; Garage Instruments; Bounded
Errors of Commission Method 4-67
4-42. Variation of E , EQ, and FF with CO Cut-Point;
CEV Fleet; Unloaded 2500 rpm Test; Bounded
Errors of Commission Method 4-68
4-43. Variation of E_, EQ, and FF with CO Cut-Point;
CEV Fleet; Unloaded 2500 rpm Test; Garage
Instruments; Bounded Errors of Commission
Method 4-69
4-44. Variation of Ec, EQ, and FF with NOX Cut-Point;
CEV Fleet; Federal Short Cycle Test; Bounded
Errors of Commission Method 4-73
4-45. Variation of EC, EQ, and FF with NOX Cut-Point;
CEV Fleet; NY/NJ Composite Test; Bounded Errors
of Commission Method 4-74
4-46. Variation of EC, EQ, and FF with NOX Cut-Point;
CEV Fleet; Key Mode Test; Bounded Errors of
Commission Method 4-75
4-47. Variation of Ec, Eo, and FF with NOX Cut-Point;
CEV Fleet; Federal Three-Mode Test; Bounded
Errors of Commission Method 4-76
4-48. Variation of Ec, Eo, and FF with NOX Cut-Point;
CEV Fleet; Unloaded 2500 rpm Test; Bounded
Errors of Commission Method 4-77
xxi
-------
FIGURES (Continued)
4-49. Variation of E and EQ for Key Mode and Weighted
Key Mode Tests; CEV Fleet; 1975 FTP HC Level =
0.41 gm/mi; Bounded Errors of Commission Method .... 4-80
4-50. Variation of Eo and EC for Key Mode and Weighted
Key Mode Tests; CEV Fleet; 1975 FTP CO Level =
3.4 gm/mi; Bounded Errors of Commission Method 4-81
4-51. Variation of EQ and Ec for Key Mode and Weighted
Key Mode Tests; CEV Fleet; 1975 FTP NOX Level =
3. 1 gm/mi; Bounded Errors of Commission Method .... 4-82
4-52. Variability of Predicted Population Results 4-85
4-53. Computation Flow Chart 4-86
4-54. Variation of Actual Ec, Eo, and FF with Predicted EC;
Federal Short Cycle; Three-Constituent Test; Bounded
Errors of Commission Method; CEV Fleet; 1975
FTP Level 1 4-88
4-55. Variation of Actual E^ Eo, and FF with Predicted EC;
Federal Short Cycle; Three-Constituent Test; Bounded
Errors of Commission Method; CEV Fleet; 1975
FTP Level 2 4-88
4-56. Variation of Actual E^ EQ, and FF with Predicted Ec;
Federal Short Cycle; Three-Constituent Test; Bounded
Errors of Commission Method; CEV Fleet; 1975
FTP Level 3 4-89
4-57. Variation of Actual E^ E0, and FF with Predicted Ec;
Federal Short Cycle; Three-Constituent Test; Bounded
Errors of Commission Method; CEV Fleet; 1975
FTP Level 4 4-89
4-58. Variation of Actual Ec, Eo, and FF with Predicted Ec;
Federal Three-Mode; Laboratory Instruments; Three-
Constituent Test; Bounded Errors of Commission
Method; CEV Fleet; 1975 FTP Level 1 4-90
4-59. Variation of Actual Ec, EQ, and FF with Predicted Ec;
Federal Three-Mode; Laboratory Instruments; Three-
Constituent Test; Bounded Errors of Commission
Method; CEV Fleet; 1975 FTP Level 2 4-90
xxn
-------
FIGURES (Continued)
4-60. Variation of Actual Ec, EQ, and FF with Predicted Ec;
Federal Three-Mode; Laboratory Instruments; Three-
Constituent Test; Bounded Errors of Commission
Method; CEV Fleet; 1975 FTP Level 3 4-91
4-61. Variation of Actual EC, EQ, and FF with Predicted EC;
Federal Three-Mode; Laboratory Instruments; Three-
Constituent Test; Bounded Errors of Commission
Method; CEV Fleet; 1975 FTP Level 4 4-91
4-62. Variation of Actual EC, EQ, and FF with Predicted EC;
Federal Three-Mode; Garage Instruments; Three-
Constituent Test; Bounded Errors of Commission
Method; CEV Fleet; 1975 FTP Level 1 4-92
4-63. Variation of Actual Ec, Eo, and FF with Predicted EC;
Federal Three-Mode; Garage Instruments; Three-
Constituent Test; Bounded Errors of Commission
Method; CEV Fleet; 1975 FTP Level 2 4-92
4-64. Variation of Actual EC, EQ, and FF with Predicted EC;
Federal Three-Mode; Garage Instruments; Three-
Constituent Test; Bounded Errors of Commission
Method; CEV Fleet; 1975 FTP Level 3 4-93
4-65. Variation of Actual EC, EQ, and FF with Predicted Ec;
Federal Three-Mode; Garage Instruments; Three-
Constituent Test; Bounded Errors of Commission
Method; CEV Fleet; 1975 FTP Level 4 4-93
5-1. Variation of Ec, Eo, and FF with HC and NOX Cut-
Point; 1974 Model Year Fleet; Federal Short Cycle
Test; Bounded Errors of Commission Method 5-11
5-2. Variation of EC, EQ, and FF with HC and NOX Cut-
Point; 1974 Model Year Fleet; NY/NJ Composite
Test; Bounded Errors of Commission Method 5-12
5-3. Variation of Ec, Eo, and FF with HC Cut-Point;
1974 Model Year Fleet; Key Mode Test; Bounded
Errors of Commission Method 5-13
XXlll
-------
FIGURES (Continued)
5-4. Variation of EC, E , and FF with HC Cut-Point;
1974 Model Year Fleet; Key Mode Test; Garage
Instruments; Bounded Errors of Commission
Method 5-14
5-5. Variation of Ec, Eo, and FF with HC Cut-Point; 1974
Model Year Fleet; Federal Three-Mode Test; Bounded
Errors of Commission Method 5-15
5-6. Variation of Ec, Eo, and FF with HC Cut-Point; 1974
Model Year Fleet; Federal Three-Mode Test; Garage
Instruments; Bounded Errors of Commission Method .... 5-16
5-7. Variation of Ec, Eo, and FF with HC and NOX Cut-
Point; 1974 Model Year Fleet; Unloaded 2500 rpm
Test; Bounded Errors of Commission Method 5-17
5-8. Variation of EC, E , and FF with HC Cut-Point;
1974 Model Year Fleet; Unloaded 2500 rpm Test;
Garage Instruments; Bounded Errors of Commission
Method 5-17
5-9. Variation of EC, Eo, and FF with CO Cut-Point;
1974 Model Year Fleet; Federal Short Cycle Test;
Bounded Errors of Commission Method 5-20
5-10. Variation of Ec, EQ, and FF with CO Cut-Point;
1974 Model Year Fleet; NY/NJ Composite Test;
Bounded Errors of Commission Method 5-21
5-11. Variation of Ec, EQ, and FF with CO Cut-Point;
1974 Model Year Fleet; Key Mode Test; Bounded
Errors of Commission Method 5-22
5-12. Variation of Ec, Eo, and FF with CO Cut-Point;
1974 Model Year Fleet; Key Mode Test; Garage
Instruments; Bounded Errors of Commission Method. . . . 5-23
5-13. Variation of Ec, Eo, and FF with CO Cut-Point;
1974 Model Year Fleet; Federal Three-Mode Test;
Bounded Errors of Commission Method 5-24
XXIV
-------
FIGURES (Continued)
5-14. Variation of Ec, E0, and FF with CO Cut-Point;
1974 Model Year Fleet; Federal Three-Mode Test;
Garage Instruments; Bounded Errors of Commission
Method 5-25
5-15. Variation of EC, E , and FF with CO Cut-Point;
1974 Model Year Fleet; Unloaded 2500 rpm Test;
Bounded Errors of Commission Method 5-26
5-16. Variation of EC, E and FF with CO Cut-Point;
1974 Model Year Fleet; Unloaded 2500 rpm Test;
Garage Instruments; Bounded Errors of Commission
Method 5-27
5-17. Variation of EC, EQ, and FF with NOX Cut-Point;
1974 Model Year Fleet; Key Mode Test; Bounded
Errors of Commission Method 5-29
5-18. Variation of Ec, Eo, and FF with NOX Cut-Point;
1974 Model Year Fleet; Federal Three-Mode Test;
Bounded Errors of Commission Method 5-30
5-19. Variation of EC, Eo, and FF with Instrument Type;
HC; 1974 Model Year Fleet; Key Mode Test; High
Speed Mode; Bounded Errors of Commission Method .... 5-33
5-20. Variation of Ec, Eo, and FF with Instrument Type;
CO; 1974 Model Year Fleet; Key Mode Test; High
Speed Mode; Bounded Errors of Commission Method .... 5-34
5-21. Variation of Ec, Eo, and FF with Instrument Type;
HC; 1974 Model Year Fleet; Key Mode Test; Low
Speed Mode; Bounded Errors of Commission Method .... 5-35
5-22. Variation of Ec, Eo, and FF with Instrument Type;
CO; 1974 Model Year Fleet; Key Mode Test; Low
Speed Mode; Bounded Errors of Commission Method .... 5-36
5-23. Variation of Ec, Eo, and FF with Instrument Type;
HC; 1974 Model Year Fleet; Key Mode Test; Idle
Mode; Bounded Errors of Commission Method 5-37
xxv
-------
FIGURES (Continued)
5-Z4. Variation of Ec, Eo, and FF with Instrument Type;
CO; 1974 Model Year Fleet; Key Mode Test; Idle
Mode; Bounded Errors of Commission Method 5-38
5-25. Variation of Ec, Eo, and FF with Instrument Type;
HC; 1974 Model Year Fleet; Federal Three-Mode Test;
High Speed Mode; Bounded Errors of Commission
Method 5-39
5-26. Variation of Ec, E0, and FF with Instrument Type;
CO; 1974 Model Year Fleet; Federal Three-Mode Test;
High Speed Mode; Bounded Errors of Commission
Method 5-40
5-27. Variation of Ec, Eo, and FF with Instrument Type;
HC; 1974 Model Year Fleet; Federal Three-Mode Test;
Low Speed Mode; Bounded Errors of Commission
Method 5-41
5-28. Variation of Ec, Eo, and FF with Instrument Type;
CO; 1974 Model Year Fleet; Federal Three-Mode Test;
Low Speed Mode; Bounded Errors of Commission
Method 5-42
5-29. Variation of Ec, Eo, and FF with Instrument Type;
HC; 1974 Model Year Fleet; Federal Three-Mode Test;
Idle Mode; Bounded Errors of Commission Method 5-43
5-30. Variation of Ec, Eo, and FF with Instrument Type;
CO; 1974 Model Year Fleet; Federal Three-Mode Test;
Idle Mode; Bounded Errors of Commission Method 5-44
5-31. Variation of Ec, Eo, and FF with Instrument Type;
HC; 1974 Model Year Fleet; Unloaded 2500 rpm Test;
Bounded Errors of Commission Method 5-45
5-32. Variation of Ec, EQ, and FF with Instrument Type;
CO; 1974 Model Year Fleet; Unloaded 2500 rpm Test;
Bounded Errors of Commission Method 5-46
5-33. Variation of Actual EC, Eo, and FF with Predicted
Ec; Federal Short Cycle; Three-Constituent Test;
Bounded Errors of Commission Method; 1974 Model
Year Fleet 5-48
XXVI
-------
FIGURES (Continued)
5-34. Variation of Actual Ec, E0, and FF with Predicted
Ec; Federal Three-Mode; Three-Constituent Test;
Laboratory Instruments; Bounded Errors of Com-
mission Method; 1974 Model Year Fleet 5-49
5-35. Variation of Actual Ec, Eo, and FF with Predicted
Ec; Federal Three-Mode; Three-Constituent Test;
Garage Instruments; Bounded Errors of Commission
Method; 1974 Model Year Fleet 5-50
5-36. Impact of Percent Population Sampled on CO Removed
(Illustrative Example Only) 5-56
6-1. Variation of Ec, Eo, and FF with HC Cut-point;
Original CEV Fleet; Key Mode Test; 1975 FTP
Level = 0.41 gm/mi; Bounded Errors of Commission
Method 6-8
6-2. Variation of Ec, Eo, and FF with HC Cut-point;
Defect Tests Only; Key Mode Test; 1975 FTP
Level = 0. 41 gm/mi; Bounded Errors of Commission
Method 6-8
6-3. Variation of Ec, Eo, and FF with CO Cut-point;
Original CEV Fleet; Key Mode Test; 1975 FTP
Level = 3.4 gm/mi; Bounded Errors of Commission
Method 6-9
6-4. Variation of Ec, Eo, and FF with CO Cut-point;
Defect Tests Only; Key Mode Test; 1975 FTP
Level =3.4 gm/mi; Bounded Errors of Commission
Method 6-9
6-5. Variation of Ec, Eo, and FF with NOX Cut-point;
Original CEV Fleet; Key Mode Test; 1975 FTP
Level = 3.1 gm/mi; Bounded Errors of Commission
Method 6-10
6-6. Variation of Ec, E0, and FF with NOX Cut-point;
Defect Tests Only; Key Mode Test; 1975 FTP
Level = 3.1 gm/mi; Bounded Errors of Commission
Method 6-11
XXVll
-------
1. SUMMARY
-------
1. SUMMARY
A series of statistical analyses was performed to determine
the degree of "correlation" that exists between five specific short tests (STs)
and the federal emission certification test procedure (FTP) for new vehicles.
This work was performed to determine if "reasonable correlation with certi-
fication test procedures" exists; this is a condition precedent to the promulga-
tion of regulations that impose the in-use warranty provisions of Sec. 207 (b)
of the Clean Air Act of 1970 upon the motor vehicle manufacturers.
The basis for the analyses was ST and FTP test data from
three vehicle fleets:
A catalyst-equipped experimental vehicle fleet (40 vehicles)
An in-use 1974 model year vehicle fleet (147 vehicles)
A catalyst-equipped defect test fleet (5 vehicles)
Each of the vehicles in these fleets was tested by the FTP and the following STs:
Federal Short Cycle
New York/New Jersey (NY/NJ) Composite
Clayton Key Mode
Federal Three-Mode
Unloaded 2500 rpm
The first two of these STs are CVS (constant volume sampling) or bag-type
tests wherein a test technician drives the car on the dynamometer in accord-
ance with a prescribed driving pattern. The vehicle exhaust is diluted by
the CVS procedure, and a single sample bag of diluted exhaust is collected
for the ST. The latter three STs are categorized as modal or volumetric.
In these tests, the test technician operates the vehicle on a dynamometer at
a fixed vehicle speed and dynamometer load, or unloaded at a fixed engine
rpm, or at idle. The vehicle tailpipe exhaust is sampled directly, and the
concentration of each pollutant is measured and recorded. The Clayton Key
Mode and the Federal Three-Mode STs each have high-speed, low-speed,
1-1
-------
and idle modes. The Unloaded 2500 rpm ST is a high-speed test with the
transmission in neutral at 2500 engine rpm.
Hydrocarbon (HC) and carbon monoxide (CO) measurements
were recorded with both laboratory analyzers and garage-type instruments
for most of the volumetric tests (Key Mode, Federal Three-Mode, Unloaded
2500 rpm). All oxides of nitrogen (NO ) measurements were made with
laboratory analyzers.
Two different statistical analysis methods were used to assess
"correlation" --a conventional correlation analysis, and a contingency table
analysis.
The principal results of the study are summarized in the
following sections. Because of the many variables involved (three test fleets,
five STs, three emission constituents, two types of measurement instruments,
etc.), the results are presented first as a function of fleet type; then overview
statements or findings are presented which provide more general conclusions,
where appropriate.
1. 1 CATALYST-EQUIPPED EXPERIMENTAL
VEHICLE FLEET
1.1.1 Direct Relatability Results
A conventional correlation analysis was performed for the
catalyst-equipped vehicle (CEV) fleet for each of the five short tests; a sum-
mary of the ST/FTP correlation coefficients obtained is given in Table 1-1.
The correlation coefficient (r) is the quantitative measure of relatability
between the results of the short test and the FTP- The closer r is to 1, the
better the relation. No relationship is indicated by r = 0. Negative r indi-
cates an inverse relation between the observed test results. For a test
sample size (N) of 40 or 39, a computed correlation of less than 0.35 indicates
that the ST and the FTP pollutants are uncorrelated with 95 percent confidence.
For N = 25 or 26, this threshold is approximately 0.4.
1-2
-------
Table 1-1. ST/FTP Correlation Summary (CEV Fleet)
Short Test
Federal Short Cycle
NY/NJ Composite
Key Mode
(Laboratory)
Key Mode
(Garage)
Federal Three -Mode
(Laboratory)
Federal Three -Mode
(Garage)
2500 rpm
(Laboratory)
2500 rpm
(Garage)
Good
Data
Set(a)
First
Second
Average
First
Second
Average
First
Second
First
Second
First
Second
First
Second
First
Second
First
Second
Test
Mode
High
Low
Idle
High
Low
Idle
High
Low
Idle
High
Low
Idle
High
Low
Idle
High
Low
Idle
High
Low
Idle
High
Low
Idle
N(b)
39
25
39
39
25
40
40
26
40
26
31
26
40
26
40
26
40
26
"r" -ST/FTP Correlation'0*
Coefficient
HC
0.87
0.91
0.93
0.92
0.92
0.95
0.61
0.53
0.92
0.57
0.53
0.97
0.73
0.73
0.88
0.51
0.39*
0.32*
0.87
0.79
0.80
0.68
0.52
0.94
0.76
0.73
0.78
0.69
0.42
0.62
0.47
0.37*
0.50
0.36*
CO
0.81
0.42
0.83
0.77
0.71
0.68
0.26*
0.39
0.54
0.30*
0.31*
0.40
0.37
0.21*
0.52
0.08*
0.09*
-0.03*
0.08*
0.22*
0.48
0.20*
0.27*
0.34*
0.24*
0.21*
0.52
0. 12*
0.03*
0.39*
0.30*
0.25*
0. 14*
0. 25*
N0x
0.62
0.47
0.53
0.61
0.51
0.61
0.79
0.20*
0.27*
0.86
0 . 04*
0 . 04*
0.89
0.03*
0. 13*
0.92
-0.28*
0.08*
0.23*
0.23*
(a) First Good Data: This data set contains the observations of the first FTP and ST,
both of which are valid.
Second Good Data: This data set contains the second pair of FTP and ST obser-
vations, both of which are valid.
Average Data: This data set contains the average of the FTP and ST observations
on each car (for the Federal Short Cycle and NY/NJ Composite only).
(b) Number of cars in data set
(c) The correlation is statistically significant at the 95% confidence level except when
indicated by an asterisk.
1-3
-------
1.1.1.1 Hydrocarbon Emission
The bag-type STs (Federal Short Cycle and NY/NJ Composite)
and the idle mode of the volumetric or modal tests (Key Mode, Federal Three-
Mode) in general show superior HC tracking characteristics. However, on
the Federal Three-Mode, the high-speed mode has a slightly higher correla-
tion in some instances than the idle mode. The Unloaded 2500 rpzn ST has
much poorer HC correlation.
1.1.1.2 Carbon Monoxide Emission
The bag-type STs exhibit the superior CO tracking character-
istics, but of a lower correlation level than that achieved for HC. The idle
mode of the volumetric tests has higher correlation than the high and low
speed modes, but with a rather poor correlation coefficient level. The
Unloaded 2500 rpzn ST is essentially uncorrelated for CO.
1.1.1.3 Oxides of Nitrogen Emission
The high-speed modes of the volumetric tests display the best
ability to track NO . The bag-type tests correlate with NO , but at a much
3fc Ji
lower coefficient level. The idle and low-speed volumetric modes and the
Unloaded 2500 rpm ST are uncorrelated with NO .
Ji
1.1.1.4 Modal vs Bag Tests
On the basis of HC and CO correlation, as noted above, the
bag tests (Federal Short Cycle and NY/NJ Composite) are preferable to the
modal- or volumetric-type ST. The volumetric STs show acute deficiencies
in tracking CO. However, the high-speed modes of the volumetric ST have
superior NO correlation.
JL
An analysis of variance indicated that the percent error due to
testing was higher for bag tests than many of the modal tests (using the same
laboratory instruments). This higher testing error may be due to variations
of vehicle operation while trying to follow the driving profile of the short
driving test procedure, rather than due to the bag collection method, per se.
The lower testing error of the volumetric tests, on the other hand, may be
1-4
-------
due to the simplicity of the test procedure itself, in that the measurements
are taken at stabilized engine operating conditions.
1.1.1.5 Weighted Modal Tests
A multiple regression analysis was performed for the three-
mode volumetric tests on the first good data set. The purpose of this analysis
was to empirically determine the linear combinations of like constituents of
the three-mode readings that have maximum correlation with the FTP. The
results are shown in Table 1-2, along with the maximum correlation using
only a single reading on each constituent. As can be seen from the table,
the weighted combination correlations are not significantly greater than the
correlation of the best single reading.
1. 1. 1. 6 Laboratory Analyzers vs Garage Instruments
The largest differences between the correlation results of the
two measurement techniques occur on the second good data sets. However,
the sample size of the second good data set, 26 cars, is risky for inference
purposes. In general, there is a greater variation in the correlation esti-
mates of first good data and second good data for the garage analyzer than
for the laboratory analyzer, as shown in Table 1-1.
The most striking difference between laboratory and garage
data is for HC on the Federal Three-Mode. The laboratory measurements for
first good data indicate the best mode to be high speed, while the correspond-
ing garage instrument readings indicate the idle mode as superior. This is
inconsistent with the results for HC on the Clayton Key Mode where both
instrument types indicated the idle mode as superior. Firm inferences are
tenuous due to differences in sample size.
CO correlation deficiency is common to both measurement
techniques. Due to the low concentration of CO being emitted in the CEV
fleet, this may be a measurements problem, in general, rather than a defi-
ciency in ST structure.
1-5
-------
Table 1-2. Correlations for Weighted Mode Tests (CEV Fleet)
Short Test
Key Mode
Laboratory
Garage
Federal Three -Mode
Laboratory
Garage
N(a)
40
40
31
40
Weighted Correlation(b)
Coefficient
HC
0.93
0.91
0.91
0.81
CO
0.55
0.58
0.48
0.53
N0x
0.83
0.90
Best Single -Mode^c'.,.
Correlation Coefficient* '
HC
0.92 (I)
0.88 (I)
0.87 (H)
0.78 (I)
CO
0.54 (I)
0.52 (I)
0.48 (I)
0.52 (I)
N0x
0.79 (H)
0.89 (H)
(a) Number of cars in first good data set; the first pair of FTP and ST observations, both
of which are valid
(b) Correlations are statistically significant at the 95% confidence level
(c) H = high speed mode
I = idle mode
-------
1.1.1.7
Sensitivity of Correlation Re suit s
Selected extreme data points were deleted and the correlation
coefficient recalculated for the Federal Short Cycle ST to illustrate variability
due to the data sample. As shown in Table 1-3, the correlation coefficient
is extremely sensitive to a small percentage of the data points.
Table 1-3. Correlations for Selected Car Deletions:
Federal Short Cycle vs FTP (CEV Fleet)
Number of Cars Deleted
0
1
2
3
4
(a)
Correlation Coefficient*
HC
0.872
0.657
0.656
--
--
CO
0.810
0.673
0.639
--
--
NO
X
0.621
0.690
0.633
0.823
0.755
(a) Significant at the 95% confidence level
1. 1. 1.8
ST Correlation Ratings
The following qualitative rating scale was used to rate the ST:
Rating
(U) Unacceptable
(P) Poor
(F) Fair
(G) Good
(E) Excellent
Description
Constituent is uncorrelated at the 95 percent
confidence level
Constituent is correlated at the 95 percent
confidence level, but with correlation less
than 0.6
Correlation between 0.6 and 0.7
Correlation between 0.7 and 0.9
Correlation between 0.9 and 1.0
1-7
-------
For rating the three-mode volumetric ST, the mode with the
highest rating was used. Table 1-4 shows the ratings of the ST on each
pollutant on this basis.
Table 1-4. ST Correlation Ratings (CEV Fleet)
Short Test
Federal Short Cycle
NY/NJ Composite
Key Mode
Laboratory
Garage
Federal Three -Mode
Laboratory
Garage
2500 rpm Unloaded
Laboratory
Garage
Rating
HC
G
E
E (I)(a)
G (I)
G(H)
G (I)
P
P
CO
G
G
P (I)
P (I)
P(I)
P (I)
U
U
NOX
F
P
G(H)
G(H)
U
(a) I = idle mode, H = high speed mode
In general, the STs have less difficulty tracking HC than CO
and NO . Excluding the Unloaded 2500 rpm ST (which has either "P" or "U"
j£
ratings for all three pollutants), the bag-type and modal STs all have "G" to
"E" ratings for HC. In the case of CO, the bag-type STs have "G" ratings,
whereas the modal STs are rated in the "P" category. This situation is re-
versed in the case of NO , where the modal STs have "G" ratings and the
bag-type STs are rated "F" to "P". Hence, the choices among the STs for
CO and NO implementation may be more limited than for HC.
1-8
-------
1. 1.2
Contingency Table Analysis Results
The contingency table analysis technique was used to establish
the ST pass-fail levels for each pollutant. The contingency table is defined in
Table 1-5, along with its associated parameters. A pictorial demonstration
of its application to a given data set is shown in Figure 1-1. It can be seen
that, for a given data set, part of the analysis is concerned with the criteria
used to select the ST cut-points. In this regard, the bounded errors of com-
mission method was used extensively to establish trends for the variations in
E , E , FF, and PP. In this method, the ST cut-points are selected to mini-
c o c
mize E while holding the E below a specified level. It thus permits a direct
Table 1-5. Contingency Table
II .u
T3 M
9J O
OH w
Pass
Fail
Total
True = FTP
Pass
a
c
a + c
Fail
b
d
b + d
Total
a + b
c + d
n = a + b
+ c + d
a = number of correctly passed vehicles (PP)
b = number of errors of omission (E )
c = number of errors of commission (E )
d = number of correctly failed vehicles (FF)
Sensitivity = a/(a + c)
Specificity = b/(b + d)
False positive error = b/(a + b)
False negative error = c/(c + d)
Correlation index =
ad - be
[(a + b)(a + c)(b + d)(c + d)]
1/2
1-9
-------
FF, CORRECTLY
FAILED VEHICLES
Er, ERROR OF
c
COMMISSION
ST CUT-POINT
t
oc
to
a
Ji uui rv/im 1
PP, CORRECTLY
PASSED VEHICLES
O_
t
EQ| ERROR OF
OMISSION
FTP MEASUREMENT
Figure 1-1. Contingency Table Representation
1-10
-------
answer to the question, "For a given permissible level of E , what level of
c
EQ is associated with the ST, and with what impact on air quality (inferred
from number of FF and EQ vehicles)?" This method is illustrated in Fig-
ure 1-2. The policy decision is the maximum allowable E .
c
With regard to procedural technique, a bivariate normal dis-
tribution model was fitted to a particular data set by incorporating the corre -
lation coefficient, mean values, and standard deviations of the data set. The
ST cut-points were then determined by using the model for the predicted
population of the CEV fleet.
As the appropriate FTP standards to which the CEV fleet was
designed were uncertain, four sets of FTP cut-points were used in the analysis,
as specified in Table 1-6. The bound of the errors of commission was varied
from 5 percent to 1 percent in 1-percent increments, with the values 0.5
percent and 0. 1 percent also included.
1. 1.2.1
Hydrocarbon Emission
The variation of E E , and FF as a function of HC cut-point
C O
was graphically determined for each ST examined. The results for the Federal
Short Cycle are shown in Figure 1-3 to indicate the general nature of the
Table 1-6. Assumed FTP Levels for CEV Fleet
Level
I
II
in
IV
Emission Levels, gm/mi
HC
0.41
0.60
0.75
0.90
CO
3.4
5.0
7.0
9-0
NO
X
3. 1
3. 1
3. 1
3.1
1-11
-------
MINIMIZE EA SUBJECT TO Er < /%
o \f
I
CO
NOT TO
EXCEED Y%
ST
CUT-POINT , '
MINIMIZE
Q.
t
FTP MEASUREMENT
Figure 1-2. Bounded Errors of Commission Method
1-12
-------
0.3 0.4
0.5
0.6 0.7 0.8
HC CUT-POINT, gm/mi
1.0 1.1
Figure 1-3. Variation of Ec, Eo, and FF with HC Cut-Point;
CEV Fleet; Federal Short Cycle; Bounded Errors of Com-
mission Method
1-13
-------
tradeoffs available for policy formulation. Reducing the EC increases EQ
and decreases FF. All STs had similar trends. To illustrate specific values
and trends among the STs, Tables 1-7 and 1-8 summarize data from the
graphical displays at HC FTP levels of 0.41 and 0.90. On the average, at
both FTP levels, the bag tests (Federal Short Cycle and NY/NJ Composite)
have lower E and higher FF at the fixed E = 5 percent than do the volumetric
tests. However, the idle mode of the Clayton Key Mode (with either laboratory
or garage instruments) test produces similar results. The Unloaded 2500 rpm
test is very poor on a comparative basis.
1.1.2.2 Carbon Monoxide Emission
The variation of E , E , and FF as a function of CO cut-point
CO
was also graphically determined for each ST examined, and for the range of
CO FTP values selected in Table 1-6 (CO = 3.4 to 9). Figure 1-4 indicates
results for the Federal Short Cycle. As in the preceding case of hydrocarbon
emissions, these displays indicated the tradeoffs possible between E , E ,
and FF. However, for CO FTP levels above 3.4, the general or average CO
levels of the CEV fleet were sufficiently low; i.e., a very high percentage of
the vehicles exceeded the 5-, 7-. and 9-gm/mi requirements, so that both
E and FF percentage values were very small for all of the short test pro-
cedures. This characteristic is summarized in Table 1-9 for the CO FTP
level of 9 gm/mi; the E and FF values are less than 1 percent for all the STs.
At the 3.4 level, however, as shown in Table 1-10, the bag
tests were sufficiently discriminatory to identify FF values above 20 percent,
with E values in the 14- to 16-percent range. The volumetric tests, on the
other hand, all had high E values (30- to 40-percent range), with very low
FF values (<16).
1.1.2.3 Oxides of Nitrogen Emission
The variations of EC, E t and FF as a function of NO cut-
point were also graphically determined for each ST examined, for the single
NO FTP value of 3. 1 gm/mi examined in the study. Figure 1-5 illustrates
results for the Federal Short Cycle.
1-14
-------
Table 1-7. Comparison of Selected ST Hydrocarbon Results:
CEV Fleet, Bounded Errors of Commission Analysis,
HC FTP Level =0.41 gm/mile (E = constant = 5%)
Short Test
Federal Short Cycle
NY/NJ Composite
Clayton Key Mode (Laboratory)
Idle
Low Speed
High Speed
Clayton Key Mode (Garage)
Idle
High Speed
Federal Three -Mode (Laboratory)
Idle
Low Speed
High Speed
Federal Three -Mode (Garage)
Idle
Low Speed
High Speed
2500 rpm Unloaded (Laboratory)
2500 rpm Unloaded (Garage)
Parameter, %
E
o
11
7
7
35
30
10
21
17
17
11
18
22
20
38
37
FF
56
60
61
32
37
57
45
52
38
51
46
44
47
28
30
1-15
-------
Table 1-8. Comparison of Selected ST Hydrocarbon Results:
CEV Fleet, Bounded Errors of Commission Analysis,
HC FTP Level = 0.90 gm/mile (E = constant = 5%)
Short Test
Federal Short Cycle
NY/NJ Composite
Clayton Key Mode (Laboratory)
Idle
Low Speed
High Speed
Clayton Key Mode (Garage)
Idle
High Speed
Federal Three -Mode (Laboratory)
Idle
Low Speed
High Speed
Federal Three -Mode (Garage)
Idle
Low Speed
High Speed
2500 rpm Unloaded (Laboratory)
2500 rpm Unloaded (Garage)
Parameter, %
E
o
9
6.5
6
21
19.5
9.5
16
15
15
10
14
16
15
23
23
FF
22
25
22
9
10
22
16
21
21
25
17
16
17
8
9
1-16
-------
30
25
20
o
10
EQ
FF
V /
A /
A
*
*
»
1975 FTP
LEVEL
7.0
-9.0
AT 5.0. 7.0. AND 9.0
FF < 5%
CO CUT-POINT, gm/mi
Figure 1-4. Variation of Ec, Eo, and FF with CO Cut-Point;
CEV Fleet; Federal Short Cycle; Bounded Errors of Com-
mission Method
1 17
-------
Table 1-9. Comparison of Selected ST Carbon Monoxide Results:
CEV Fleet, Bounded Errors of Commission Analysis,
CO FTP Level =9.0 gm/mi (E = constant = 5%)
Short Test
Parameter, %
FF
Federal Short Cycle
NY/NJ Composite
Clayton Key Mode (Laboratory)
Idle
Low Speed
High Speed
Clayton Key Mode (Garage)
Idle
Low Speed
High Speed
Federal Three-Mode (Laboratory)
Idle
Low Speed
High Speed
Federal Three-Mode (Garage)
Idle
Low Speed
High Speed
2500 rpm Unloaded (Laboratory)
2500 rpm Unloaded (Garage)
1 18
-------
Table 1-10. Comparison of Selected ST Carbon Monoxide Results:
CEV Fleet, Bounded Errors of Commission Analysis,
CO FTP Level = 3.4 gm/mi (E = constant = 5%)
Short Test
Federal Short Cycle
NY/NJ Composite
Clayton Key Mode (Laboratory)
Idle
Low Speed
High Speed
Clayton Key Mode (Garage)
Idle
Low Speed
High Speed
Federal Three -Mode (Laboratory)
Idle
Low Speed
High Speed
Federal Three -Mode (Garage)
Idle
Low Speed
High Speed
2500 rpm Unloaded (Laboratory)
2500 rpm Unloaded (Garage)
Parameter, %
E
0
14
16
29
33
36
29
36
33
33
40
43
29
36
36
35
38
FF
22
20
15
11
8
14
7
10
16
8
7
14
7
7.5
8
6
1-19
-------
25
20
15
j°
"o
10
o
LU
Q_
FF<5.
0
5 6
NOX CUT-POINT, gm/mi
Figure 1-5. Variation of Ec, Eo, and FF with
NOX Cut-Point; CEV Fleet; Federal Short
Cycle; Bounded Errors of Commission Method
1-20
-------
The significant results at the E level of 5 percent are
summarized in Table 1-11 for comparative purposes. As can be noted, the
high-speed mode of the volumetric tests (Clayton Key Mode and Federal
Three-Mode) produced the highest FF values and the lowest E values, and
are thus indicated to be superior for NO discrimination purposes.
Ji
1.1.2.4 Weighted Three-Mode Tests
Contingency table analyses were also made for two different -
weighted Key Mode tests. The results indicated that the weighted volumetric
tests are not significantly better than the best single mode, as was also con-
cluded from conventional correlation analyses (see Sec. 1. 1. 1.5).
1.1.2.5 Variance Effects
Since the variations in E , E , and FF with ST cut-point noted
co r
previously are predictions from the data, the variability of the predictions
was analyzed. The uncertainty in the predicted results increases when
decreasing the bounds of the errors of commission, as illustrated in Figure 1-6
for HC on the Federal Short Cycle.
1.1.2.6 Laboratory vs Garage Instruments
Tables 1-7 through 1-10 indicate that generally similar levels
of E and FF were obtained with both laboratory and garage analyzers for the
HC and CO ranges examined for the CEV fleet.
1.1.2.7 Modal vs Bag Tests
In terms of HC and CO discrimination, as noted above, the bag
tests are superior to the modal ST. The modal STs all have high EQ and low
FF values. In terms of NO discrimination, the high-speed mode of the
Ji
volumetric ST was superior. These results agree with those predicted from
conventional correlation analysis in Sec. 1. 1. 1.4.
1.1.2.8 Multiple-Constituent Tests
In addition to analyzing each pollutant individually, an analysis
was made for three-constituent tests. In a three-constituent test, a car fails
1-21
-------
Table 1-11. Comparison of Selected ST NO Results:
CEV Fleet, Bounded Errors or Commission Analysis,
NO FTP Level = 3.1 gm/mi (E = constant = 5%)
Short Test
Federal Short Cycle
NY/NJ Composite
Clayton Key Mode (Laboratory)* '
Idle
Low Speed
High Speed
fa)
Federal Three -Mode (Laboratory)* '
Idle
Low Speed
High Speed
2500 rpm Unloaded (Laboratory)^'
Parameter, %
E
o
9.5
10
13
14
6.5
11
11
3
13
FF
5.5
5
2
<2
8.5
1
1
8.5
2
(a) Garage-type analyzers for NO were not available for ST
evaluation.
x
the ST if any of its HC, CO, and NO measurements exceed the previously
Ji
determined cut-points. These tests are applicable to the bag tests, the
unloaded test, and the individual modes of the three-mode volumetric tests.
The three-constituent test results for the Federal Short Cycle
and the Federal Three-Mode (high speed and idle modes only) were computed
and graphically summarized as a function of predicted E . Table 1-12 sum-
marizes these results for the predicted E value of 5 percent. Both the
laboratory and garage instrument results are displayed for the Federal Three-
Mode short test.
1-22
-------
01
I
ERROR BAR INDICATES PLUS AND MINUS ONE
STANDARD DEVIATION OF THE ESTIMATE
234567
HC FEDERAL SHORT CYCLE, gm/mi
Figure 1-6. Variability of Predicted Population Results
1-23
-------
Table 1-12. Comparison of Three-Constituent Test Results:
CEV Fleet, Bounded Errors of Commission Analysis,
(Predicted E = constant = 5%)
Short Test
Federal Short Cycle
Federal Three -Mode
(Laboratory)
Idle
High Speed
Federal Three -Mode
(Garage)
Idle
High Speed
FTP
Level
1
2
3
4
1
2
3
4
1
2
3
4
1
2
3
4
1
2
3
4
Parameter, %
Actual EC
5
8
7.5
7.5
0
3
3
13
3.5
9.5
9.5
13
10
2.5
0
5
10
15
10
5
Actual E
o
16
16
8
6
36
28
16
16
16
13
10
7
36
25
14
13.5
30
18
6
6.5
Actual FF
48
25
20
17
36
17
16
7
54
35
23
16
30
18
16
11.5
36
25
24.5
18
1-24
-------
With laboratory instrument measurements, as the FTP
cut-points increase from level Set I to level Set IV. the resulting actual errors
of commission tend to increase for the given predicted level of errors of com-
mission. This trend is not present for the garage instrument results shown.
A comparison of the modes on the Federal Three-Mode test
shows that, for the fixed predicted percent E , the high speed mode has a
higher percent FF and lower percent E than does the idle mode. This is true
regardless of instrumentation or FTP level. However, the actual percent E
is generally lower on the idle mode than on the high-speed mode, but this
difference is not always significant.
A comparison of different modes or ST should be made on a
fixed actual percent E basis. This is, of course, difficult to do because of
the computational procedure followed. It can be approximately performed,
however. Consider comparing the Federal Short Cycle to the Federal Three-
Mode. At FTP level I, the actual percent E is approximately the same for
the high-speed mode and the Federal Short Cycle (statistically, they are the
same). Now, comparing the percent FF and percent E values, percent FF
and percent E are both higher on the high-speed mode than the Federal Short
Cycle. This difference is not statistically significant at the 95 percent confi-
dence level, and the two tests would have to be judged as equal. Also, at the
95 percent confidence level, the high-speed mode is superior to the idle mode.
The differences between laboratory and garage instruments
are quite predictable, based upon the previous results from individual pollu-
tants. For the fixed predicted percent E , on their respective modes,
a. Actual percent E is higher for garage instruments than for
laboratory instruments
b. Actual percent FF is lower for garage instruments than for
laboratory instruments
c. Actual percent E is higher for garage instruments than for
laboratory instruments.
1-25
-------
1.2 IN-USE 1974 MODEL. YEAR VEHICLE FLEET
1.2.1 Direct Relatability Results
A conventional correlation analysis was made for the 1974
model year fleet to assess direct relatability between the five short tests
and the FTP. The method was the same as described for the CEV fleet in
Sec. 1.1.1. The resulting ST/FTP correlation coefficients for HC, CO,
and NO are summarized in Table 1-13 for the three individual inertia test
ji
weight groups (A = 4000 Ib, B = 2750 Ib, and C = 5500 Ib) and for the pooled
vehicle population (combined groups A, B, and C).
1.2. 1. 1 Hydrocarbon Emission
For the pooled fleet, the bag-type STs (Federal Short Cycle
and NY/NJ Composite), the idle mode of the modal STs with laboratory
analyzers, and the Unloaded 2500 rpm ST with laboratory analyzers in gen-
eral exhibit the better HC tracking characteristics.
For Group A, similar results apply.
For Group B, the results are similar to the pooled fleet
except that in some instances the low-speed mode of the Key Mode and the
low and high-speed modes of the Federal Three-Mode test have a. slightly
higher correlation coefficient than the idle mode.
For Group C, none of the STs are able to track HC with any
reasonably high degree of correlation.
1.2.1.2 Carbon Monoxide Emission
For the pooled fleet, the bag-type STs, the idle and low-speed
modes of the modal tests with laboratory analyzers, and the Unloaded 2500 rpm
ST with laboratory analyzers in general exhibit the better CO tracking
characteristics.
For Groups A and B, similar results apply except that the
low-speed mode is superior to the idle mode in the modal tests.
1-26
-------
Table 1-13. Correlation Coefficient Summary:
1974 Model Year Fleet
Short Test
Federal
Short
Cycle
NY/NJ
Composite
Key Mode
(Laboratory)
Key Mode
(Garage)
Vehicle
Group 'a'
Pooled
A
B
C
Pooled
A
B
C
Pooled
A
B
C
Pooled
A
Test
Mode
High
Low
Idle
High
Low
Idle
High
Low
Idle
High
Low
Idle
High
Low
Idle
High
Low
Idle
N(b)
147
50
48
49
147
50
48
49
147
50
48
49
145
50
ST/FTP Correlation
Coefficient*0)
HC
0.932
0.933
0.897
0.383
0.906
0.9H
0.920
0.513
0.757
0.776
0.793
0.590
0.595
0.723
0.812
0.868
0.825
0.238*
0.228*
0.460
0.528
0.545
0.455
0.228*
0.151*
0.245*
CO
0.905
0.972
0.897
0.476
0.890
0.950
0.857
0.498
0.518
0.769
0.739
0.514
0.827
0.704
0.262*
0.738
0.650
-0.195*
0.435
0.757
0.507
0.472
0.470
0.563
0.652
0.372
NO
X
0.355
0.780
0. 104*
0.674
0.060*
0.733
0.005*
0.611
0.521
0.419
0.463
0.562
0.495
0.381
0.731
0.635
0.548
0.555
0.580
0.571
1-27
-------
Table 1-13. Correlation Coefficient Summary:
1974 Model Year Fleet (Continued)
Short Test
Federal
Three -Mode
(Laboratory)
Federal
Three -Mode
(Garage)
Vehicle
Group'3-'
B
C
Pooled
A
B
C
Pooled
A
Test
Mode
High
Low
Idle
High
Low
Idle
High
Low
Idle
High
Low
Idle
High
Low
Idle
High
Low
Idle
High
Low
Idle
High
Low
Idle
N(b)
46
49
147
50
48
49
145
50
ST/FTP Correlation
Coefficient^)
HC
0.478
0.765
0.692
0. 191*
0.198*
0. 100*
0.766
0.771
0.803
0. 507
0.523
0.709
0.890
0.859
0.851
0.522
0.533
0.252*
0.474
0.531
0.632
0. 138*
0. 107*
0.660
CO
0.362
0.540
0.560
-0.221*
-0.091*
0.229*
0.604
0.729
0.734
0.717
0.801
0.724
0.278*
0.737
0.622
0.159*
0.592
0.733
0.387
0.409
0.476
0.533
0.597
0.397
N0x
0.467
0.453
0.411
0.492
0.664
0.369
0.722
0.611
0.665
0.552
0.707
0.639
.
1-28
-------
Table 1-13. Correlation Coefficient Summary:
1974 Model Year Fleet (Continued)
Short Test
2500 rpm
Unloaded
(Laboratory)
2500 rpm
Unloaded
(Garage)
Vehicle
Group**)
B
C
Pooled
A
B
C
Pooled
A
B
C
Test
Mode
High
Low
Idle
High
Low
Idle
N
46
49
147
50
48
49
147
50
46
49
ST/FTP Correlation
Coefficient^)
HC
0.536
0.763
0.717
0.095*
-0.008*
-0.060*
0. 809
0.832
0.865
0. 107*
0.574
0.487
0.781
-0.064*
CO
0.268*
0.539
0.550
-0.083*
0.239*
0.392
0.740
0.812
0.724
0.350
0.447
0.676
0.684
-0.051*
NO
X
0.447
0.524
0. 577
0.679
(a)A = Chrysler (4000 Ib)
B = Ford (2750 Ib)
C = Chevrolet (5500 Ib)
Pooled = Groups A + B + C
'Number of cars in the data set
(c)
'The correlations are statistically significant at the 95 percent confidence
level except where indicated by an asterisk. ST and FTP uncorrelated
for correlations below 0.28.
1-29
-------
For Group C, the idle mode of the Key Mode and Federal
Three-Mode (with laboratory analyzers) are superior, although the low-
speed modes of these STs and the bag-type tests are correlated with CO at
lower correlation coefficient levels. The other STs are essentially uncor-
related with CO for Group C.
1.2.1.3 Oxides of Nitrogen Emission
For the pooled fleet, all modes of the Key Mode and Federal
Three-Mode and the Unloaded 2500 rpm ST have similar correlations in the
0. 41 and 0. 52 range; the Federal Short Cycle correlates at a lower value
(0. 36), while the NY/NJ Composite bag test is uncorrelated.
For Group A, however, the bag-type STs have the highest
correlation coefficients observed for NO (0.73 to 0.78), while the modal
2£
and Unloaded 2500 rpm ST results are similar to those of the pooled fleet.
For Group B, the results are similar to those for the pooled
fleet except that the bag-type STs are not correlated at all for NO . Here the
J*.
range of correlation coefficients for the modal and Unloaded 2500 rpm STs is
from 0.55 to 0.73, with the highest values obtained in the high-speed mode.
The Group C results are similar to those for Group A.
There is no single ST with good NO correlation across the
JL
1974 model year fleet population.
1.2.1.4 Modal vs Bag Tests
In terms of HC and CO emissions correlation, the bag-type
STs are superior for Groups A, B, and the pooled population. For Group C
the idle mode of the Key Mode and Federal Three-Mode STs has the higher
correlation for CO; the idle mode of the Key Mode and the low- and high-
speed modes of the Federal Three-Mode are essentially the same as the bag-
type STs in terms of HC discrimination capability.
In terms of NO correlation, the bag-type and modal STs are
essentially equivalent for Group C, whereas the bag-type STs are clearly
superior for Group A. The modal tests are superior for Group B and the
pooled fleet.
1-30
-------
1.2.1.5 Laboratory Analyzers vs Garage Instruments
From an HC and CO correlation viewpoint, the garage
analyzers are inferior to the laboratory analyzers in that they have lower
correlation coefficients than the laboratory analyzers for HC and CO in each
corresponding test mode. They do, however, tend to identify the same
superior test modes as the laboratory analyzers, and can have reasonably
high correlation coefficients, although there is a wide variation for the three
groups examined in the 1974 model year fleet.
To illustrate, consider the idle mode of the Federal Three-
Mode ST, which for the pooled fleet resulted in representatively high corre-
lation coefficients for HC and CO with laboratory analyzers: 0.80 and 0.73,
respectively. With garage instruments, these correlation coefficients
dropped to 0.63 and 0.48, respectively.
In the case of Group A, the laboratory analyzer HC and CO
values were 0.71 and 0.72, whereas the garage instrument values were
reduced to 0.66 and 0.40.
Group B HC and CO values for laboratory analyzers were 0. 85
and 0.62. With garage instruments, they were lowered to 0.72 and 0.55.
In the case of Group C, the HC correlation coefficient of 0. 25
with laboratory instruments was not statistically significant at the 95 percent
confidence level, while the CO correlation coefficient was 0.73. With garage
instruments, these values dropped to -0. 06 and 0. 39, respectively. In addi-
tion, all other ST test modes with garage instruments were uncorrelated for
HC and CO for Group C. This was the only group exhibiting these charac-
teristics, although it also had generally poorer HC and CO correlation
coefficients than the other groups when laboratory analyzers were used.
This group-peculiar characteristic raises the issue as to
whether it is related to inertia test weight factors or to vehicle manufacturer,
since each inertia test weight group was made by a different automotive
company. There are insufficient data to evaluate this issue at this time;
however, a comparison can be made between the 2750-lb Pintos of Group B
1-31
-------
above and the 5000-lb Galaxies of the CEV fleet in Section 1. 1, since both
were manufactured by the Ford Motor Company.
Again, using the idle mode of the Federal Three-Mode ST for
comparison purposes, the use of garage instruments instead of laboratory
instruments for the CEV Galaxies reduced the HC correlation coefficient
from 0.80 to 0.78, and increased the CO correlation coefficient from 0.48
to 0.52 (see Table 1-1). These ranges are similar to those reported above
for Group B (Pintos), even though the Galaxies were catalyst-equipped and
the Pintos were not.
Thus, it appears that additional examinations may be required
of possible manufacturer-related effects (e.g., idle fuel-air ratio tolerance
bands and quality control measures) in order to fully understand their impact
upon measurement instrument type for short test purposes.
1. 2. 1.6 ST Ratings
ST ratings, using the scale established for the CEV fleet in
Sec. 1.1.1.8, are given in Table 1-14. As can be seen, no single ST per-
forms consistently well on all three individual groups, or on a pooled basis.
Generally, the STs are unable to track HC and CO emission levels on
Group C.
As with the CEV fleet, the bag-type STs have higher ratings
than the volumetric tests. The Unloaded 2500 rpm ST shows substantially
higher correlation for the 1974 model year fleet than for the CEV fleet (as
shown in Table 1-4). The extreme CO tracking deficiency for the CEV fleet
data is not evident for the 1974 model year fleet.
1.2.2 Contingency Table Analysis Results
A contingency table analysis, using the bounded errors of
commission method described in Sec. 1. 1.2 for the CEV fleet, was also
performed for the 1974 model year fleet, with the results as discussed
below.
1-32
-------
Table 1-14. ST Ratings: 1974 Model Year Fleet
Short Test
Federal Short Cycle
NY/NJ Composite
Key Mode
(Laboratory)
Key Mode
(Garage)
Federal Three -
Mode (Laboratory)
Federal Three -
Mode (Garage)
Vehicle
Group(a)
Pooled
A
B
C
Pooled
A
B
C
Pooled
A
B
C
Pooled
A
B
C
Pooled
A
B
C
Pooled
A
B
C
Ratings (b)
HC
E
E
G
P
E
E
E
P
G (I)(c)
G (I)
G (L)
P (D
P (D
U
G (L)
U
G (I)
G (I)
G (H)
P (L)
F (I)
F (I)
G (L)
U
CO
E
E
E
P
G
E
G
P
G (L)
G (L)
G (L)
G (I)
P (H)
F (L)
P (D
U
G (I)
G (L)
G (I)
G (I)
P (D
P (L)
P (D
P (D
NO
X
P
G
U
F
U
G
U
F
P (H)
P (H)
G (H)
P (L)
P (H)
F (L)
G (H)
G (L)
1-33
-------
Table 1-14. ST Ratings: 1974 Model Year Fleet
(Continued)
Short Test
2500 rpm Unloaded
(Laboratory)
2500 rpm Unloaded
(Garage)
Vehicle
Group la>
Pooled
A
B
C
Pooled
A
B
C
Ratings *b)
HC
G
G
G
U
P
P
G
U
CO
G
G
G
P
P
F
F
U
NO
X
P
P
P
F
(a)A = Chrysler (4000 Ib)
(b)
B = Ford (2750 Ib)
C = Chevrolet (5500 Ib)
Pooled = Groups A + B + C
Rating scale as in Sec. 1.1.1.8
I . idle
L = low speed mode
H = high speed mode
1-34
-------
1.2.2.1 Hydrocarbon Emission
The variation of EC> EQ, and FF as a function of HC cut-point
was graphically determined for each ST. The results for the Federal Short
Cycle are shown in Figure 1-7 to indicate the general trends. All STs had
similar trends. To illustrate specific values and trends among the STs,
Table 1-15 summarizes data for the E value of 5 percent.
The bag tests (Federal Short Cycle and NY/NJ Composite)
have lower EQ and higher FF at the fixed E = 5 percent condition than do
the volumetric tests. There is little difference shown between the various
volumetric STs.
1.2.2.2 Carbon Monoxide Emission
The variations of E , E , and FF as a function of CO cut-point
co r
were also graphically determined. Figure 1-8 indicates results for the
Federal Short Cycle.
To illustrate specific values and trends among the STs,
Table 1-16 summarizes data from the figures for the E value of 5 percent.
The bag-type STs (Federal Short Cycle and NY/NJ Composite)
exhibit excellent CO tracking characteristics; the EQ values are considerably
better (lower) than the volumetric tests, and the FF values are the highest.
When garage-type instruments are used, the E values are essentially doubled
(over laboratory instrument values) and FF values are significantly reduced.
1.2.2.3 Oxides of Nitrogen Emission
Figure 1-7 also indicates the variation of EC, EQ, and FF as
a function of NO cut-point for the Federal Short Cycle.
x c
The significant results for each ST at the EC level of 5 per-
cent are summarized in Table 1-17 for comparative purposes. As can be
noted, all STs identified very low percentages of correctly failed vehicles
(FF), less than 5 percent, while having significant errors of omission,
approximately 15 percent.
1-35
-------
30
25
20
o
at.
uu
10
0
\
*
*
'.HC
\
\
\
X
/
/
/
/
/
/
/HC
/ \
FF
NOV
0
HC AND NOX CUT-POINT, gm/mi
Figure 1-7. Variation of Ec, Eo, and FF with HC and NOX
Cut-Points; 1974 Model Year Fleet; Federal Short Cycle;
Bounded Errors of Commission Method
1-36
-------
Table 1-15. Comparison of ST Hydrocarbon Results: 1974 Model Year
Fleet, Bounded Errors of Commission Analysis
(E = constant = 5%)
Short Test
Parameter, %
FF
Federal Short Cycle
NY/NJ Composite
Clayton Key Mode (Laboratory)
Idle
Low Speed
High Speed
Clayton Key Mode (Garage)
Idle
Low Speed
High Speed
Federal Three-Mode (Laboratory)
Idle
Low Speed
High Speed
Federal Three-Mode (Garage)
Idle
Low Speed
High Speed
2500 rpm Unloaded (Laboratory)
2500 rpm Unloaded (Garage)
6.5
8.5
16
17
18
11.5
14
13
15.5
17.5
18
17
14
12
16
16
34.5
32
24. 5
23.6
22.5
29
27
28
25
23
23
24
27
29
26
26
1-37
-------
651-
60 -
55 -
£ 50
45 -
40 -
»
- 25
UJ
-o 20
z
UJ°
I
~~ O A5
1 1
o.
10
5
Q
N '
X
N /
% X
\ /
\ /
\ /
\ /
A
/ \
f \
\
\
\
X »
X \
X >
x^
i i i r* 1
15 20 25 30 35 40
0
FF
\
45
50
CO CUT-POINT, gm/mi
Figure 1-8. Variation of Ec, Eo, and FF with CO Cut-Point;
1974 Model Year Fleet; Federal Short Cycle; Bounded Errors
of Commission Method
1-38
-------
Table 1-16. Comparison of ST Carbon Monoxide Results: 1974 Model
Year Fleet, Bounded Errors of Commission Analysis
(E = constant = 5%)
c
Short Test
Parameter^ %
E
FF
Federal Short Cycle
NY/NJ Composite
Clayton Key Mode (Laboratory)
Idle
Low Speed
High Speed
Clayton Key Mode (Garage)
Idle
Low Speed
High Speed
Federal Three-Mode (Laboratory)
Idle
Low Speed
High Speed
Federal Three-Mode (Garage)
Idle
Low Speed
High Speed
2500 rpm Unloaded (Laboratory)
2500 rpm Unloaded (Garage)
7
8
19
18
35
35
35
37
20
20
29
35
31
30
19
33
65
64
53
54
38
38
38
35
53
52
43
37
41
42
53
40
1-39
-------
Table 1-17. Comparison of ST NOX Results: 1974 Model Year Fleet,
Bounded Errors of Commission Analysis
(E = constant = 5%)
Short Test
Federal Short Cycle
NY/NJ Composite
Clayton Key Mode (Laboratory)
Idle
Low Speed
High Speed
Federal Three -Mode (Laboratory)
Idle
Low Speed
High Speed
2500 rpm Unloaded (Laboratory)
Parameter, %
E
o
14.5
16.5
13.5
14
13.5
14
14
14
14
FF
3
1.5
<5
<5
<5
<5
<5
<5
4
1.2.2.4 Single-Constituent Test Results
On the average, the bag-type tests have lower E and higher
FF for a fixed rate of E than do the volumetric tests. However^ FF rates
in the 30 percent range can be achieved with any of the tests. For a fixed
percent FF, the percent E is determined since the sum of FF and E is the
FTP rejection rate. Thus, the "best" test for fixed percent FF is the one
with the lowest percent E . In general, the bag-type STs are better in this
respect. However, the actual level of percent E on the volumetric tests is
still quite low. For example, at 30 percent FF on the CO Federal Short
Cycle, the percent E is essentially zero. For CO on the Key Mode low-
speed mode, percent E is 0.65 percent for laboratory instruments and
3. 85 percent for garage instruments.
1-40
-------
1.2.2.5
Multiple-Constituent Tests
In addition to analyzing each pollutant individually, an analysis
was made for multiple-constituent tests. The method of analysis and compu-
tational procedures were the same as for the CEV fleet, as discussed in
Sec. 1. 1.2.8.
Three-constituent test results for the Federal Short Cycle
and the Federal Three-Mode (high-speed and idle modes only) were com-
puted and graphically summarized as a function of predicted E . Table 1-18
summarizes these results using laboratory instruments for predicted E
values <2 percent.
For the actual percent E less than 2 percent, the laboratory
results of the Federal Three-Mode and the Federal Short Cycle are com-
parable. Table 1-18 indicates the minimum and maximum for percent FF
and percent E , while percent E is less than 2 percent. There is little
difference between the idle mode and the Federal Short Cycle. Over this
range of percent E , the idle mode would appear favorable to the Federal
Short Cycle due to the lower value of percent E on the idle mode.
Table 1-18. ST Comparison: 1974 Model Year Fleet,
Multiple Constituent Tests
(Actual E S 2%)
Short Test
Federal Short Cycle
Federal Three -Mode
Idle
High
% FF
Min
25
22
5
Max
36
38
42
% E
o
Min
44
42
38
Max
55
58
75
A comparison of instrument types showed that the laboratory
instruments are generally preferable.
1-41
-------
1. 3 DEFECT DATA FROM CATALYST-EQUIPPED
EXPERIMENTAL VEHICLE FLEET
1.3.1 Nature of Defects and Statistical Impact
Upon completion of the FTP and ST tests performed on the
CEV fleet as described in Sec. 1. 1, 95 defect tests were performed on 5 of
the vehicles from the 40-vehicle CEV fleet.
The 95 defect tests simulated a wide variety of malfunctions
which could occur in typical passenger cars. The general categories of
defects are defective ignition components; changes in ignition timing, dwell,
and spark advance; faulty carburetion; defective valves; clogged air filters;
and faulty emission control components. The defects were introduced indi-
vidually and mixed.
Correlation analyses were performed to determine the statis-
tical character of the defect test data. Many of the defect tests were either
replications or produced similar data. The HC correlations are consistently
higher, over 0. 9, among the defect data than the previous 40-car CEV fleet.
Addition of all defect data to the original CEV fleet data would significantly
distort the population characteristics with regard to HC. CO and NO dis-
x
tortion would also occur, although not as pronounced as with HC.
As the assumption of independence of the observations is
crucial to contingency table analysis, the 95 defect tests were statistically
pruned to 24 tests representing 24 independent defective vehicles. These
data are considered to represent a population distinct from the original
40-car population. Of these 24, 6 had no Federal Three-Mode (laboratory)
data, and 5 had no Key Mode (laboratory) data.
1.3.2 Contingency Table Analysis Results
The analysis proceeded in two stages. The original CEV fleet
population was first analyzed, using first good data. The analysis method
was the bounded errors of commission procedure which established the ST
1-42
-------
cut-points (see Sec. 1. 1.2). Percent E was varied from 10 percent to
1 percent in 1-percent increments, with the addition of points at 0. 5 percent
and 0. 1 percent. Immediately following analysis of the original CEV fleet,
the defect population was analyzed. The contingency table results were cal-
culated for this population using the cut-points previously determined from
the original CEV fleet population. The computations were performed at
each of the EC settings. Thus, the analysis is merely an assessment of how
well a test constructed using data with an unknown mix of normal and defect
operation will perform on a population of defective vehicles known to represent
extreme departures from normal operation. A summary of the analysis on
each constituent is given in Table 1-19. The ST cut-points were established
for E less than or equal to 5 percent, and the FTP level was Level I.
(HC = 0.41 gm/mi, CO = 3.4 gm/mi, NO = 3. 1 gm/mi).
Ji
1.3.2. 1 Hydrocarbon Emission
In all cases, each ST produced significantly higher FF values
and lower EQ values for the defect fleet than for the original CEV fleet. The
percent E for the defect fleet was generally lower and varied from 0. 97 to
8. 68 percent, as compared with the 5 percent level used in the CEV fleet to
select the HC cut-point values.
1.3.2.2 Carbon Monoxide Emission
Each ST produced significantly higher FF values and lower E
values for the defect fleet than for the original CEV fleet, except for the
Unloaded 2500 rpm ST with garage analyzers, where the E values were
essentially the same. The percent E for the defect fleet was generally
somewhat higher than the 5 percent level used in the CEV fleet, varying
from 4.48 to 16. 5 percent.
1.3.2.3 Oxides of Nitrogen Emission
Both FF and E values were significantly higher for the defect
fleet than for the original CEV fleet, for each ST. The percent E for the
1-43
-------
Table 1-19. Defect Analysis Comparison Summar
Population [% E = 5, (a) FTP Level
Predicted
Short Test
Federal Short Cycle
NY/NJ Composite
Key Mode (Laboratory)
Key Mode (Garage)
Federal Three-Mode
(Laboratory)
Federal Three -Mode
(Garage)
2500 rpm Unloaded
(Laboratory)
2500 rpm Unloaded
(Garac'-l
Test
Mode
High
Low
Idle
High
Low
Idle
High
Low
Idle
High
Low
Idle
'
No. of
Defect
Cars
24
24
i9
24
18
24
24
24
Pollu-
tant
HC
CO
N0x
HC
CO
NO
X
HC
CO
NO
X
HC
CO
NO
X
HC
CO
NO
X
HC
CO
HC
CO
HC
CO
HC
CO
NO
HC
CO
NO
HC
CO
NO
X
HC
CO
HC
CO
HC
CO
HC
CO
NO
X
HC
CO
Original
CEV Fleet
%EQ
11.0
14. 1
9.60
7.24
16. 1
9.77
30.4
36.0
6.87
35.3
33.0
.13.8
6.79
28.6
13.4
21.8
33. 3
22. 3
36.5
10. 38
29.2
10.8
43. 5
2.93
16.9
40.9
10.9
16.6
33.4
10.6
19.5
36. 1
22.0
36.5
18.0
29.2
38.7
34.9
12.9
37.0
37.7
% FF
55.9
22. 1
5.36
59.6
20. 1
5.19
36.8
7.75
8.69
37 0
10.8
1.76
60.5
15.2
2.20
45.4
10. 1
44.9
6.79
56.8
14. 1
58. 1
6. 10
8.75
52.0
8.68
6.73
52.3
16.1
1.05
47.7
7.21
45.2
6.81
49. 1
14. 1
28.5
8.46
1.83
30.2
5.62
Defect
Fleet
%E0
5.40
6.28
36.6
5.31
7.85
18 4
6.47
22.2
8.55
6.36
17.2
45.0
6.01
10.8
45.4
8.02
23.9
8. 16
32.0
8.03
11.7
9.85
16. 14
5.65
10.6
20. 1
50.5
10.5
17.0
54. 1
8.47
23.6
8. 16
30.6
6.81
12.9
13.7
21.0
47.7
15.0
39.9
% FF
69.0
65.2
16.9
69.1
63.4
35. 19
67.6
48.3
52.2
67.7
53.2
15.8
68. 1
59.7
15.4
66.4
47.4
66.3
39.3
66.4
59.6
71.6
52.4
53.6
70.8
48.5
8.78
70.9
51.6
5. 17
66.0
47.7
65.8
40.7
67.6
58.4
60.7
50.4
5.93
59. 5
31. 37
%*p
Ec
4.21
6. 11
1.22
6.02
9.3
10. 5
2.84
11.4
9.31
2.42
13.8
11. 3
5.56
6.26
8. 34
3.63
12.03
5. 37
16.5
8.68
7.29
4.74
7.48
6.05
4. 54
10. 1
2.30
6.55
10.6
0.88
3.75
11.6
5. 16
13.7
6. 13
4.48
0. 97
10.3
2.26
1. 55
8.74
1.11,
= ^r,
> constant for original CEV fleet
Ib)
HC = 0.41 cm
CO 5 4 cm/fi
NO 3 I Km/r,
1-44
-------
defect fleet was generally higher than the 5 percent level used in the CEV
fleet, varying from 0.88 percent to 11.3 percent.
1. 3.2.4
Multiple-Constituent Tests
The results of a three-constituent test for the Key Mode
(laboratory) and a nine-constituent test for the Key Mode (laboratory) are
shown in Table 1-20. These results are typical for all the multi-constituent
test analyses made. As can be seen, the multiple-constituent ST had notice-
able improvements in FF discrimination over values obtained for the original
CEV fleet, with essentially no E .
Table 1-20. Key Mode Composite TesVa' (Laboratory Data)
Test Type
Three -Constituent
High Speed
Low Speed
Idle
Nine - Cons tituent
Original
CEV Fleet
% FF
27. 5
22. 5
60.0
62. 5
%EC
5.00
5.00
5.00
12. 50
% E0
37. 5
42.5
5.00
2. 50
Defect
Fleet
% FF
89. 5
73. 7
89. 5
94.7
% Ec
0
0
0
0
% E0
10. 5
26. 3
10. 5
5.26
(a)
% Ec < 5; FTP Level I (HC = 0.41 gm/mi, CO = 3.4 gm/mi, NOx =
3. 1 gm/mi)
1. 3.2 5
General Comments
A review of the above typical results illustrates that the short
tests perform well at isolating a population of defective cars. This is noted
1-45
-------
by the general tendency for percent FF to increase and percent E to decrease
in the defect population. Although percent E decreased for HC, this was not
generally true for CO and NO .
The sources of the errors of commission and omission are
twofold. The first and usual source is that of the test procedures; i.e. ,
measurement errors. The second source is due to mixing of defects. An
observation was classified as a defective car if any component of the vehicle
was defective. Hence, all the NO data analyzed are not representative of
NO defects, for example. The multiple-constituent tests (which tend to
eliminate mixing errors) show a very high probability, greater than 70 per-
cent, of detecting defect vehicles (note that all the defective cars failed the
FTP at Level I).
In conclusion, the ST/FTP tracking of defective vehicles is
very good.
1-46
-------
1.4 GENERAL OVERVIEW REMARKS
1.4.1 Mode vs Bag ST
1.4.1.1 Individual Pollutants
In all the analyses conducted, the bag tests were shown to be
technically superior in analyzing HC and CO. Mode-type STs are preferable
to bag STs when considering their relative performances on NO . However,
all STs are deficient in analyzing NO . As the dominant variables in both
ji.
fleets are HC and CO, the bag tests are preferable under these conditions.
The complexity of implementation of bag-type STs could be a
major deterrent to their universal acceptance. The mode STs are more
desirable in this respect, especially if garage-type instruments are deemed
suitable. A clear choice is not possible without a full analysis in which the
objectives and constraints of an implementation procedure are specified.
1.4.1.2 Multiple-Constituent Tests
The clear superiority possessed by the bag-type ST is not
present when comparing tests on a multiple-constituent basis. In both the
CEV fleet and the 1974 model year fleet, the Federal Short Cycle is
approximately equivalent to the high-speed mode of the Federal Three-Mode
with laboratory instruments.
1.4.2 Single Mode vs Weighted Mode Tests
Analysis of weighted mode tests shows only very minor im-
provements in correlation over a single-mode ST. As a weighted-mode ST
would be of increased complexity; this option demands little attention.
1.4.3 Garage Instrument vs Laboratory Analyzer
The garage instruments offer additional tradeoffs within the
volumetric test area. Garage instruments reduce the technical sophistication
of the ST while, at the same time, reducing the complexity of implementation.
Technically, the garage instrument tests are inferior to the laboratory
1-47
-------
instrument tests in that the garage instruments have higher percent testing
errors for a given modal test. However; they provide additional options
under a full-scale tradeoff study.
1.4.4 Correlation Coefficient vs Contingency Table Analysis
The usefulness of the correlation coefficient is confined to
measuring direct relatability. It is useful in identifying critical areas for
further research and in providing a relative overview, such as ranking of
the ST.
For analyzing the tradeoff between impact on air quality and
cost to the public, a contingency table analysis which admits a policy decision
is most favorable. The public costs are defined as those incurred by the
manufacturer and/or those incurred by the environment. Constraints are
easily incorporated and, thus, an appropriate policy or set of policies can
be identified. The method of bounded errors of commission is recommended
as the procedure for contingency table analysis. The policy decision is the
bound on percent of allowable errors of commission. The effect of the
policy is measured in percent FF and percent errors of omission. Other
measures such as relative impact, discussed below, are also available. In
short, it allows the policy-maker to control quantifiable economic costs and
to assess the impact on air quality.
1-48
-------
1.4.5 Relative Impact on Air Quality
1.4.5.1 By Individual Pollutant
The FTP standards, or cut-points, can be interpreted as
establishing the desired impact on air quality in that the FTP cut-points fix
the percent of the population classified as high-polluting vehicles. If the
FTP were used as the test procedure in an inspection/maintenance program
which tested all vehicles (i. e. , as the ST), the relative impact on air quality
would ideally be 100 percent; that is, all the vehicles that are failures are
in fact identified as such.
Similarly, the effectiveness of the various STs can also be
used as a measure of impact on air quality, where "ST effectiveness" is
defined as:
or_ ,r .. % FF for the short test ,, ..
ST effectiveness = fl FTP failures in same population (U1)
% FF
% FF + % E
o
Thus, on this basis, the ST is always less effective than the FTP, in
proportion to the percent of errors of omission (E ) associated with a given
ST. Table 1-21 shows the ST effectiveness values for the 1974 model year
fleet for an E rate of 5 percent. These values indicate the relative impact
on air quality of the ST as compared with the impact of the FTP on air quality,
for the E conditions shown.
c
Actual benefit or impact is dependent upon the user's needs
and constraints. One measure of benefit would be the tons of pollutant re-
moved from the atmosphere on an annual basis in a given region by the use
of an ST in an inspection/maintenance program. This can be approximated
by the relationship:
Tons removed = ST effectiveness X A pollutant to be removed
in population X % population sampled (1-2)
1-49
-------
Table 1-21. Short Test Effectiveness; E = 5%
1974 Model Year Fleet
Short Test
Federal Short Cycle
NY/NJ Composite
Key Mode
Laboratory
Garage
Federal Three -Mode
Laboratory
Garage
2500 rpm Unloaded
Laboratory
Garage
ST Effectiveness^
HC
0.83
0.78
0.58
0.34
0.61
0.41
0.61
0.39
CO
0.90
0.88
0.76
0. 51
0.72
0.48
0.73
0.47
NOX
0. 17
0.06
0.28
0.22
0.22
%FF
HC
34
32
24 (I)(b)
14 (L)
25(1)
17 (I)
25
16
CO
65
64
55 (L)
37 (H)
52 (I)
35 (I)
53
34
NOX
3
1
5 (I)
4(H)
4
(a)
ST Effectiveness =
where
FF
FTP Fails
(b)
FTPHC Fails = 41.09%
FTP CO Fails = 72. 35%
FTP NOX Fails = 17. 8%
I = idle mode
L = low speed mode
H = high speed mode
1-50
-------
where
effectiveness =
and
A pollutant to be removed in population = average value for the
population of HC, CO,
or NOX, in tons/year,
in excess of that per-
mitted by the FTP
standard; it is based
on the FTP failures
and corresponding
emission values ob-
served in the popula-
tion, and vehicle-miles-
traveled characteristics
This relationship ignores those additional benefits likely to occur if the failed
vehicles were repaired and achieved emission levels below the FTP standards
after repair.
Equation (1-2) indicates areas of tradeoff that should be ex-
amined prior to the implementation of a specific inspection/ maintenance
program. Figure 1-9 depicts one aspect of such tradeoffs. This figure is
an illustrative plot of Eq. (1-2) for two different ST (Federal Short Cycle,
and Unloaded 2500 rpm -with garage instruments) as used for CO emissions.
As indicated in Table 1-21, their effectiveness values are 0.90 and 0.47,
respectively; i. e. , as compared with the CO discrimination capability of the
FTP procedure, they are 90 and 47 percent as effective as the FTP in iden-
tifying vehicles which fail the FTP test on CO. Thus, to achieve the same
benefit in total CO pollutant removal, the percentage of the population that
must be sampled by the Unloaded 2500 rpm ST is approximately double that
which must be sampled with the Federal Short Cycle ST. Alternatively stated,
for any given percent sampling of the population, the use of the Federal Short
Cycle ST would result in approximately double the amount of CO removed.
The complexity of program implementation can be measured
in annual cost. The cost components would include such items as annual
1-51
-------
IX)
i
o
o
FEDERAL
SHORT
CYCLE
2500 rpm
UNLOADED
PERCENT VEHICLE POPULATION SAMPLED
Figure 1-9- Impact of Percent Population Sampled on CO
Removed (Illustrative Example Only)
1-52
-------
operating expenses, maintenance expenses, and amortized initial development
and installation expenses. The ST requiring laboratory instrumentation would
have substantial initial procurement costs, and higher annual maintenance and
operating expenses than those using garage instruments. The bag-type ST
requires more skilled personnel and a CVS station. The bag ST and multi-
mode tests also require a dynamometer. Thus, the ST can be ranked
according to cost as follows:
Federal Short Cycle, NJ/NY Composite
Three-Mode volumetric with laboratory instruments
Three-Mode volumetric with garage instruments
2500 rpm Unloaded with laboratory instruments
2500 rpm Unloaded with garage instruments
For those inspection/maintenance programs targeted to 100 percent inspection
*
of all vehicles, the above ranking of ST by cost would appear valid. However,
if less than 100 percent inspection is envisioned for some reason, then addi-
tional factors should be considered. For example, the unit cost of a program
(per vehicle) would be expected to decrease as the percent of the population
sampled increases. Thus, in the example of Figure 1-9, if the program were
targeted to a defined level of CO removal, a cost-benefit analysis might be an
appropriate method to select the ST and the percentage sampled for minimum
cost purposes. The type of constraint normally imposed on a tradeoff study
would typically be total annual cost; however, additional constraints on per-
cent E or percent rejected (E plus FF) are also admissible under this
approach. Other areas of consideration are effective sampling and site
selection, importance of the pollution source as a function of geographic
location, social impact, etc.
1.4.5.2 Multiple Constituent Tests
Short test effectiveness is also a useful measure of test quality
for the multiple-constituent test, although the pollutant removal implications
of Eq. (1-2) must apply on an individual pollutant basis. Shown in Table 1-22
1-53
-------
Table 1-22. Short Test Effectiveness Values for Multiple
Constituent Tests; 1974 Model Year Fleet *a>
Short Test
Federal Short Cycle
Federal Three-Mode
(Laboratory Instruments)
Idle
High
Federal Three-Mode
(Garage Instruments)
Idle
High
ST Effectiveness
0.77
0.373
0.314
0.483
0.568
0.330
0.374
Percent E
Predicted^b)
5
0.05
0.01
5
5
5
5
Actual
8.84
2.04
0.68
0.00
2.72
0.00
0.69
(a)FTp failures =
Using bounded errors of commission method of analysis
are the effectiveness values for the Federal Short Cycle and the Federal
Three-Mode. Comparison of the test-to-test effectiveness values should,
of course, be made at points where the actual percent E is equal; however,
this can be only approximated with the existing data.
The technical favorability of the Federal Short Cycle is
diminished when comparing on the basis of equivalent percent E . Although
the Federal Short Cycle effectiveness is 0.77 at actual percent E equal to
8. 84, it is reduced to 0. 373 and 0. 314 for actual percent E values of 2. 04
and 0. 68, respectively. However, as shown in Table 1-22, the effectiveness
values of the high-speed mode of the Federal Three-Mode ST with laboratory
1-54
-------
and garage instruments are 0. 568 (actual percent E = 2. 72) and 0. 374 (actual
percent E = 0. 69), respectively. Comparable effectiveness values for the
idle mode with laboratory and garage instruments are 0. 483 and 0. 330, re-
spectively, both with actual percent E equal to 0. Thus, in the actual per-
cent E range below approximately 3, the Federal Three-Mode ST with
garage instruments (idle or high-speed mode) is essentially equivalent to the
Federal Short Cycle in effectiveness while the Federal Three-Mode ST with
laboratory instruments has a higher effectiveness than the Federal Short
Cycle.
Although the favorability of the laboratory instruments over
the garage instruments persists under this method of comparison, considera-
tion of program complexity could bias test desirability in favor of the Federal
Three-Mode with garage instruments.
1-55
-------
2. INTRODUCTION
-------
2. INTRODUCTION
2. 1 BACKGROUND AND OBJECTIVES
With regard to compliance by vehicles and engines in actual
use with the certification emission standards established for a vehicle at
the time of its manufacture, the Clean Air Act of 1970 stipulates in
Sec. 207 (b):
"If the Administrator determines that
(i) there are available testing methods and procedures to
ascertain whether, when in actual use throughout its
useful life .... , each vehicle and engine to which
regulations .... apply complies with the emission
standards of such regulations,
(ii) such methods and procedures are in accordance with
good engineering practices, and
(iii) such methods and procedures are reasonably capable
of being correlated with tests conducted under
section 206 (a) (1), then --
"(1) he shall establish such methods and pro-
cedures by regulation, and
"(2) at such time as he determines that inspec-
tion facilities or equipment are available
for purposes of carrying out testing
methods and procedures established under
paragraph (1), he shall prescribe regula-
tions which shall require manufacturers
to warrant the emission control device or
system of each new motor vehicle or new
motor vehicle engine .... for its useful
life."
Thus, there are the essential requirements of "availability," "conformance
with good engineering practices, " and "reasonable correlation with certifi-
cation test procedures" which must be met prior to the promulgation of
regulations which impose the in-use warranty provisions of Sec. 207 (b)
upon the motor vehicle manufacturers.
2-1
-------
The states of New York and New Jersey have developed
short emission tests for potential use in inspection/maintenance (I/M)
programs in their areas. The Clayton Manufacturing Company also devel-
oped a short test procedure for use in I/M programs. More recently, the
EPA has developed short tests similar to those of New York, New Jersey,
and Clayton. Thus, there are a number of tests "available" to determine
the exhaust emissions of in-use vehicles; these test methods and procedures
"conform with good engineering practices11 in that they utilize well-
recognized emission-measurement equipment and techniques.
These tests are "short" in duration (approximately 3 to 5
minutes) in order to (a) minimize the inconvenience of the motoring public
(and thereby maximize cooperation), and (b) minimize capital costs of
inspection stations by maximizing the number of vehicles a given facility
could test. They have been structured for "simplicity" in order to (a)
reduce the potential for procedural errors, and (b) to reduce test costs.
As a result, all such tests require that the vehicle be tested in a "hot"
condition; i.e., at its normal operating temperature.
There remains the requirement to demonstrate "reasonable
correlation with certification test procedures, " i. e. , with the Federal Test
Procedure (FTP) used in the certification of new motor vehicles. Therefore,
the present study was performed with the principal objective of analyzing
emission data from both short tests (STs) and FTP tests of the same vehicles
in order to determine the degree of "correlation" which exists between vehicle
exhaust emissions as determined by an ST and the FTP. A second objective
was to analyze continuous trace data from these tests to form the basis for
the development of a new and "better correlating" short test procedure,
should the need occur.
2.2 STUDY SCOPE
The basis for the analyses was ST and FTP data from three
vehicle fleets:
2-2
-------
a. Catalyst-Equipped Experimental Vehicle (CEV) Fleet
This fleet comprised 40 catalyst equipped "1975-
prototype" models that had been operated in California
in Ford vehicle test programs. These vehicles were
tested by Olson Laboratories in Anaheim, California.
b. In-Use 1974 Model Year Vehicle Fleet
This fleet comprised 147 in-use 1974 model year cars
in three groups of approximately 50 cars each, repre-
senting different inertia weight classes (subcompact,
intermediate, and full size) and three different auto
manufacturers. These vehicles were procured by
Olson Laboratories, Livonia, Michigan, from the
greater Detroit area and tested by EPA in the Ann
Arbor test facility.
c. Defect Test Fleet
This fleet comprised five of the catalyst-equipped Ford
vehicles from the CEV fleet noted above. Approximately
95 "defect" tests were conducted on these vehicles. The
defect tests included such items as spark plug misfiring,
carburetor misadjustment, defective valves, and
degraded catalysts. These tests were performed by
Olson Laboratories, Anaheim, California.
Each of the above vehicles was tested by the FTP and the
following STs:
Federal Short Cycle
NY/NJ Composite
Clayton Key Mode
Federal Three-Mode
Unloaded 2500 rpm
For the volume trie-type tests (Clayton Key Mode, Federal Three-Mode,
and Unloaded 2500 rpm), both laboratory and garage-type instruments were
used to record HC and CO measurements. Garage-type instruments were
included in the event that higher-accuracy laboratory analyzers would not
be compatible with the working environment of a typical automotive garage
or a large-scale vehicle testing station. All the NOx readings were made
2-3
-------
with laboratory analyzers due to the unavailability of an appropriate
garage-type NO instrument.
2.3 METHOD OF APPROACH
The primary thrust of the work performed under this contract
was statistical in nature. Two complementing methods were employed to
assess Sec. 207 (b) correlation --a conventional correlation analysis and a
contingency table analysis. The conventional correlation analysis addresses
the question of direct relatability between the ST and the FTP by examining
the relationships present in the data. The results are of great usefulness
in indicating the extent to which each ST tends to track the FTP. The
contingency table analysis addresses the relatability of ST and FTP on a
pass-or-fail level. Each data point is examined, and a determination is
made as to whether the auto passed or failed the FTP and passed or failed
the ST. Thus, errors of commission (E ), errors of omission (E ),
correct passes by each test (PP), and correct fails by each test (FF) are
identified. Hence the technique allows for the study of the tradeoffs between
errors and correct identifications.
The conventional correlation analysis, being purely an
analysis of the data, does not permit policy decision as a variable or
parameter. Contingency table analysis, on the other hand, permits the
integration of policy decision in that it provides for the determination of
the ST pass/fail cut-points. Thus, policy decision entered the analysis as
a quantifiable variable, and a study indicating the impact of various policies
was performed in the contingency table analysis. One important method
reflecting impact to policy is that of the method of bounded errors of com-
mission. In this scheme, limits are set on the maximum permissible per-
centage of errors of commission, and the ST cut-points are selected to
yield minimum errors of omission within this constraint. This analysis
permits a direct answer to the question, "For a given permissible level of
errors of commission, what level of errors of omission is associated with
2-4
-------
a given test, and with what impact on air quality (inferred from the
percentage of FF and EQ vehicles)?11
These two methods of analysis, each representing different
interpretations of Sec. 207 (b) correlation, were applied to both the CEV
fleet and the 1974 model year in-use fleet. They were also applied to the
defect test fleet to (a) determine the statistical character of the specific
defect tests, and (b) to examine the ability of the STs to detect defective
vehicles of this nature.
2.4 ORGANIZATION OF REPORT
The results of the study are reported in the following order
and context:
Section 3 - Test Characteristics and Procedures
Defines the five short tests used, describes
the test conditions and procedures, and dis-
cusses the composition of the three test
fleets.
Section 4 - Catalyst-Equipped Experimental Vehicle Fleet
Defines and discusses, for the CEV fleet, the
statistical analysis techniques and results for
the correlation and contingency table analyses
conducted.
Section 5 - In-Use 1974 Model Year Vehicle Fleet
Defines and discusses, for the 1974 model year
fleet, the statistical analysis techniques and
results for the correlation and contingency table
analyses conducted.
Section 6 - Defect Data from Catalyst-Equipped Experimental
Vehicle Fleet
Defines and discusses the analysis techniques and
results from the analyses made to determine the
statistical character of the defect tests and to
examine the ability of the various short tests to
detect defective vehicles.
2-5
-------
3. TEST CHARACTERISTICS AND PROCEDURES
-------
3. TEST CHARACTERISTICS AND PROCEDURES
In this program, five short tests (STs) and the 1975 Federal
Test Procedure (FTP) were performed on three test fleets. This section
defines the various STs, describes the test conditions, and discusses the
composition of the test fleets.
3.1 SHORT TESTS
3. 1. 1 General
Two classes of STs were involved, and these may be catego-
rized as (a) modal or volumetric and (b) as driving trace or CVS. Both sets
of nomenclature are used in this report, depending upon the aspect of the
test structure that is pertinent to the discussion. In the modal tests, the
test technician operates the vehicle on a dynamometer at a fixed vehicle
speed and dynamometer load, or unloaded at a fixed engine rpm, or at idle.
The vehicle tailpipe exhaust is sampled directly, and the concentration of
each pollutant is measured and recorded in percent, of in parts per million,
of the undiluted exhaust. Three modal STs were used:
Clayton Key Mode
Federal Three-Mode
Unloaded 2500 rpm
The Clayton Key Mode and Federal Three-Mode STs each had high speed,
low speed, and idle modes.
For the second class of ST, the test technician drives the
car on the dynamometer in accordance with a prescribed driving pattern on
a driving trace. The vehicle exhaust is diluted by the CVS procedure, and
a single sample bag of diluted exhaust is collected for the whole ST. The
dilute sample is analyzed and the results usually expressed in grams per
mile. This procedure requires the same equipment, sampling procedure,
3-1
-------
and analytical equipment as the Federal Test Procedure (FTP) used in the
certification of new vehicles. The difference is that the driving trace for
the ST is much shorter and simpler. Two CVS-type STs were used:
Federal Short Cycle
Composite of New Jersey Acid and New York Short
Test (NY/NJ)
Both classes of ST involved approximately 2 or 3 minutes
of driving time on the dynamometer. All STs were performed with the
engine at its normal operating temperature; i.e., "hot" tests.
The HC and CO content of the exhaust gas in the volumetric
tests was measured from samples taken at the same time by two different
classes of instruments. One set, called "laboratory analyzers," was
identical (except for range) with the high-accuracy analyzers used in CVS
certification testing. The second set, called "garage instruments," used
a lower-cost, lower-accuracy and precision instrument of the type currently
in use by many automotive service stations for routine diagnostic work.
The structure of each test is given below.
3.1.2 ST Definition
3.1.2.1
Clayton Key Mode
The Clayton Key Mode is a well-known test which has been
in use for several years for diagnostic emissions testing.
Vehicle
Weight
Class,
Ib
2000 to
2750
2800 to
3750
3800
and up
Transmission
Range /Gear
In lower
gear (3rd)
Drive or
high gear
Drive or
high gear
Dynamomete r
Load, hp
@ mph
15 @ 38
24@46
30 @ 50
Modes
High
Speed
Cruise,
mph
36 to 38
44 to 46
48 to 50
Low
Speed
Cruise,
mph
22 to 25
29 to 32
32 to 35
Idle
Automatic
transmission
drive
3-2
-------
3.1.2.2
Federal Three-Mode
The Federal Three-Mode differs from the Clayton Key Mode
in that it uses dynamometer loadings simulating the average power that
occurs at the appropriate speeds in the FTP where the vehicle is accelerat-
ing (decelerations are not included). This results in a higher dynamometer
loading for the Three-Mode as compared with the Key Mode at the low speed
condition, and, for vehicles with an inertia weight greater than 4500 Ib, at
the high speed setting also.
Vehicle
Weight
Class, Ib
Up to
2500
2501 to
3500
3501 to
4500
Above
4500
Transmi s si on
In lower gear
for 30-mph
test (3rd gear)
Drive or
high gear
Drive or
high gear
Drive or
high gear
High Speed
Mode
Speed,
mph
50
50
50
50
Load,
hp
21
26
31
36
Low Speed
Mode
Speed,
mph
30
30
30
30
Load,
hp
9
12
15
18
Idle Mode
Automatic
transmission
in neutral
3.1.2.3
neutral.
3.1.2.4
Unloaded 2500 rpm
This is a high-speed test: 2500 rpm, transmission in
Federal Short Cycle
The Federal Short Cycle was derived from the FTP. Accel-
erations and decelerations are representative of those encountered in the
FTP, and average speed is nearly the same as the three-bag FTP driving
cycle (21. 70 mph and 21. 27 mph, respectively).
3-3
-------
Time in Mode, sec
This is a nine-mode, 125-sec CVS test that follows the
driving schedule shown below and plotted in the top half of Figure 3-1.
Mode
0-16 mph acceleration
16 - 29 mph acceleration
29 mph cruise
29-37 mph acceleration
37 - 42 mph acceleration
42 - 37 mph deceleration
37 - 20 mph deceleration
20-0 mph deceleration
Idle
The test does not include engine startup or shutdown. The dynamometer
loadings follow the procedure as required for the FTP.
3.1.2.5 Composite NY/NJ
This is a six-mode, 75-sec CVS test that follows the driving
cycle shown below and plotted in the lower half of Figure 3-1.
Mode Time in Mode, sec
Idle 22
0-30 mph acceleration 15
30 mph cruise 15
30 - 10 mph deceleration 12
10 mph cruise 7
10-0 mph deceleration 4
75
3-4
-------
FEDERAL SHORT CYCLE
60
TIME, sec
120
COMPOSITE NY/NJ SHORT CYCLE
CO
30
20
10
20 40 60
TIME, sec
80
Figure 3-1. Federal Short Cycle and Composite NY/NJ
Short Cycle Test Driving Schedules
3-5
-------
The test does not include engine startup or shutdown. All vehicles are
tested at an inertia weight of 3000 Ib and a dynamometer loading of 3. 5 hp
at 30 mph.
3.1.3 Short Test Sequence
A short test sequence consists of the following tests and
soak periods in the order shown.
a. Completion of 1975 FTP
b. Soak - 6 minutes
c. Clayton Key Mode
d. Soak - 6 minutes
e. Federal Three-Mode
f. High-speed Unloaded 2500 rpm test
g. Soak - 6 minutes
h. Federal Short Cycle
i. Soak - 6 minutes
j. Composite NY/NJ
The 6-minute soak procedure is performed as follows: after completion of
the preceding test, the vehicle engine is stopped, the vehicle hood is closed
if it was open, and the auxiliary air cooling fan is turned on if it was not
previously in use. The fan remains in operation for 3 minutes. At the end
of 3 minutes, the auxiliary air fan is turned off and the vehicle's engine is
started. The engine is allowed to idle in neutral for 3 minutes. Upon the
completion of this 3-minute idle period, the next test in the sequence is
initiated.
During the entire ST, the vehicle hood is closed and the
auxiliary cooling fan is not in operation.
In the modal tests, the car is to be operated in each mode
until the emissions stabilize. In the CVS tests, driving trace procedures
and tolerance (and transmission shift points, if applicable) are the same
as for the FTP.
3-6
-------
3.2 TEST FLEETS
3.2.1 Catalyst-Equipped Experimental Vehicle
Fleet (CEV)
3.2.1.1 Type of Car
These 40 vehicles were all 1973 Ford Galaxies, owned by the
Ford Motor Company. They were equipped by Ford with an oxidizing
catalyst for control of HC and CO. The emission control system also
included air injection and exhaust gas recirculation (EGR). All had
8-cylinder, 400 CID engines with two-barrel carburetors and automatic
transmissions. Manufacturer's specifications for ignition timing and dwell
were 12° BTDC and 24° to 30°, respectively. Axle ratio was 3.0, and tire
size was HR 78-15. The FTP inertia weight at which the vehicles were
tested was 5000 Ib.
3.2.1.2 Prior Use
At the time of receipt of these vehicles by the testing labora-
tory (Olson Laboratories, Inc., Anaheim, California), the odometer readings
ranged from 7000 to 36,000 miles, with an average of 21,000. Prior to these
tests, the cars were primarily used by Federal and California state
employees in a number of locations throughout California. The driving
pattern was highly variable, ranging from primarily stop-and-go city traffic
to primarily high-speed highway driving. Vehicle and emission system
maintenance was performed essentially in accordance with Ford Motor
Company recommended procedure. Emission system maintenance on some
vehicles was performed by Ford Motor Company, while for others it was
performed by local Ford dealers or motor pool personnel, following pro-
cedures established and monitored by Ford Motor Company.
3.2.1.3 Test Conditions
Upon receipt of a vehicle, the as-received fuel was drained
and test fuel (indolene clear) was added. The car was operated for
3-7
-------
approximately 10 minutes, at which time a vehicle inspection was
performed. This consisted primarily of measuring engine tune, idle CO
and HC, inspecting fluid levels, and verifying the existence and operation
of emission control devices.
For the first 20 vehicles of this fleet, a short test sequence
(as defined in Sec. 3. 1.3) was performed right after this inspection, after
which the vehicle was placed in cold soak for the first FTP. Immediately
following the FTP, another short test sequence was performed, and the
vehicle placed in cold soak for the second FTP. A third short test sequence
was performed immediately upon completion of the second FTP. For the
latter 20 cars of this fleet, the first short test sequence (after the vehicle
inspection) was deleted.
All test conditions, instrumentation, and procedures for the
1975 FTP were as prescribed in the Federal Register, with one addition. The
concentrations of HC, CO, NO , and CO- in the undiluted exhaust were also
X b
continuously measured and recorded during each FTP. The sampling train
and analyzers used for this were the same as those used for the volumetric
short test cycles. These continuous trace data were used to gain insight
into the emission generation characteristics of various portions of the FTP.
The first group of 20 cars was tested during the period
8 September to 25 October 1974. A group of 10 cars was tested between
11 and 16 December 1974. A final group of 10 cars was tested between
22 January and 19 February 1975.
3.2.2 Defect Test Fleet
Upon completion of the CEV fleet vehicle tests described
above, 95 defect tests were performed on 5 of the 40 vehicles of the CEV
fleet. These simulated a wide variety of malfunctions that could occur in
a typical passenger car: defective ignition components; changes to ignition
timing, dwell, and spark advance; faulty carburetion; defective valves;
clogged air filters; and faulty emission control components. A detailed
listing of all defects is given in the Appendix.
3-8
-------
For each defect, one FTP was performed, followed by a
short test sequence. For each FTP, additional continuous trace recordings
were made of the concentration of HC, CO, NO , and CO_ in the undiluted
X Ct
vehicle exhaust, as was described for the normal vehicle tests. In 20 of the
tests, catalyst bed temperature and exhaust flow rate were measured and
recorded for the duration of the FTP and each ST.
3.2.3 In-Use 1974 Model Year Vehicle Fleet
3.2.3.1 Types of Cars
This fleet comprised in-use 1974 model year vehicles. There
were 49 Ford Pintos, 49 Chevrolets (Caprice and Impala), and 49 Dodge/
Plymouths (Coronet, Charger, Satellite). The Pintos were 140 CID, tested
at 2750-lb inertia weight, the Dodge/Plymouths were 318 CID, tested at
4000-lb inertia weight, and the Chevrolets were 400 CID, tested at 5500-lb
inertia weight.
All cars had automatic transmission. The emission control
systems were EGR plus air injection for the Chevrolets and Pintos, and
EGR for the Plymouth/Dodges.
The rear axle ratio was 2. 73 for the Chevrolets and 3.40 for
the Pintos. The Plymouth/Dodges had a ratio of 2.94.
3.2.3.2 Prior Use
These vehicles were all privately owned, and were from the
greater Detroit metropolitan area. The as-received odometer readings
ranged from 3000 to 20,000 miles, with an average of 11,000. There was
no significant difference in the odometer readings between any of the manu-
facturers' subgroups of 49 cars. No information is available concerning
the detailed driving pattern or maintenance history for any of the cars.
3.2.3.3 Test Conditions
Testing was performed by the EPA Emissions Laboratory
at Ann Arbor. Each car was tested once by the 1975 FTP, immediately after
3-9
-------
which a short test sequence (as defined in Sec. 3.1.3, with one exception)
was performed. The one exception pertains to the Key Mode test. For the
1974 model year fleet, the Key Mode tests were run at a fixed set of speeds.
These speeds were 48 to 50 mph for the high speed mode, and 32 to 34 mph
for the low speed mode, regardless of the test vehicle inertia weight.
Thus, the Pintos were the only vehicles affected, as all other vehicles fall
in the same weight class for the Key Mode test. The Key Mode tests for
both the 1974 model year fleet and the CEV fleet were thus all run at the
same sets of speeds. The dynamometer inertia and horsepower settings
were made in accordance with the test vehicle inertia weight, per the
standard Key Mode format.
Twenty-five of the cars that failed the FTP were tuned by
EPA and retested by the same procedure described above. This tuneup was
parametric in that adjustments were made as required in an effort to bring
ignition timing, dwell, etc., within manufacturer's specifications, but no
new components were installed, regardless of the condition of the existing
ones.
3-10
-------
4. CATALYST-EQUIPPED EXPERIMENTAL
VEHICLE FLEET
-------
4. CATALYST-EQUIPPED EXPERIMENTAL VEHICLE FLEET
This section summarizes the results of statistical analyses
conducted to determine the degree of correlation existing between the various
short tests (STs) and FTP tests conducted on the catalyst-equipped experi-
mental vehicle (CEV) fleet. Preliminary analyses are discussed in Sec. 4. 1;
the principal statistical analysis techniques and results are summarized in
Sec. 4. 2.
4. 1 PRELIMINARY STATISTICAL ANALYSES
Preliminary analyses were made to assess data quality and
statistical structure. Of specific concern were the following goals:
a. Determine the data acceptable for further processing.
b. Determine the variation within each test procedure.
c. Determine the vehicle-to-vehicle variation.
d. Determine the intrinsic variables and statistical structure
of each of the tests.
e. Determine the distribution properties of the test data.
Goals a and e were met by simple data screening techniques.
A multivariable analysis of variance was used to meet goals b and c, while d
was addressed by a canonical correlation analysis. These techniques/analyses
are briefly discussed below.
4. 1. 1 Data Screening
All of the basic test data for the CEV fleet were received by
The Aerospace Corporation for processing. A screening procedure was
developed to evaluate these data, and to provide an annotated data base for
subsequent statistical analyses. All inputs to the statistical data base were
derived directly from the test data traces. All apparent anomalies and/or
discrepancies in the data were examined and an effort made to reconcile
them. Discussions were held with the testing laboratory and with EPA/ECTD,
4-1
-------
as appropriate, to resolve these situations. In some cases, certain tests,
or portions thereof, were deleted from the data base. Of the 40 cars of the
CEV fleet, the final data base contained 26 cars with two valid FTP tests,
and 14 cars with one valid FTP test. A few short test (ST) results were
deleted, as were various isolated values for a given pollutant for a specific
mode.
After the test data were put on tape, various descriptive
characteristics of the data were used to detect gross errors in the observa-
tions, in coding and keypunching, and in including inappropriate cases.
Generally, this was accomplished by checking for improper symbols or
characters, such as characters were numbers should be, for outliers or
blunders, and for missing observations. Erroneous data were reconstructed
where possible; otherwise, the case was flagged as inappropriate for pro-
cessing. Table 4-1 summarizes the number of cases available for statistical
analysis.
Table 4-1. Number of Cases Available for Statistical
Analysis (CEV Fleet)
Test
Federal Test Procedure
Federal Short Cycle
NY/NJ Composite
Clayton Key Mode
Laboratory instruments
Garage instruments
Federal Three-Mode
Laboratory instruments
Garage instruments
2500 rpm Unloaded
Laboratory instruments
Garage instruments
No. of Cases
40
39
39
40
40
31
40
40
40
4-2
-------
The mathematical model employed in the contingency table
analysis requires that the data follow a bivariate normal or log-normal
distribution. This assumption was checked using a combination of histograms,
normal probability plots, and scatter plots (Ref. 4-1). Generally, the log of
the data appears normally distributed.
4.1.2 Multivariate Analysis of Variance
A multivariate analysis of variance (Ref. 4-2) with estimation
of the variance components (Ref. 4-3) was made for the CEV fleet. The pur-
pose of the analysis of variance is the comparison of means when the data are
grouped or classified in one or more ways. The CEV fleet data were grouped
according to replication. The groups were termed first good data and second
good data, reflecting the original testing sequence. No difference in the data
groups was discernible.
The purpose in estimating the variance components is the
quantification of multiple sources of variation. The sources of variation
identified in the CEV fleet were fluctuations between cars and measurement
errors within each test. The results of the variance components analysis
are shown in Table 4-2. Since there were 14 cars in the CEV fleet which had
only one valid FTP, the number of cars in the analysis of Table 4-2 is less
than the number previously indicated in Table 4-1 because replicates are re-
quired to analyze variance components.
Normalized dispersion is defined as
_ _ standard deviation of population (S)
population mean (M)
which is a dimensionless quantity. "D" provides an effective measure of the
variability of the population as observed by a test. As the fluctuations be-
tween cars are legitimate, a good indicator of test quality is the percent of
the variation due to testing (ex). This indicator is defined as
01 =i±Vx 100
s^
4-3
-------
Table 4-2. Summary of Variance Components (CEV Fleet)
Test
FTP
Federal Short Cycle
NY/NJ Composite
Key Mode
Federal Three -Mode
2500 rpm Unloaded
No. of Cars(a)
26
25
25
25L
26G
17L
Z6G
26L
Z6G
Test Mode
High speed
Low speed
Idle
High speed
Low speed
Idle
Pollutant'*'
HC
CO
N0x
HC
CO
NO
X
HC
CO
N0x
HCL
HCG
coL
COQ
N0x
HCL
HCG
coL
COG
N0x
HCL
HCG
COL
COQ
N0x
HCL
HCG
COL
coG
NO
X
HCL
HCG
coL
coG
N0x
HCL
HCG
COL
coG
N0x
HCL
HCG
coL
coc
N0x
M
0. 69
2.68
2.57
0. 59
1.06
3.70
31.7
25.4
27.0
199
38.3
0.043
0.046
831
181
38.6
0.012
0.03
1418
289
39.8
0.0075
0.024
193
195
36.6
0.045
0.044
1008
271
42.9
0.018
0.033
2406
202
38.5
0.009
0.028
90. 5
375
59.5
0.021
0.038
464
Units
gm/mi
gm/mi
gm/mi
gm/mi
gm/mi
gm/mi
ppm
ppm
ppm
ppm
ppm
%
%
ppm
ppm
ppm
%
%
ppm
ppm
ppm
%
%
ppm
ppm
ppm
%
%
ppm
ppm
ppm
%
%
ppm
ppm
ppm
%
%
ppm
ppm
ppm
%
%
ppm
D
0.88
0.5
0. 26
1. 15
0.88
0.23
1.49
1.27
0.26
0.47
0.35
0.70
0.42
0.58
0.43
0.54
0.83
0.23
0. 19
1.0 .
0.54
0.64
0.25
0.27
0.65
0.29
0. 67
0.32
0.47
1.03
0.54
0.67
0.33
0. 13
1.61
0.67
0.67
0.31
0.19
2.39
1.66
0.81
0.45
0.30
o.%
3
7
16
22
44
30
19
76
53
10
30
3
32
4
3
65
1
84
27
1
70
12
--
20
6
28
3
50
2
67
77
3
69
14
1
57
4
--
52
4
1
1
50
48
(a)
Subscripts L and G denote laboratory and garage analyzers
4-4
-------
where
ST = standard deviation of testing errors
S = standard deviation of population.
Missing values for a in Table 4-2 indicate the computationally
degenerate case where ST is computed to be larger than S . The ot values
shown in Table 4-2 indicate that the garage analyzers are of lower quality
(higher a) than the corresponding laboratory instruments. The ST bag tests
have higher a than many of the volumetric test procedures using laboratory
instruments. The high ot in the bag tests may be due to variations within the
driving procedure rather than to instrumentation, while the low ot associated
with volumetric tests with laboratory analyzers may be due to simplicity of
the procedure plus instrument accuracy.
4.1.3 Canonical Correlation Analysis
Canonical correlation analysis (Refs. 4-2, 4-4) examines the
relationship between two sets of variables. The problem is to find a linear
combination of a set, X, of variables that has maximum correlation with a
linear combination of another set, Y, of variables. The resulting correlation
is called the canonical correlation coefficient, and the linear combinations
are termed the canonical variables. A second pair of linear combinations
is then looked for that has a maximum correlation and is uncorrelated with
the first pair of linear combinations. The number of pairs of linear combina-
tions of the X and Y sets is equal to the number of variables in the smaller
set (X or Y, whichever is smaller). The technique is useful in testing for
independence of two sets of variables and in predicting information about a
hard-to-measure set of variables from a set that is easier to measure.
The canonical correlation coefficients for each ST versus the
FTP are shown in Table 4-3 together with original correlation coefficients.
The observations used were the first good data set. For the EPA Short Cycle
and the NY/NJ Composite, the canonical correlation coefficients do not differ
significantly from the correlation coefficients of the original data. Slight
improvements can be seen in the three-mode volumetric tests. However, the
4-5
-------
Table 4-3. Canonical Correlation Coefficients Between the FTP and ST
for the CEV Fleet (first good data set)
Test
Federal Short Cycle
NY/NJ Composite
Key Mode
( Laboratory)
Key Mode
(Garage)
Federal Three -Mode
(Laboratory)
Federal Three -Mode
(Garage)
2500 rpm Unloaded
(Laboratory)
2500 rpm Unloaded
(Garage)
No. of Cars
39
39
40
40
31
40
40
40
Test Mode
High speed
Low speed
Idle
High speed
Low speed
Idle
High speed
Low speed
Idle
High speed
Low speed
Idle
Pollutant
HC
CO
NO
X
HC
CO
NO
HC
CO
N0x
HC
CO
NO
X
HC
CO
NO
X
HC
CO
HC
CO
HC
CO
HC
CO
N0x
HC
CO
NO
X
HC
CO
NO
X
HC
CO
HC
CO
HC
CO
HC
CO
NO
X
HC
CO
N0x
Conventional
Correlation
Coefficient
0.87
0.81
0.62
0.92
0.77
0.61
0.74
0.23
0.79
0.70
0.38
0.1 6
0.94
0.04
0.24(»)
0.73
0.37
0.73
0.21
0.88
0.52
0.77
0.16
0.83
0.74
0. 25
0.02
0.78
0.52
0.08
0.73
0.21
0.78
0.52
0.47
0.30
0.23(a>
0.50
0. I4
0. 20
Canonical
Variable
1
2
3
1
2
3
1
2
3
.
1
2
3
-
1
2
3
1
2
3
1
2
3
1
2
3
Canonical
Correlation
Coefficient
0.89
0.80
0. 61
0.92
0.82
0.61
0.96
0.86
0.65
--
.-
..
--
..
0.94
0.85
0.65
--
--
--
0.98
0.93
0.55
--
...
--
0.90
0.84
0.58
..
--
--
0.63
0.49
0.21
0.58
0.38
0. 26<»»
'a)Not significantly different from 0 at the 95% confidence level.
4-6
-------
physical interpretation of the canonical variables is illusive for these tests.
The canonical correlation coefficients for the unloaded test indicate the
inferior correlation properties of this type procedure. With the exception of
the unloaded test, the canonical correlation coefficients are significantly
different from zero; i.e., the tests are correlated to some degree.
4.1.4 Summary of Preliminary Analysis Results
The HC and CO observations are generally more variable than
the NO readings, as indicated by the dispersion results for the FTP (see
X.
Table 4-2). The test-to-test variation (or) can be quite high and, hence,
repeatability of the test procedures can be poor.
Canonical variables may offer some advantages in further
analysis. However, their interpretation is difficult, and in a first analysis
the original variables seem appropriate.
A model of the distribution properties of the test data appears
most likely to be log-normal. This type of model appears appropriate for
predicting a contingency table for the total vehicle population.
4-7
-------
4.2 PRINCIPAL STATISTICAL ANALYSIS TECHNIQUES
AND RESULTS
The statistical procedures utilized consist of two comple-
menting classes: correlation analysis and contingency table analysis.
Correlation analysis addresses the direct relatability of the ST with the
FTP. Correlation analysis is an important aid in identifying STs that
have acute deficiencies. Contingency table analysis approaches the question
of relatability from the viewpoint of the possible tradeoffs between impact
on air quality and cost to the public (both direct and indirect). It is an
important tool to aid in policy formulation and cost-benefit analysis. The
following sections briefly define each such analysis technique and summa-
rize its associated results.
4.2.1 Correlation Analysis
4.2.1.1 Conventional Method
A conventional correlation analysis includes the calculation
of the sample correlation coefficient r, and an a-percent confidence interval
for the population correlation coefficient p, on paired observations. Letting
(x., y.) i = 1, , N denote the observations, r is defined by
N
S (*j - Mx)(y. - My)
S S
x y
where M , S , and M S are the mean and standard deviation of the obser-
xx y y
vations x. and y^, respectively. An a-percent confidence interval is given by
(r-, r+), where the probability that the interval covers p is a/100. For the
95 percent interval used in this study
4-8
-------
where z = 1/2 ln(-j-^-^J, (Fisher's Z statistic, Ref. 4-4).
The sample correlation coefficient is used as the prime quanti-
tative measure of relatability. The closer r is to 1, the better the relation.
A lack of relationship is indicated by r = 0. Negative r indicates an inverse
relation between the observations, i.e., if one observation is high, the
other is low, and vice versa. The confidence interval is viewed as reflecting
the sensitivity of the calculations to the data. The wider the interval, the
less predictable is the correlation coefficient and, hence, the relatability.
A scattergram is also an important device for assessing
direct relatability. A scattergram is merely a two-dimension plot of the
data pairs (x., y.). This provides for visual examination of the data, which
is crucial in any relatability study. A sample scattergram and the associ-
ated statistics are shown in Figure 4-1. Here HC on the Federal Short Cycle
is plotted versus the HC on the FTP for the CEV fleet. The number of cases
(N) is 39. The sample correlation coefficient (COR) is 0.872, while the
95 percent interval is (0.768, 0.931). The regression line of "y" on "x"
(ST on the FTP) is produced by drawing a straight line between the two points
marked Y on the right and left borders of the plot. This line represents
a least squares fit of the data (as measured in the y direction). Similar
scatter grams for each ST and each emission constituent (HC, CO, NO )
X.
were examined in the course of the study.
Since the data included replications on some cars, the data
were organized into the following structure for conventional correlation
analysis:
a. First Good Data
This data set contains the observations of the first
FTP and ST, both of which are valid.
b. Second Good Data
This data set contains the second pair of FTP and
ST observations, both of which are valid.
4-9
-------
. . T
+
+ «^« JE
I
H*
o
u
C
s
c
(gm/mi)
3.0
2.5 *
2.0 *
1.5 *
1.0 »
.50
C.O
X* +* ^« * ^# ** «^ ^*
.20 .60 1.0 1.4 1.8 2.2 2.6
.40 .80 1.2 1.6 2.0 2.4 2.8
M* 39
COB= .872
95* IHIEBVAL: ( 0. 76B, 0.931)
3.0
3.2
UC FTP (gm/mi)
HEAD ST.DBV. BEGBBSSIOM LIIE B2S.HS.
X .64338 .53578 X* .99967*1* .13485 .07045
T .50870 .46756 1= .76129*1* .01891 .05365
Figure 4-1. Correlation Analysis Scatter gram; CEV Fleet;
Federal Short Cycle HC vs FTP HC
-------
c. Average Data
This data set contains the average of the FTP and ST
observations on each car (for the Federal Short Cycle
and NY/NJ Composite only).
For each ST a correlation analysis was performed on first
good data, second good data, and average data where appropriate. The
following sections briefly summarize the significant results.
4.2.1.1.1 FTP Composite Emissions vs Individual FTP Bags
To gather insight on correlatable STs, the correlations
between the FTP composite emissions and the individual FTP bag data were
examined. The composite values were computed in the standard manner.
The bag data were computed in grams of each pollutant per bag. Addition-
ally, the sum of the bag 2 and bag 3 constituents were computed and the
correlation coefficient with the composite data calculated. The analysis
was conducted on both first and second good data.
Table 4-4 shows the FTP intra-correlations for like con-
stituents. Additionally, cross correlation coefficients between dissimilar
pollutants were computed (not shown). NO appeared to be uncorrelated
with HC and CO. The cross correlation of HC and CO was typically 0.4 to
0.6. The results shown in Table 4-4 indicate that both cold (bag 1) and hot
(bag 2, bag 3, bag 2 + 3) test procedures have a high correlation with the
FTP composite. Thus, it may be possible to develop prototype STs using
sections of the FTP-
4.2.1.1.2 ST vs FTP Composite Emissions
A summary of ST/FTP correlation coefficients is given in
Table 4-5. For N = 40 or 39, a computed correlation coefficient greater
than 0.35 indicates that the ST and FTP pollutants are statistically corre-
lated at the 95 percent confidence level. For N = 25 or 26, this threshold
is approximately 0.4.
4-11
-------
Table 4-4. FTP Composite vs Bag Correlation Summary
(CEV Fleet)
FTP
Bag No.
1
2
3
2+3
Good
Data
Set'*'
First
Second
First
Second
First
Second
First
Second
Composite vs Bag ...
Correlation Coefficient' '
HC
0.90
0.91
0.94
0.99
0.84
0.97
0.98
0.99
CO
0.96
0.93
0.90
0.81
0.86
0.87
0.95
0.90
N0x
0.95
0.91
0.87
0.79
0.95
0.97
0.99
0.98
(a)
(b)
First good data contained 40 cars
Second good data contained 26 cars
The correlations are statistically significant at the
95% confidence level.
4.2. 1.2
Multiple Regression Analysis
A regression analysis evaluates the relationship between a
dependent variable and one or more independent variables. This technique
was used to predict the FTP results from three-mode volumetric obser-
vations. For example, the constants h_, a,, a_, and a_ in
HC
FTp
are determined so that the correlation between the predicted HC-,-,-, as
r T IT
given above and the observed HCpTp is maximum. The procedure is step-
wise in that an independent variable is added one at a time in order of their
largest contribution to the correlation (Ref. 4-1). Hence the order of
4-12
-------
Table 4-5. ST/FTP Correlation Summary (CEV Fleet)
Short Test
Federal Short Cycle
NY/NJ Composite
Key Mode
(Laboratory)
Key Mode
(Garage)
Federal Three -Mode
(Laboratory)
Federal Three -Mode
(Garage)
2500 rpm
(Laboratory)
2500 rpm
(Garage)
Good
Data
/ \
Set
39
25
39
39
25
40
40
26
40
26
31
26
40
26
40
26
40
26
"r" -ST/FTP Correlation'0'
Coefficient
HC
0.87
0.91
0.93
0.92
0.92
0.95
0.61
0.53
0.92
0.57
0.53
0.97
0.73
0.73
0.88
0.51
0.39*
0.32*
0.87
0.79
0.80
0.68
0.52
0.94
0.76
0.73
0.78
0.69
0.42
0.62
0.47
0.37*
0.50
0.36*
CO
0.81
0.42
0.83
0.77
0.71
0.68
0.26*
0.39
0. 54
0.30*
0.31*
0.40
0.37
0.21*
0.52
0.08*
0.09*
-0.03*
0.08*
0.22*
0.48
0.20*
0.27*
0.34*
0.24*
0.21*
0.52
0. 12*
0.03*
0.39*
0.30*
0.25*
0. 14*
0. 25*
NO
X
0.62
0.47
0.53
0.61
0.51
0.61
0.79
0.20*
0.27*
0.86
0 . 04*
0 . 04*
0.89
0.03*
0.13*
0.92
-0.28*
0.08*
0.23*
0.23*
(a) First Good Data: This data set contains the observations of the first FTP and ST,
both of which are valid.
Second Good Data: This data set contains the second pair of FTP and ST obser-
vations, both of which are valid.
Average Data: This data set contains the average of the FTP and ST observations
on each car (for the Federal Short Cycle and NY/NJ Composite only).
(b) Number of cars in data set
(c) The correlation is statistically significant at the 95% confidence level except when
indicated by an asterisk.
4-13
-------
inclusion indicates the mode's relative importance. The ordering of the
modes varies depending on the ST and pollutant under study.
A multiple regression analysis was performed for the three-
mode volumetric tests on first good data. The purpose of this analysis was
to empirically determine the linear combinations of the three-mode readings
that have maximum correlation with the FTP. The linear combinations are
composed of like constituents. Thus, each linear combination can be con-
sidered as a weighted observation on HC, CO, and NOx- The results are
shown in Table 4-6, along with the maximum correlation coefficient using
only a single reading on each constitutent. As can be seen from Table 4-6,
the weighted combination correlation coefficients are not significantly
higher than the correlation coefficient of the best single reading.
Table 4-6. ST/FTP Correlations for Weighted Mode Tests
(CEV Fleet) (first good data only)
Short Test
Key Mode
Laboratory
Garage
Federal Three-
Mode
Laboratory
Garage
N(a>
40
40
31
40
Weighted Corre-^b)
lation Coefficient
HC
0.93
0.91
0.91
0.81
CO
0.55
0.58
0.48
0.53
NO
X
0.83
0.90
tc\
Best Single -Mode* ' .
Correlation Coefficient* '
HC
0.92 (I)
0.88 (I)
0.87 (H)
0.78 (I)
CO
0.54 (I)
0.52 (I)
0.48 (I)
0.52 (I)
NO
X
0.79 (H)
0.89 (H)
(a)
(b)
(c)
Number of cars in data set
Correlations are statistically significant at the 95% confidence level
H - high speed mode
I = idle mode
4-14
-------
4.2. 1.3
Correlation Sensitivity Analysis
As previously mentioned, the sensitivity of the analysis to the
data used can be assessed by using the confidence interval. To dramatize
this sensitivity, a worst case approach was examined by deleting selected
extreme data points from the existing data. Recalculation of the correlation
coefficient was performed to illustrate the variability due to the sample.
This was done in a sequential manner for the Federal Short Cycle. A sum-
mary of the analysis results for the Federal Short Cycle-FTP correlations
is shown in Table 4-7. A review of Table 4-7 values indicates that the re-
sults are extremely sensitive to a small percentage of the data points.
4.2.1.4 Discussion of Selected Correlation Analysis Results
4.2.1.4.1
Shortcomings of the Correlation Coefficient
The main usage of the correlation coefficient is as an indicator
of direct relatability between ST and FTP. In this respect it has a number of
deficiencies. The computed correlation coefficient is sensitive to the location
Table 4-7. Correlation Coefficients for Selected Car Deletions;
Federal Short Cycle vs FTP (CEV Fleet)
Number of Cars Deleted
0
1
2
3
4
(a)
Correlation Coefficient
HC
0.872
0.657
0.656
--
--
CO
0.810
0.673
0.639
--
--
NO
X
0.621
0.690
0.633
0.823
0.755
^Significant at the 95% confidence level
4-15
-------
of a small percentage of the data, as shown in Table 4-7. It is a summary
statistic in that all the information contained in the data is compressed into
a single number (this is alleviated to some degree by examination of the
scattergrams).
It is difficult to infer air quality impact from correlation
statistics except in the broadest sense, and a tradeoff analysis is virtually
impossible based solely on correlation coefficients.
4.2.1.4.2 Mode Tests vs Bag Tests
On the basis of HC and CO correlation, the bag tests (Federal
Short Cycle and NY/NJ Composite) are preferable to the mode- or volumetric-
type ST. The volumetric STs, in general, show deficiencies in tracking CO.
The high-speed modes, however, have superior NO correlation.
JL
4.2.1.4.3 Laboratory Analyzers vs Garage Instruments
The largest difference between the correlation results of the
two measurement techniques occurs on the second good data sets. There
is a greater variation in the correlation estimates of first good data and
second good data for the garage analyzer than for the laboratory analyzer,
as shown in Table 4-5. This is most likely due to the combination of low
CO values for the CEV fleet, small sample size, and less accurate
instrumentation.
The most striking difference between laboratory and garage
data is for HC on the Federal Three-Mode. The laboratory measurements
indicate the best mode to be high speed, while the garage readings indicate
the idle mode as superior. This is inconsistent with the results for HC on
the Clayton Key Mode, and may be attributed to the difference in the sample
sizes of the Federal Three-Mode and the Clayton Key Mode tests.
CO correlation deficiency is common to both measurement
techniques. Due to the low concentration of CO being emitted, this may be
a measurements problem, in general, rather than a deficiency in ST
structure.
4-16
-------
4.2.1.4.4 ST Correlation Ratings
The following qualitative rating scale was used to rate the ST:
Rating Description
(U) Unacceptable Constituent is uncorrelated at 95%
confidence level
(P) Poor Constituent is correlated at the 95%
confidence level, but with correla-
tion less than 0.6
(F) Fair Correlation between 0. 6 and 0. 7
(G) Good Correlation between 0.7 and 0. 9
(E) Excellent Correlation between 0. 9 and 1. 0
For rating the three-mode volumetric ST, the mode with the
highest rating was used. Table 4-8 shows the ratings of the ST on each
pollutant on this basis.
In general, the STs have less difficulty tracking HC than CO
and NO . Excluding the Unloaded 2500 rpm ST (which has either "P11 or "U"
JL
ratings for all three pollutants), the bag-type and modal STs all have "G" to
"E" ratings for HC. In the case of CO, the bag-type STs have "G" ratings,
whereas the modal STs are rated in the "P" category. This situation is re-
versed in the case of NO , where the modal STs have "G" ratings and the
x
bag-type STs are rated "F" to "P". Hence, the choices among the STs for
CO and NO implementation may be more limited than for HC.
4.2.2 Contingency Table Analysis
The contingency table analysis is used to establish the ST
pass-fail levels for each pollutant. The contingency table is defined in
Table 4-9, along with its associated parameters. A pictorial demonstration
of its application to a given data set is shown in Figure 4-2. This figure
4-17
-------
shows that, for a given data set, ST cut-points must be established in order
to compute the elements of the contingency table. Four basic approaches
for cut-point determination were considered, which are described as follows.
Table 4-8. ST Correlation Ratings
Short Test
Federal Short Cycle
NY/NJ Composite
Key Mode
Laboratory
Garage
Federal Three -Mode
Laboratory
Garage
2500 rpm Unloaded
Laboratory
Garage
Rating
HC
G
E
E (I)(a)
G(I)
G(H)
G(I)
P
P
CO
G
G
P(D
P(D
P(D
P(D
U
U
N0x
F
P
G(H)
G(H)
U
I = idle mode, H = high speed mode
4-18
-------
Table 4-9. Contingency Table
Predicted =
Short Test
Pass
Fail
Total
True = FTP
Pass
a
c
a + c
Fail
b
d
b + d
Total
a + b
c + d
n = a + b
+ c + d
a = number of correctly passed vehicles (PP)
b = number of error of omission (E )
c = number of error of commission (E )
d = number of correctly failed vehicles (FF)
Sensitivity = a /(a + c)
Specificity = b/(b + d)
False positive error = b/(a + b)
False negative error = c/(c + d)
4
Correlation index =
ad - be
[(a + b)(a + c)(b + d)(c + d)]
1/2
4-19
-------
FF, CORRECTLY
FAILED VEHICLES
Er, ERROR OF
c
COMMISSION
ST CUT-POINT
i
.
i
i
EQ, ERROR OF
^ , OMISSION
13
<
PP, CORRECTLY
PASSED VEHICLES
ai
ti
FTP MEASUREMENT
Figure 4-2. Contingency Table Representation
4-20
-------
4.2.2.1 Analysis Methods Examined
4.2.2.1.1 Maximum Correlation Method
In this method, the ST cut-point is selected so that the corre-
lation index (as defined in Table 4-9) is maximized. This is an impartial
procedure for finding the STs that give the best correlation .with the FTP
under the terms of the contingency table. Figure 4-3 graphically illustrates
the procedure. This method provides for no policy decision.
4.2.2.1.2 Bounded Errors of Commission Method
The ST cut-points are selected to minimize the errors of
omission while holding the errors of commission below a specified level.
This method permits a direct answer to the question, "For a given permis-
sible level of errors of commission, what level of errors of omission must
be accepted, and with what impact on air quality (inferred from the number
of FF vehicles) ?" This method is pictorially demonstrated in Figure 4-4.
The policy decision is the maximum allowable errors of commission.
4.2.2.1.3 Weighted Errors Method
The strategy used in this method is, as indicated by
Figure 4-5, to minimize a linear combination of the errors of commission
and the errors of omission. The linear combination represents cost to the
public, where the weights indicate the relative importance of the two types
of costs: those incurred by manufacturers, versus those due to deterioration
of air quality. Air quality impact is inferred from the level of FF vehicles.
The policy decision is the cost structure; that is, the specification of the
weights.
4.2.2.1.4 Percent Rejection Method
The ST cut-points are determined so that a specified per-
centage of the population is failed by the ST. This is shown graphically in
Figure 4-6. The policy decision is the percentage to be rejected by the ST.
4-21
-------
PARAMETRIC
TECHNIQUE
USE LINEAR REGRESSION
AS A MODEL
X2 -
+ B
DATA ANALYTIC
TECHNIQUE
SELECT SHORT TEST LEVEL
SUCH THAT THE TABLE
CORRELATION IS MAXIMUM
SOLVE FOR C2 WHEN
I.E..
\s>
ST . [ :S:
CUT-POINT -
FTP MEASUREMENT
Figure 4-3. Maximum Correlation Method
MINIMIZE EQ SUBJECT TO EC < Y%
PARAMETRIC TECHNIQUE ONLY. USE BIVARIATE
NORMAL PROBABILITY MODEL
NOT TO |
EXCEED Y% l
ST
CULJLOJJiT*r J. "
T 1
T"1
MINIMIZE
Q.
t
FTP MEASUREMENT
Figure 4-4. Bounded Errors of
Commission Method
4-22
-------
MINIMIZE aEr + 0E
C Q
PARAMETRIC
TECHNIQUE
USE BIVARIATE NORMAL
PROBABILITY MODEL
WEIGHT a
ST
DATA ANALYTIC
TECHNIQUE
SELECT THE SHORT TEST LEVEL
SO THAT aEc + #EO IS MINIMUM
WEIGHT/?
i a.
IE
FTP MEASUREMENT'
Figure 4-5. Weighted Errors Method
REJECT a* OF THE SAMPlf
PARAMETRIC
TECHNIQUE
USE BIVARIATE NORMAL
PROBABILITY MODEL
DATA ANALYTIC
TECHNIQUE
SELECT THE SHORT TEST LEVEL UNTIL
SPECIFIED PERCENT IS REJECTED
ST
SHORT TEST
REJECTION
REGION
l!
FTP MEASUREMENT
Figure 4-6. Percent Rejection Method
4-23
-------
4.2.2.2 Procedural Techniques Utilized
The techniques used to compute the ST cut-points and the
contingency table entries are classified as data analytic and parametric.
The data analytic technique uses the data directly without resort to a model.
The parametric procedure uses a model of the data.
4.2.2.2.1 Data Analytic Technique
The cut-point for each pollutant is determined individually.
For each ST cut-point, the table entries are calculated by counting the
number of data points in each of the appropriate regions indicated in
Figure 4-2. Each ST cut-point is then iteratively varied until the objective
of the particular strategy is achieved. This set of ST cut-points is then
taken as the solution to the strategy under study.
This procedure was not applied to the method of bounded
errors of commission. The bound typically ranged from 5 percent to 0.1
percent. In terms of actual counts, this range is 2. 0 to 0.04 cars for the
CEV fleet. The results would thus be sensitive to a very small portion of
the data.
4.2.2.2.2 Parametric Technique
The data are first modeled by using a bivariate normal
distribution as shown in Figure 4-7. Thus the correlation coefficient, mean
values, and standard deviations are computed from the data and substituted
into the model. The ST cut-points are then determined by using the model.
Figure 4-8 indicates the pertinent probability calculations for predicting the
table entries. The predicted table entries are shown in Figure 4-9.
Figure 4-10 shows the equations to be solved to determine the ST cut-points.
After the ST cut-points have been determined, the contingency
table results are calculated using both the actual data points and the model
of the data.
4-24
-------
BIVARIATE NORMAL DISTRIBUTION
e U(l -/><
WHERE nvfji2 * POPULATION MEANS
0-^0-2 POPULATION STANDARD DEVIATIONS
P CORRELATION COEFFICIENT
Xj - FTP MEASUREMENT
X2 - ST MEASUREMENT
Figure 4-7. Parametric Model
PROBABILITY OF ERROR OF COMMISSION
X2)
Pr{ XjC2} j j D (Xr
-C2}-// D(X1.X2)«1d!
WHERE Cj CRITICAL FTP LEVEL
C2 - CRITICAL ST LEVEL
Figure 4-8. Probability Equations
4-25
-------
EXPECTED ERRORS OF COMMISSION AND OMISSION
EC - N x [ PROBABILITY OF ERROR OF COMMISION ]
E N x [ PROBABILITY OF ERROR OF OMISSION ]
o
EXPECTED CORRECT FAILURES
FF N x [ PROBABILITY OF CORRECT FAILURE ]
PP = N - FF - E - Ert
c o
WHERE N - NUMBER OF CARS USED
Figure 4-9. Expected Values
SOLVE FOR C2
PERCENT REJECTION
+00 +00
0=Pr^X>Cl=l /
u rr \*2>^2> J J
-C0c2
WEIGHTED ERRORS
MIN (QE. + J8EJ; a + £ 1
/» co
BOUNDED ERRORS OF COMMISSION:
MIN(E); E
-------
4.2.2.3
Selected Analysis Methods
Only two of the above four cut-point-level selection strategies
were investigated in any detail: the maximum correlation method and the
bounded errors of commission method. The maximum correlation method
was chosen for comparison with the previous correlation analysis. The infor
mation contained in an analysis under the other two strategies is identical
for varying policy decisions. That is, as the policy is varied under each
strategy, the resulting loci of EC, EQ> and FF are identical. Hence, the
bounded errors of commission method was chosen for its particular rele-
vance to the cost to manufacturers and air quality impact.
As the emissions standard to which the CEV fleet was de-
signed is uncertain, four sets of FTP cut-points were used in the analyses.
These are specified in Table 4-10.
Table 4-10. Assumed FTP Levels (CEV Fleet)
Level
I
II
HI
IV
Emission Level, gm/mi
HC
0.41
0.60
0.75
0.90
CO
3.4
5.0
7.0
9.0
NO
X
3.1
3.1
3.1
3.1
4-27
-------
4.2.2.4 Maximum Correlation Analysis Results
The problem of presenting the results can best be seen while
observing relationships of EC, EQ, and FF to changing FTP level. For
example, Figure 4-11 illustrates a typical plot for HC, using the data
analytic calculation technique. Similarly, Figure 4-12 shows the same
results using the parametric calculational technique on the actual data points
only, while Figure 4-13 shows the results as predicted from a model of the
data. Trends are clearly more visible in the predicted population results.
Although these trends are an intrinsic component of the model, the actual
magnitudes and rates of change of the trends are due to the data.
A summary of the results of the maximum correlation
analysis for the predicted population of the CEV fleet is shown in Tables 4-11
through 4-14. For N = 40 or 39, a. computed table correlation coefficient
greater in magnitude than 0.31 indicates that the ST and FTP pollutants are
statistically correlated at the 95 percent confidence level. For N = 31,
this threshold is 0.35. Figures 4-13 to 4-24 depict the relationship of
EC, EQ> and FF to changing FTP level for HC and CO on the predicted
population basis. Figure 4-25 shows the variation of E and E for NO .
r ocx
The correlation index of the contingency table, as defined in
Table 4-9, is substantially different than the computed correlation coeffi-
cients of Sec. 4.2.1. Although the relative ranking of the ST may be similar
to that of Table 4-8, experience has shown that contingency table correlation
index is an unreliable indicator of relatability. For example, consider the
extreme case where EQ = 0.0, EC = 0.0, and FF = 0.01%. In this case the
correlation index will be 1.0; however, 99. 99% of the data are in the
correctly passed group, and the correlation index tells nothing about 99.99%
of the data. This example also indicates that the correlation index is a
function of the ST and FTP cut-points. Although this is desirable for policy
analysis, tradeoffs are best inferred by directly observing the pertinent
quantities.
4-28
-------
O FSC
A NY/NJ
D 2500 rpm
UNLOADED
FF
0.4 0.5 0.6 0.7 0.8 0.9
1975 FTP HC LEVEL, gm/mi
1.0
Figure 4-11. Variation of Ec, Eo, and FF with HC
FTP Level; Maximum Correlation Method; Data
Analytic Technique; CEV Fleet
4-29
-------
o
50
45
40
35
30
25
20
15-
10
O FSC
A NY/NJ
O 2500 rpm UNLOADED
Ec
EQ
FF
>a
I
I
.4 0.5 0.6 0.7 0.8
1975 FTP HC LEVEL, gm/mi
0.9
1.0
Figure 4-12. Variation of Ec, Eo, and FF with HC
FTP Level; Maximum Correlation Method; Para-
metric Technique; CEV Fleet
4-30
-------
O FSC
A NY/NJ
D 2500 rpm UNLOADED
FF
0.4
0.5 0.6 0.7 0.8
1975 FTP HC LEVEL, gm/mi
Figure 4-13. Variation of Ec, EQ> and FF with HC
FTP Level; Maximum Correlation Method; Pre-
dicted Population Technique; CEV Fleet
4-31
-------
Table 4-11. Maximum Correlation Summary,
FTP Level I (CEV Fleet)
Short Test
Federal
Short Cycle
NY/NJ
Composite
Key Mode
(Laboratory)
Key Mode
(Garage)
Federal
Three-Mode
(Laboratory)
Federal
Three-Mode
(Garage)
2500 rpm Unloaded
2500 rpm Unloaded
(Garage)
N
39
39
40
40
31
40
50
40
Test
Mode
High
Low
Idle
High
Low
Idle
High
Low
Idle
High
Low
Idle
Pollutant
HC
CO
NOX
HC
CO
NO
X
HC
CO
N0x
HC
CO
NOX
HC
CO
NOX
HC
CO
HC
CO
HC
CO
HC
CO
N0x
HC
CO
NOX
HC
CO
NOX
HC
CO
HC
CO
HC
CO
HC
CO
NOX
HC
CO
*Ec
6.49
10.8
16.3
5.38
12.0
16.7
10.6
23.1
10.0
11.5
20.4
33.6
5.21
17.2
30.4
8.95
20.9
9.05
24.2
6.30
17.8
6.29
23.9
5.49
7.70
21.7
42.9
7.64
17. 1
37.8
8.47
23.7
8.99
24.1
8. 17
17.9
12.2
22.4
32.4
11.8
25.6
*Eo
8.53
8.23
5.26
6.72
8.94
5.32
17.2
18.5
4.40
19.4
16.6
7.13
6.51
14.4
6.9
13.4
16.8
13.6
18.9
8.27
14.6
8.60
23.5
2.72
11.39
21.3
5.78
11.3
16.9
5.55
12.4
18.6
13.4
18.9
II. 7
14.7
21.1
17.7
6.67
20.2
19.9
%FF
58.3
28.0
9.71
60.1
27.3
9.65
50.1
25.3
11.2
47.8
27.2
8.43
60.8
29.4
8.69
53.9
26.6
53.6
24.4
58.9
28.7
60.3
26.0
8.96
57.5
28.2
5.90
57.6
32.7
6.14
54.9
24.7
53.7
24.4
55.5
28.7
46.1
25.6
8.09
46.9
23.5
Table
Correlation
Index<">
0.667
0.596
0.372
0.730
0.556
0.363
0.405
^
0. 166
0.532
0.344
0.257'
0.105°
0.737
0.363
0.147
0.514
0.242'
0.507
0.136
0.675
0.346
0.661
0.050"
0.644
0.571
o. MO'
0.012'
0.575
0.321
0.063
0.544
0. 151
0.511
0. 137 ""
0.563
0.344
0.300
0.194*
0.121°
0.323
0.089°
(a)
The correlation ii statistically significant at the 95% confidence level except where Indicated
by an asterisk
4-32
-------
Table 4-12. Maximum Correlation Summary,
FTP Level II (CEV Fleet)
Short Test
Federal
Short Cycle
NY/NJ
Composite
Key Mode
(Laboratory)
Key Mode
(Garage)
Federal
Three-Mode
(Laboratory)
Federal
Three-Mode
(Garage)
2500 rpm Unloaded
(Laboratory)
2500 rpm Unloaded
(Garage)
N
39
39
40
40
31
40
40
40
Test
Mode
High
Low
Idle
High
Low
Idle
High
Low
Idle
High
Low
Idle
Pollutant
HC
CO
HC
CO
HC
CO
HC
CO
HC
CO
HC
CO
HC
CO
HC
CO
HC
CO
HC
CO
HC
CO
HC
CO
HC
CO
HC
CO
HC
CO
HC
CO
%EC
7.90
7.10
6.42
8.38
14.0
30.8
15.5
25.4
6.28
19.6
11.4
26.2
11.6
32.7
7.69
20.5
7.73
35.6
9.68
30.8
9.60
21.8
10.7
31.7
11.5
32.7
10.3
20. S
16.3
29.1
15.8
35.7
%Eo
8.31
2.36
6.70
2.50
15.2
7.58
16.8
7.03
6.50
6.35
12.3
7.01
12.5
7.64
8. 10
6.37
8.56
11.3
10.99
10.5
10.9
8.89
11.5
7.55
12.4
7.63
11.1
6.38
18.2
7.30
17.5
7.90
% FF
44.9
5.63.
46.5
5.48
37.7
9.51
36.0
10.1
46.3
10.7
41.1
9.80
40.9
9.18
45.4
10.4
47.8
12.2
45.4
13.0
45.4
14.6
41.9
9.27
41.1
9.18
42.4
10.4
35.3
9.52
35.9
8.91
Table
Correlation
Index(a)
0.675
0.510
0.737
0.467
0.415
0.142*
0.354
0.223*
0.744
0.321
0.523
0.209*
0.519
0.116*
0.683
0.304*
0.670
0.46
0.582
0.127*
0.585
0.297*
0.553
0.129*
0.521
0.116*
0.572
0.302*
0.309
0.166*
0.332
0.075*
'a'The correlation is statistically significant at the 95% confidence level except where indicated
by an asterisk
4-33
-------
Table 4-13. Maximum Correlation Summary,
FTP Level HI (CEV Fleet)
Short Teat
Federal
Short Cycle
NY/NJ
Composite
Key Mode
(Laboratory)
Key Mode
(Garage)
Federal
Three -Mode
(Laboratory)
Federal
Three-Mode
(Garage)
2500 rpm Unloaded
(Laboratory)
2500 rpm Unloaded
(Garage)
N
39
39
40
40
31
40
40
40
Teat
Mode
High
Low
Idle
High
Low
Idle
High
Low
Idle
High
Low
Idle
Pollutant
HC
CO
HC
CO
HC
CO
HC
CO
HC
CO
HC
CO
HC
CO
HC
CO
HC
CO
HC
CO
HC
CO
HC
CO
HC
CO
HC
CO
HC
CO
HC
CO
%EC
8.49
1.16
6.79
1.57
16.2
29.4
18.1
20.7
6.61
6.98
12.8
21.9
13.0
32.5
8.26
14.1
8.46
42.8
10.8
33.4
10.7
18.7
11.9
30.8
12.9
32.4
11.4
14.2
19.2
26.4
18.5
37.9
*Eo
7.50
0.11
6.14
0.12
12.7
1.20
13.9
1.13
5.91
13.1
10.7
1.12
10.9
1.19
7.33
1.05
7.93
2.59
9.97
2.47
9.89
2.19
10.1
1.18
10.8
1.19
9.71
1.05
15.1
1.15
14.6
1.22
%FF
34.6
0.21
36.0
0.21
28.4
1.40
27.2
1.47
35.12
1.54
31.5
1.42
31.3
1.35
34.9
1.49
38.0
2.74
36.0
2.86
36.1
3.14
32.1
1.36
31.4
1.35
32.5
1.49
27.1
1.39
27.6
1.32
Table
Correlation
IndexW
0.673
0.312*
0.736
0.267*
0.413
0.082*
0.352
0.134*
0.742
0.207*
0.522
0.125*
0.515
0.066*
0.682
0.193*
0.670
0.031*
0.582
0.086*
0.586
0.214*
0.552
0.074*
0.519
0.067*
0.570
0.192*
0.308
0.097*
0.330
0.043*
''The correlation ia statistically significant at the 95% confidence level except where indicated
by an asterisk
4-34
-------
Table 4-14. Maximum Correlation Summary,
FTP Level IV (CEV Fleet)
Short Teat
Federal
Short Cycle
NY/NJ
Composite
Key Mode
(Laboratory)
Key Mode
(Garage)
Federal
Three-Mode
(Laboratory
Federal
Three-Mode
(Garage)
2500 rpm Unloaded
(Laboratory)
2500 rpm Unloaded
(Garage)
N
39
39
40
40
31
40
40
40
Teat
Mode
High
Low
Idle
High
Low
Idle
High
Low
Idle
High
Low
Idle
Pollutant
HC
CO
HC
CO
HC
CO
HC
CO
HC
CO
HC
CO
HC
CO
HC
CO
HC
CO
HC
CO
HC
CO
HC
CO
HC
CO
HC
CO
HC
CO
HC
CO
%EC
8.49
0.05
6.66
0.09
17.4
22.3
19.7
12.3
6.40
5.49
13.4
13.5
13.6
26.5
8.24
6.26
8.69
41.8
11.4
29.0
11.2
11.0
12.4
24.2
13.5
26.3
11.7
6.32
21.3
18.7
20.4
33.9
%Eo
6.28
0.0
5.22
0.0
9.85
0.08
10.7
0.08
4.93
0.07
8.69
0.07
8.79
0.08
6.14
0.07
6.89
0.30
8.51
0.29
8.44
0.26
8.23
0.08
8.73
0.08
7.94
0.07
11.8
0.08
11.4
0.08
% FF
25.3
0.00
26.4
0.00
20.2
0.09
19.4
0.09
25.1
0.10
22.9
0.09
22.8
0.08
25.4
0.09
28.9
0.31
27.3
0.32
27.4
0.35
23.3
0.09
22.9
0.08
23.6
0.09
19.8
0.09
20.1
0.08
Table
Correlation
Index!*)
0.665
0.138*
0.729
0. 104*
0.401
0.030*
0.341
0.053*
0.734
0.092*
0.512
0.048*
0.505
0.024*
0.674
0.083*
0.665
0.014*
0.576
0.041*
0.580
0.113*
0.542
0.027*
0.509
0.024*
0.561
0.083*
0.299*
0.036*
0.321
0.015*
'a*The correlation i« »tatistically significant at the 95% confidence level except where indicated
by an aateriik
4-35
-------
501~
45
40
35
\
UJ
o
25
20
15
10
'
I
O HIGH SPEED
A LOW SPEED
D IDLE
FF
\N
-O
1.4 0.5 0.6 0.7 0.8
1975 FTP HC LEVEL, gm/mi
0.9
1.0
Figure 4-14. Variation of Ec, Eo, and FF with HC
FTP Level; Federal Three-Mode Test; Maximum
Correlation Method; Predicted Population of CEV
Fleet
4-36
-------
O HIGH SPEED
A LOW SPEED
D IDLE
FF
0.5 0.6 0.7 0.8
1975 FTP HC LEVEL, gm/mi
Figure 4-15. Variation of Ec, E0, and FF
with HC FTP Level; Key Mode Test;
Maximum Correlation Method; Predicted
Population of CEV Fleet
4-37
-------
LU
O
CC
UJ
Q.
15
10
5 -
O FSC
A NY/NJ
D 2500 rpm
UNLOADED
345 678
1975 FTP CO LEVEL, gm/mi
Figure 4-16. Variation of Ec, Eo, and FF with CO FTP Level;
Maximum Correlation Method; Predicted Population of CEV
Fleet
10
4-38
-------
451-
O HIGH SPEED
A LOW SPEED
D IDLE
5678
1975 FTP CO LEVEL, gm/mi
Figure 4-17. Variation of Ec, Eo, and FF with CO FTP Level;
Federal Three-Mode Test; Maximum Correlation Method; Pre-
dicted Population of CEV Fleet
4-39
-------
30
25
<
UJ
UJ
O
OS
20
15
10
O HIGH
A LOW SPEED
D IDLE
678
1975 FTP CO LEVEL, gm/mi
10
Figure 4-18. Variation of Ec. Eo, and FF with CO FTP Level;
Key Mode Test; Maximum Correlation Method; Predicted
Population of CEV Fleet
4-40
-------
50
40
o
FF
< 30
o
o
UJ
D_
20
10
0
0.4
0.5
0.6 0.7 0.8
1975 FTP HC LEVEL, gm/mi
0.9
1.0
Figure 4-19. Variation of Ec, Eo> and FF with HC FTP Level;
Unloaded 2500 rpm Test; Garage Instruments; Maximum Cor-
relation Method; Predicted Population of CEV Fleet
4-41
-------
60
50
40
o
<
30
h«M _^
o 20
10
O HIGH SPEED
A LOW SPEED
D IDLE
k
'Ik
\
FF
I
0.4
0.5
0.6
0.7
0.8
0.9
1975 FTP HC LEVEL, gm / mi
1.0
Figure 4-20. Variation of Ec, Eo, and FF with HC FTP Level;
Federal Three-Mode Test; Garage Instruments; Maximum Cor-
relation Method; Predicted Population of CEV Fleet
4-42
-------
UJ
O
a:
60
50
40
£ 20
10
R
XX
O HIGH SPEED
A LOW SPEED
D IDLE
---- E
^-6
0
0.4
0.5 0.6 0.7 0.8
1975 FTP HC LEVEL, gm / mi
0.9
1.0
Figure 4-21. Variation of Ec, E0, and FF with HC FTP Level;
Key Mode Test; Garage Instruments; Maximum Correlation
Method; Predicted Population of CEV Fleet
4-43
-------
o
O£
401-
30
J" 20
10
\
FF
678
1975 FTP CO LEVEL, gm/mi
Figure 4-22. Variation of Ec, Eo, and FF with CO FTP
Level; Unloaded 2500 rpm Test; Garage Instruments;
Maximum Correlation Method; Predicted Population of
CEV Fleet
40 r-
O HIGH SPEED
A LOW SPEED
D IDLE
5678
1975 FTP CO LEVEL, gm/mi
10
Figure 4-23. Variation of Ec, E0, and FF with CO FTP
Level; Federal Three-Mode Test; Garage Instruments;
Maximum Correlation Method; Predicted Population of
CEV Fleet
4-44
-------
o
O£
UJ
O.
10-
C -.
O HIGH SPEED
A LOW SPEED
D IDLE
5678
1975 FTP CO LEVEL, gm/ml
Figure 4-24. Variation of Ec, Eo, and FF with CO FTP Level;
Key Mode Test; Garage Instruments; Maximum Correlation
Method; Predicted Population of CEV Fleet
4-45
-------
201-
15
u,°
§ 10
°
KEY
HIGH
\
F3M
HIGH
|_
NY/NJ
NY/NJ
2500 rpm IDL£
UNLOADED F3IV1
LOW
0
10
20
30
PERCENT
40
50
60
Figure 4-Z5. Variation of Eo and Ec at NOX Level of 3. 1 gm/mi;
CEV Fleet; Maximum Correlation Method; Predicted Population
of CEV Fleet
-------
By examination of the values of E , E , and FF in Tables 4-11
through 4-14, the ST can be seen to follow the correlation ratings of Table 4-8.
For example, on the Key Mode (Laboratory) for HC
High Speed:
Low Speed:
Idle:
Ec
10.6
11.5
5.21
Eo
17.2
19.4
6.51
FF
50. 1
47.8
60.8
As the idle mode has the highest percent FF and lowest percent E and E ,
it is a superior mode for HC. This is consistent with the ratings in Table 4-8.
The summary tables and the summary graphs do not clearly
favor a single FTP level as a design level for the CEV fleet. However, the
CO plots suggest that levels II, III, and IV are too high for CO as the percent
FF dips below 15 percent on all tests.
The maximum correlation method does not admit a policy
decision after the FTP level has been set. Thus its usefulness is restricted
for purposes of tradeoff analysis.
4.2.2.5 Bounded Errors of Commission Analysis Results
4.2.2.5.1 Single-Constituent Tests
For the CEV fleet, the bound of errors of commission was
varied from 5 percent to 1 percent in 1 percent increments, with the values
0. 5 percent and 0. 1 percent also included. An analysis was made for each
of the FTP levels of Table 4-10. The results of the analysis are summarized
in the following three sections. The data plotted are for the predicted CEV
population. Since the exact FTP value is uncertain, only general observations
can be made.
4-47
-------
4.2.2.5.1.1 Hydrocarbon Emissions
The variation of E , E , and FF as a function of HC cut-point
is displayed in Figures 4-26 through 4-35 for each ST examined, and for the
range of HC FTP values selected in Table 4-10 (HC = 0.41 to 0.90). The
figures correspond to the following ST/FTP level spectrum:
Short Test
Federal Short Cycle
NY/NJ Composite
Clayton Key Mode
(Laboratory)
Clayton Key Mode
(Garage)
Federal Three -Mode
(Laboratory)
Federal Three -Mode
(Laboratory)
Federal Three -Mode
(Garage)
Federal Three -Mode
(Garage)
2500 rpm Unloaded
(Laboratory)
2500 rpm Unloaded
(Garage)
FTP HC Level
0.41
X
X
X
X
X
X
X
X
0.60
X
X
X
0.75
X
X
0.90
X
X
X
X
X
X
X
X
Figure No.
4-26
4-27
4-28
4-29
4-30
4-31
4-32
4-33
4-34
4-35
The graphical displays indicate the general nature of the tradeoff avail-
able for policy formulation. Reducing the errors of commission (Ec) increases
the errors of omission (E ) and decreases the correct failures (FF). To
illustrate specific values and trends among the STs, Tables 4-15 and 4-16
4-48
-------
60
55
45
40
t 35
o
LU
a. 25
20
15
10
8.
\
\
\
\
\
\\
\\
\\
FF
0.60
4y
/ 1975
/ r- FTP
LEVEL \ i
0.41
°- 60
\\ /
-
0.75
0.90
5 -
0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1
HC CUT-POINT, gm/mi
Figure 4-26. Variation of Ec, E0, and FF with HC Cut-Point;
CEV Fleet; Federal Short Cycle; Bounded Errors of Commis-
sion Method
4-49
-------
I
Ul
o
^M M» HV^B C
0
FF
0
0
20 30 40
HC CUT-POINT, ppm
50
60
Figure 4-27. Variation of Ec, Eo> and FF with HC Cut-Point;
CEV Fleet; NY/NJ Composite Test; Bounded Errors of Com-
mission Method
-------
60
FF
1975 FTP
1975 FTP = 0.90
AT 0.90 FF <10%
FOR HIGH SPEED AND
LOW SPEED
0.41
HIGH SPEED
LOW SPEED
IDLE
400 500
HC CUT-POINT, ppm
Figure 4-28. Variation of Ec, Eo, and FF with HC Cut-Point;
CEV Fleet; Key Mode Test; Bounded Errors of Commission
Method
-------
ts>
1975 FTP - 0.41
1975 FTP - 0.9
H HIGH SPEED
L LOW SPEED
I IDLE
30 40 50 60 70
HC CUT-POINT, ppm
Figure 4-29- Variation of Ec, E0, and FF with HC Cut-Point;
CEV Fleet; Key Mode Test; Garage Instruments; Bounded
Errors of Commission Method
-------
H - HIGH SPEED
L - LOW SPEED
IDLE
I
Ul
CO
0
..--'
I
40
80
120
200 240 280
HC CUT-POINT, ppm
320
360
400 440
Figure 4-30. Variation of Ec, E0, and FF with HC Cut-Point;
CEV Fleet; 1975 FTP HC Standard = 0.41 gm/mi; Federal
Three-Mode Test; Bounded Errors of Commission Method
-------
Ut
.*».
H HIGH SPED
L - LOW SPEED
IDLE
0
200 250 300 350 400 450 500 550
HC CUT-POINT, ppm
600
650
Figure 4-31. Variation of Ec, Eo, and FF with HC Cut-Point;
CEV Fleet; 1975 FTP HC Standard = 0.9 gm/mi; Federal
Three-Mode Test; Bounded Errors of Commission Method
-------
H - HIGH SPEED
L = LOW SPEED
I = IDLE
^
Ul
LU
25
30
35
40 45 50
HC CUT-POINT, ppm
55
60
65
70
Figure 4-32. Variation of Ec> Eo, and FF with HC Cut-Point;
CEV Fleet; 1975 FTP Standard = 0.41 gm/mi; Federal Three-
Mode Test; Garage Instruments; Bounded Errors of Commis-
sion Method
-------
40 r~
H HIGH SPEED
L LOW SPEED
I IDLE
'\ '\ ^x \
' \ V v
50 60 70
HC CUT-POINT, ppm
Figure 4-33. Variation of Ec, Eo, and FF with HC Cut-Point;
CEV Fleet; 1975 FTP Standard = 0. 9 gm/mi; Federal Three-
Mode Test; Garage Instruments; Bounded Errors of Commis-
sion Method
-------
601
50)
40]
1975 FTP
LEVELX '
x
0.41
301
201
101
0.60
0.75
0.90
0.60,
0.41 0.60
0.90^
H = HIGH SPEED
L = LOW SPEED
I - IDLE
2200
UC
Figure 4-34. Variation of Ec, EQ, and FF with HC Cut-Point;
CEV Fleet; Unloaded 2500 rpm Test; Bounded Errors of Com-
mission Method
-------
55
50
40
in
oo
1975 FTP LEVEL - 0.41
FF
H = HIGH SPEED
L LOW SPEED
I - IDLE
1975 FTP LEVEL 0.9
100
120
140 160 180
HC CUT-POINT, ppm
200
220
240
Figure 4-35. Variation of Ec, Eo, and FF with HC Cut-Point;
CEV Fleet; Unloaded 2500 rpm Test; Garage Instruments;
Bounded Errors of Commission Method
-------
Table 4-15. Comparison of Selected ST Hydrocarbon Results.-
CEV Fleet, Bounded Errors of Commission Analysis,
HC FTP Level = 0.90 gm/mile (Ec = constant = 5%)
Short Test
Federal Short Cycle
NY/NJ Composite
Clayton Key Mode (Laboratory)
Idle
Low Speed
High Speed
Clayton Key Mode (Garage)
Idle
High Speed
Federal Three -Mode (Laboratory)
Idle
Low Speed
High Speed
Federal Three -Mode (Garage)
Idle
Low Speed
High Speed
2500 rpm Unloaded (Laboratory)
2500 rpm Unloaded (Garage)
Parameter, %
Eo
9
6.5
6
21
19.5
9.5
16
15
15
10
14
16
15
23
23
FF
22
25
22
9
10
22
16
21
21
25
17
16
17
8
9
4-59
-------
Table 4-16. Comparison of Selected ST Hydrocarbon Results:
CEV Fleet, Bounded Errors of Commission Analysis,
HC FTP Level = 0.41 gin/mile (E = constant = 5%)
Short Test
Federal Short Cycle
NY/NJ Composite
Clayton Key Mode (Laboratory)
Idle
Low Speed
High Speed
Clayton Key Mode (Garage)
Idle
High Speed
Federal Three -Mode (Laboratory)
Idle
Low Speed
High Speed
Federal Three -Mode (Garage)
Idle
Low Speed
High Speed
Z500 rpm Unloaded (Laboratory)
2500 rpm Unloaded (Garage)
Parameter, %
E
o
11
7
7
35
30
10
21
17
17
11
18
22
20
38
37
FF
56
60
61
32
37
57
45
52
38
51
46
44
47
28
30
4-60
-------
summarize data from the figures at HC FTP levels of 0.41 and 0. 90. On the
average, at both FTP levels, the bag tests (Federal Short Cycle and NY/NJ
Composite) have lower EQ and higher FF at the fixed E = 5 percent than do
the volumetric tests. However, the idle mode of the Clayton Key Mode (with
either laboratory or garage instruments) test produces similar results. The
2500 rpm Unloaded test is very poor on a comparative basis.
4.2.2.5.1.2 Carbon Monoxide Emissions
The variation of E , E ,
and FF as a function of CO cut-
point are displayed in Figures 4-36 through 4-43 for each ST examined, and
for the range of CO FTP values selected in Table 4-10 (CO = 3.4 to 9.0).
The figures correspond to the following ST/FTP-level spectrum:
Short Test
Federal Short Cycle
NY/NJ Composite
Clayton Key Mode
(Laboratory)
Clayton Key Mode
(Garage)
Federal Three-Mode
(Laboratory)
Federal Three -Mode
(Garage)
2500 rpm Unloaded
(Laboratory)
2500 rpm Unloaded
(Garage)
CO FTP Level
3.4
X
X
X
X
X
X
X
X
5.0
X
X
X
7.0
X
X
X
9.0
X
X
X
X
X
X
X
X
Figure No.
4-36
4-37
4-38
4-39
4-40
4-41
4-42
4-43
4-61
-------
30
25
FF
i
o*
ro
20
u" 15
LU
O
o. 10
X
\
*
*
m
1975 FTP
LEVEL
-7.0
-9.0
AT 5.0. 7.0. AND 9.0
FF<5%
CO CUT-POINT, gm/mi
Figure 4-36. Variation of Ec, Eo, and FF with CO Cut-Point;
CEV Fleet; Federal Short Cycle; Bounded Errors of Commis-
sion Method
-------
i
<^
00
30 -
EQ AND FF < 10% AT LEVELS II, III, AND IV
CO CUT-POINT, ppm
Figure 4-37. Variation of Ec, E0, and FF with CO Cut-Point;
CEV Fleet; 1975 FTP CO Level = 3.4 gm/mi; NY/NJ Com-
posite Test; Bounded Errors of Commission Method
-------
501
1975 FTP LEVEL =3.4 AT 3.4 FF < 10% FOR HIGH SPEED AND LOW SPEED
1975 FTP- 9.0 AT9.0FF<1%
40
30
o
20
UJ
O-
FF
H HIGH SPEED
L LOW SPEED
- IDLE
300
400
500
600 700 800
CO CUT-POINT, ppm
900
1000
1100
1200
Figure 4-38. Variation of Ec, Eo, and FF with CO Cut-Point;
CEV Fleet; Key Mode Test; Bounded Errors of Commission
Method
-------
1975 FTP LEVEL - 3.4
EQ AND FF < 1% AT 1975 FTP
9.0
50 r~
40
o
-------
«^-*"
^ L _
HIGH SPEED
LOW SPEED
IDLE
400 600
800 1000 1200
CO CUT-POINT, ppm
1400 1600 1800 2000
Figure 4-40. Variation of Ec, E0, and FF with CO Cut-Point;
CEV Fleet; 1975 FTP CO Level = 3.4 gm/mi; Federal Three-
Mode Test; Bounded Errors of Commission Method
-------
50
40
. 30
o
I
o
-J
UJ
UJ
o
Ec
Eo
FF
H HIGH SPEED
L = LOW SPEED
I IDLE
H
600 650 700
CO CUT-POINT, ppm
Figure 4-41. Variation of Ec, Eo, and FF with CO Cut-Point;
CEV Fleet; 1975 FTP Level = 3. 4 gm/mi; Federal Three-
Mode Test; Garage Instruments; Bounded Errors of Com-
mission Method
-------
AT 5.0, 7.0 AND 9.0FF<3%
E
00
50
40
o
< 30
"b
LU
UJ"
£ 20
3.4
^*"
5.0
1975 FTP LEVEL
500
600 700
CO CUT-POINT, ppm
800
900
Figure 4-42. Variation of Ec, Eo, and FF with CO Cut-Point;
CEV Fleet; Unloaded 2500 rpm Test; Bounded Errors of Com-
mission Method
1000
-------
1975 FTP LEVEL
FF < 5%
3.4
50
40
-o30
20
10
EQ AND FF< 5% AT 1975 FTP - 9.0
Ec
Eo
FF
750 800 850
CO CUT-POINT, ppm
900
950
1000
Figure 4-43. Variation of Ec, E0, and FF with CO Cut-Point;
CEV Fleet; Unloaded 2500 rpm Test; Garage Instruments;
Bounded Errors of Commission Method
-------
As in the preceding case of hydrocarbon emissions, these
figures indicate the tradeoffs possible between E , E , and FF. However,
for CO FTP levels above 3.4, the general or average CO levels of the CEV
fleet were sufficiently low, i.e., a very high percentage of the vehicles
were better than the 5.0, 7.0, and 9.0 gm/mi requirements, that both
E and FF percentage values were very small for all of the short test pro-
cedures. This characteristic is summarized in Table 4-17 for the CO FTP
level of 9.0 gm/mi.
At the 3.4 level, however, as shown in Table 4-18, the bag
tests were sufficiently discriminatory to identify FF values above 20 percent,
with E values in the 14- to 16-percent range. The volumetric tests, on
the other hand, all had high E values (30- to 40-percent range) with very
low FF values (< 16).
4.2.2.5.1.3 Oxides of Nitrogen Emissions
The variations of E , E , and FF as a function of NO cut-
O C X
point are displayed in Figures 4-44 through 4-48 for each ST examined, for
the single NO FTP values of 3. 1 gm/mi examined in the study.
The significant results at the E level of 5 percent are sum-
marized in Table 4-19 for comparative purposes. As shown, the high-speed
mode of the volumetric tests (Clayton Key Mode and Federal Three-Mode)
produced the highest FF values and the lowest E values, and are thus indi-
cated to be superior for NO discrimination purposes.
4.2.2.5.1.4 Weighted Three-Mode Tests
In addition, a bounded errors analysis was made for two
weighted Key Mode tests. The first weighting factors were based on the
multiple regression analysis of Sec. 4.2.1.2. The second weighting factors
are suggested by Clayton Manufacturing Co. These latter factors were
developed for HC and CO, based on 1972 surveillance data. The weighting
factors are given in Table 4-20. The analysis was performed at FTP level I
"Exhibit G, Short Tests Versus 1975 CVS Relatability Analysis/Correlation/
Errors of Commission and Omission," May 1973
4-70
-------
Table 4-17. Comparison of Selected ST Carbon Monoxide Results:
CEV Fleet, Bounded Errors of Commission Analysis,
CO FTP Level =9-0 gm/mi (E = constant = 5%)
Short Test
Parameter, %
FF
Federal Short Cycle
NY/NJ Composite
Clayton Key Mode (Laboratory)
Idle
Low Speed
High Speed
Clayton Key Mode (Garage)
Idle
Low Speed
High Speed
Federal Three-Mode (Laboratory)
Idle
Low Speed
High Speed
Federal Three-Mode (Garage)
Idle
Low Speed
High Speed
2500 rpm Unloaded (Laboratory)
2500 rpm Unloaded (Garage)
< 1
< 1
< 1
< 1
< 1
< 1
< 1
4-71
-------
Table 4-18. Comparison of Selected ST Carbon Monoxide Results:
CEV Fleet, Bounded Errors of Commission Analysis,
CO FTP Level = 3.4 gm/mi (E = constant = 5%)
Short Test
Federal Short Cycle
NY/NJ Composite
Clayton Key Mode (Laboratory)
Idle
Low Speed
High Speed
Clayton Key Mode (Garage)
Idle
Low Speed
High Speed
Federal Three -Mode (Laboratory)
Idle
Low Speed
High Speed
Federal Three -Mode (Garage)
Idle
Low Speed
High Speed
2500 rpm Unloaded (Laboratory)
2500 rpm Unloaded (Garage)
Parameter, %
Eo
14
16
29
33
36
29
36
33
33
40
43
29
36
36
35
38
FF
22
20
15
11
8
14
7
10
16
8
7
14
7
7.5
8
6
4-72
-------
I
-4
Ul
I
UJ
o
UJ
Q_
25
20
< 15
. o
10
0
FF<5.4%
I
5 6
NOX CUT-POINT, gm/mi
J
7
Figure 4-44. Variation of Ec, Eo, and FF with NOX Cut-Point;
CEV Fleet; Federal Short Cycle Test; Bounded Errors of Com-
mission Method
-------
20 r-
I
-J
NO CUT-POINT, ppm
A
Figure 4-45. Variation of Ec, Eo, and FF with NOX Cut-Point;
CEV Fleet; NY/NJ Composite Test; Bounded Errors of Com-
mission Method
-------
25 i-
20
i
Ul
o 15
j
"L>
10
ON LOW SPEED FF < 2%
ON IDLE FF< 2.5%
EQ
FF
H = HIGH SPEED
i = LOW SPEED
I = IDLE
200 400 600 800 1000 1200 1400 1600 1800
NOY CUT-POINT, ppm
A
2000
2200
Figure 4-46. Variation of Ec, Eo, and FF with NOX Cut-Point;
CEV Fleet; Key Mode Test; Bounded Errors of Commission
Method
-------
I
-o
u. 30
u_
a
* 20
o
UJ
o
PERCENT E
=> 0
-
1
t '
V I i i ''
0 400 800 1200
Ec
Eo
FF
1
H HIGH SPEED
L LOW SPEED
I = IDLE
I
1600 2000 2400 2800 3200 3600 4000
NO CUT-POINT, ppm
A
Figure 4-47. Variation of Ec, EO) and FF with NOX Cut-Point;
CEV Fleet; Federal Three-Mode Test; Bounded Errors of
Commission Method
-------
25
20
It
Q
-ZL ...
< 15
-J
-4
10
UJ
FF < 2%
FF
800
NO CUT-POINT, ppm
A
900
1000
Figure 4-48. Variation of Ec, E0, and FF with NOX Cut-Point;
CEV Fleet; Unloaded 2500 rpm Test; Bounded Errors of Com-
mission Method
-------
Table 4-19. Comparison of Selected ST NOX Results:
CEV Fleet, Bounded Errors of Commission Analysis,
NO FTP Level =3.1 gm/mi (E = constant = 5%)
X C
Short Test
Federal Short Cycle
NY/NJ Composite
Clayton Key Mode (Laboratory)**'
Idle
Low Speed
High Speed
Federal Three -Mode (Laboratory)*31'
Idle
Low Speed
High Speed
2500 rpm Unloaded (Laboratory)'
Parameter, %
Eo
9.5
10
13
14
6.5
11
11
3
13
FF
5.5
5
2
<2
8.5
1
1
8.5
2
(a) Garage-type analyzers for NO were not available for ST
evaluation.
4-78
-------
Table 4-20. Key Mode Weighting Factors
Origin
Regression Analysis
Clayton Report
Mode
High
Low
Idle
Constant
High
Low
Idle
Constant
Weights
HC
0.00025
0.00017
0.00174
0. 154
0.8736
0.8736
0.312
CO
6.67
- 1.79
72.65
2.07
0.66
0.66
0.33
NO
X
0.00116
0.00024
0.00204
0.929
only (see Table 4-10). The results are depicted in Figures 4-49 to 4-51.
They clearly illustrate that the weighted volumetric tests are not signifi-
cantly better than the best single mode.
4.2.2.5.1.5 Variance Estimates
As the plots in Figures 4-26 to 4-48 are predictions from the
data, the variability of these predictions should be addressed. Referring to
Figures 4-7 and 4-9, the problem of estimating the ST cut-point, for a fixed
FTP level, is analogous to estimating the quantiles of a distribution function
(Ref. 4-5). Thus, the large sample standard deviation is given by
PF
US)
dy
N
(4-1)
y =
4-79
-------
o
(XL
i
oo
o
0
0
W, WEIGHTING FACTORS BASED
1 ON REGRESSION COEFFICIENTS
CLAYTON FACTORS
H - HIGH SPEED
L = LOW SPEED
I - IDLE
10
20
30
40
PERCENT E
50
60
70
80
Figure 4-49. Variation of Ec and Eo for Key Mode and Weighted
Key Mode Tests; CEV Fleet; 1975 FTP HC Level = 0.41 gm/mi;
Bounded Errors of Commission Method
-------
I L H
oo
0
w.
W, = WEIGHTING FACTORS BASED
1 ON REGRESSION COEFFICIENTS
W - CLAYTON FACTORS
H = HIGH SPEED
L = LOW SPEED
I - IDLE
0
10
20
30
PERCENT E.
40
50
60
Figure 4-50. Variation of Eo and Ec for Key Mode and Weighted
Key Mode Tests; CEV Fleet; 1975 FTP CO Level = 3.4 gm/mi;
Bounded Errors of Commission Method
-------
H
00
ro
0
0
W, - WEIGHTING FACTORS BASED
1 ON REGRESSION COEFFICIENTS
H - HIGH SPEED
L LOW SPEED
I IDLE
10 15
PERCENT E.
20
25
Figure 4-51. Variation of Eo and Ec for Key Mode and Weighted
Key Mode Tests; CEV Fleet; 1975 FTP NOX Level = 3. 1 gm/mi;
Bounded Errors of Commission Method
-------
where
PF = probability of passing the FTP
P(y) = probability of an error of commission for ST
cut-point set at value y
y = upper bound on probability of errors of commission
N = sample size or number of cars in the data set
LS = true cut-point for the population.
As PF, p(y), and LSQ are unknown, they can only be approximated from the
data. LSQ is, of course, approximated by the cut-point estimated from the
data. PF is estimated on the percent passed by FTP divided by 100. p(y) is
taken to be the locus of E versus cut-point. |dp(y)/dy|T c is taken as the
c I l1-1^©
derivative of the E versus cut-point curve evaluated at the cut-point of
interest (LS J. Equation (4-1) will be used to discuss variablility of the pre-
dicted population.
For a fixed FTP level, the standard deviation of the estimated
cut-point can be independently controlled by increasing the sample size.
Once the sample size is fixed, this standard deviation varies inversely with
the magnitude of the derivative of the E versus cut-point curve. Thus, in
regions where the curve is steep, the variability of the predictions will be
less than in regions where the curve is flat.
For example, at FTP level for HC = 0.41 and Federal Short
Cycle cut-point of 0.4, Y * 0.045, N = 39, PF = 0.33, and
dp(y)
dy
oc 0.25
0.41
Thus, Eq. (4-1) gives the approximate standard deviation of 0.044 gm/mi.
4-83
-------
For the Federal Short Cycle point of 0.75,
dp(y)
dy
0.75
asO.033 , -ySO.003
and the standard deviation increases to 0.086 gm/mi. Figure 4-52 illustrates
the effect of the cut-point uncertainty on the other computed quantities of EQ
and FF. It shows that the uncertainty in the predicted results increases with
decreasing errors of commission bounds.
4.2.2.5.2 Multiple-Constituent Tests
In addition to analyzing each pollutant individually, an analysis
was made for multiple-constituent tests, using the contingency table approach.
In a three-constituent test, a car fails the ST if any of its HC, CO, and NOx
measurements exceed the previously determined cut-points. These tests are
applicable to the bag tests, the unloaded test, and the individual modes of the
three-mode volumetric tests. Nine constituent tests are applicable only to
the three-mode volumetric tests. A car fails the ST if any one of the modes
fails on its three-constituent tests. Only data analytic and parametric actual
results were computed. A model for predicting population results was not
available.
Shown in Figure 4-53 is the computational procedure followed
in determining the multiple constituent tests. Note that the cut-point selec-
tion policy is applied at the pollutant level and not at the multiple-constituent
test level. For example, the percent E is bounded for individual pollutants
in the method of bounded errors of commission, and this bound can possibly
be exceeded on a multiple-constituent test. In forming the multiple-constit-
uent test contingency table, the following definitions apply:
Correctly passed (PP): Car passes the ST and the FTP
Correctly failed (FF): Car fails the ST and the FTP
Error of Commission (E ): Car fails the ST and passes the FTP
Error of Omission (E ): Car passes the ST and fails the FTP
4-84
-------
70
60
50
o40
UJ
£ 30
LU
a.
10
0
0
T
1
ERROR BAR INDICATES PLUS AND MINUS ONE
STANDARD DEVIATION OF THE ESTIMATE
FF
vk A
234567
HC FEDERAL SHORT CYCLE, gm/mi
Figure 4-52. Variability of Predicted Population Results
4-85
-------
START
Determine ST
cut-points for
HC, CO, and NO,
Write Results of
Contingency Table
Analysis for
HC, CO, and
Determine
3-Constituent
Test Results,
Write Results
YES
Determine
9-Constituent
Test Results,
Write Results
Set
Policy
NO
-FINISH
FINISH
Figure 4-53. Computation Flow Chart
4-86
-------
where FTP or ST failure occurs if any one of the test constituents exceeds
its respective cut-points. A car is counted once in forming the table and
falls into one, and only one, of the above categories. The percent E on a
multiple-constituent test may be larger than the largest individual pollutant
percent EC, or may be smaller than the smallest individual pollutant percent
EC, depending upon the actual data set and its particular mix of pollutant
failures. A useful observation, using the individual pollutant results, of the
actual percent E for the multiple constituent test is
c
Percent E =* (maximum pollutant ST percent FF)
X (minimum pollutant FTP percent PP)
Other useful relations for the multiple constituent tests are:
Percent FF > max (pollutant percent FF)
Percent PP < min (pollutant percent PP)
The three-constituent test results for the Federal Short Cycle
and the Federal Three-Mode (high-speed and idle modes only) are summarized
in Figures 4-54 through 4-65. The data plotted are the parametric population
results. Both the laboratory and garage instrument results are displayed for
the Federal Three-Mode short test (Figures 4-58 through 4-65).
The data were generated in the following manner: The method
of bounded errors of commission was used to determine an ST pass/fail cut-
point for each pollutant individually. The three-constituent test results were
obtained by simultaneously comparing the observed emission levels for HC,
CO, and NO of a vehicle against the determined ST and the given FTP cut-
points .
4-87
-------
LU o
u 8
at
OK>O
(
) <
> (
) (
_ <
)
1 2 3 4 5
oo
00
801-
60
40
20
401-
30
PREDICTED PERCENT E.
"i 20
o
10
FF
PREDICTED PERCENT E
Figure 4-54. Variation of Actual Ec, E0,
and FF with Predicted Ec; Federal Short
Cycle; Three-Constituent Test; Bounded
Errors of Commission Method; CEV Fleet;
1975 FTP Level I
Figure 4-55. Variation of Actual Ec, Eo,
and FF with Predicted Ec; Federal Short
Cycle; Three-Constituent Test; Bounded
Errors of Commission Method; CEV Fleet;
1975 FTP Level II
-------
O
O£
0 0
o
o
(J
LU
I 8
o
on
LLJ
Q_
i 4
«
r
IJ
c
>
5 <
) (
) (
) (
12345
I
oc
401-
30
20
10
o
FF
401-
30
20
10
I
2 3
PREDICTED PERCENT E
1 2 3
PREDICTED PERCENT E
Figure 4-56. Variation of Actual Ec, Eo,
and FF with Predicted Ec; Federal Short
Cycle; Three-Constituent Test; Bounded
Errors of Commission Method; CEV Fleet;
1975 FTP Level III
Figure 4-57. Variation of Actual Ec, Eo,
and FF with Predicted Ec; Federal Short
Cycle; Three-Constituent Test; Bounded
Errors of Commission Method; CEV Fleet;
1975 FTP Level IV
-------
rfk
o
ACTUAL PERCENT E. 0 FOR IDLE MODE
o
et
c
0 HIGH SPEED
<
i (
> (
> (
) (
> <
4 -
0' =t"^
80
60
40
20
3 4
FF
OHIGH SPEED
AIDLE
HIGH SPEED
AIDLE
PREDICTED PERCENT EC
Figure 4-58. Variation of Actual Ec, Eo,
and FF with Predicted Ec; Federal Three-
Mode; Laboratory Instruments; Three-
Constituent Test; Bounded Errors of Com-
mission Method; CEV Fleet; 1975 FTP
Level I
O HIGH SPEED
A IDLE
RANGE OF VARIATION:
- HIGH SPEED MODE
- IDLE MODE
101-
o
9
60
40
20
OHIGH SPEED
AIDLf
FF
HIGH SPEED
AIDlf
PREDICTED PERCENT E
Figure 4-59. Variation of Actual Ec> Eo,
and FF with Predicted Ec; Federal Three-
Mode; Laboratory Instruments; Three-
Constituent Test; Bounded Errors of Com-
mission Method; CEV Fleet; 1975 FTP
Level II
-------
O HIGH SPEED
A IDLE
20
10
0 0
RANGE OF VARIATION:
HIGH SPEED MODE
~ IDLE MODE
O
0 C
1'
O
40
30
20
10
OHIGH SPEED
AIDLE
FF
HIGH SPEED
AIDLE
IDLE
I
HIGH
I
I
1 2 3
PREDICTED PERCENT
Figure 4-60. Variation of Actual Ec, Eo,
and FF with Predicted EC; Federal Three-
Mode; Laboratory Instruments; Three-
Constituent Test; Bounded Errors of Com-
mission Method; CEV Fleet; 1975 FTP
Level III
LU
o
O
-------
*.
sO
O HIGH SPEED
A IDLE
I
o:
o
10 -
5
80
60
o
20
RANGE OF VARIATION:
HIGH SPEED MODE
- IDLE MODE
-
(
> (
(
> ,
i
7
(.
)
*
r /
i t
t j
i
01234
O HIGH SPEED
AI OLE
FF
HIGH SPEED
A IDLE
PREDICTED PERCENT E_
o
u
20
10
40
t
cj
30
20
10
O HIGH SPEED
A IDLE
RANGE OF VARIATION:
- HIGH SPEED MODE
- IDLE MODE
toc
i
4 5
FF
OHIGH SPEED
AI OLE
HIGH SPEED
IDLE
^"TA
^ * \
***" IDLE
I
1234
PREDICTED PERCENT E.
Figure 4-62. Variation of Actual Ec, E0,
and FF with Predicted Ec; Federal Three-
Mode; Garage Instruments; Three-
Constituent Test; Bounded Errors of Com-
mission Method; CEV Fleet; 1975 FTP
Level I
Figure 4-63. Variation of Actual Ec, Eo>
and FF with Predicted Ec; Federal Three-
Mode; Garage Instruments; Three-
Constituent Test; Bounded Errors of Com-
mission Method; CEV Fleet; 1975 FTP
Level n
-------
o
ec.
o
10
ACTUAL PERCENT L 0 FOR IDLE MODE
c
OHIGH SPEED MODE
O
401
---- FF
HIGH SPEED MODE
IDLE MODE
1 2 3
PREDICTED PERCENT E,
Figure 4-64. Variation of Actual Ec, Eo,
and FF with Predicted EC; Federal Three-
Mode; Garage Instruments; Three-
Constituent Test; Bounded Errors of Com-
mission Method; CEV Fleet; 1975 FTP
Level III
10
O HIGH SPEED
A IDLE
RANGE OF VARIATION:
HIGH SPEED MODE
~ IDLE MODE
ft
40
30
20
10
O HIGH SPEED
AIDLE
FF
HIGH SPEED
AIDLE
PREDICTED PERCENT E.
Figure 4-65. Variation of Actual Ec, Eo,
and FF with Predicted Ec; Federal Three-
Mode; Garage Instruments; Three-
Constituent Test; Bounded Errors of Com-
mission Method; CEV Fleet; 1975 FTP
Level IV
-------
The results in the figures are presented as follows: for each
three-constituent test, the actual errors of omission, actual correct failures,
and actual errors of commission are plotted versus the predicted error of
commission that was the original bound on the individual pollutants. For
example, in Figure 4-54, if policy was set at a maximum of 2 percent errors
of commission on each individual pollutant, the actual results of a three-
constituent test are about 30 percent errors of omission, 36 percent correct
failures, and 2.6 percent errors of commission. For the actual errors of
commission, plus or minus one standard error of the estimate is indicated
by an error bar, with the value of the estimate in the center of the bar. In
the above example, plus one standard error of the estimate gives about
5 percent E , while the minus side shows about 0. 1 percent E , with the
observed value being 2.6 percent E .
4.2.2.5.2.2 Variance Estimates
For fixed ST and FTP cut-points, the cell counts in a 2 X 2
contingency table are binomially distributed when the observations are inde-
pendent (Ref. 4-6). Since the ST cut-points are computed from data prior to
forming the contingency table, there is statistical dependence between the ST
cut-points and the resulting table. Hence, the binomial distribution will be
an approximation to the true distribution. Thus, the approximate standard
deviation is
X (100 - X
N
where
X = cell count in percent
N = total table count
4-94
-------
For example, if the percent errors of omission is computed to be 50 percent,
then, with 40 cars, the standard deviation is 7.9 percent. Table 4-21 shows
the approximate cell standard deviations for the range of cell percentages
assuming N = 40. This procedure was also used to calculate the standard
error of the estimate depicted in Figures 4-54 through 4-65 by the error
bars on the actual errors of commission.
4.2.2.5.2.3 Discussion of Results
As the FTP cut-points increase from level Set I to level Set IV,
the resulting actual errors of commission tend to increase for a given pre-
dicted level of errors of commission. For example, on the Federal Short
Cycle at 2 percent predicted errors of commission, the actual errors are
2.5 percent, 5 percent, 5 percent, and 7.5 percent for FTP levels I, II, III,
and IV. respectively. This trend is not present for the garage instrument
results as shown in Figures 4-62 through 4-65.
Table 4-21. Approximate Standard Deviation
for Three-Constituent Tests -
CEV Fleet, N = 40
Cell
Percentage
60
50
40
30
20
12.5(a)
10
5
-------
A comparison of the modes on the Federal Three-Mode test
shows that, for a fixed predicted percent EC, the high speed mode has a higher
percent FF and Lower percent E than does the idle mode. This is true
regardless of instrumentation or FTP level. However, the actual percent E
is generally lower on the idle mode than on the high-speed mode, but this
difference is not always significant.
A comparison of different modes or ST should be made on a
fixed actual percent E basis. This is, of course, difficult to do because of
the computational procedure followed. It can be approximately performed,
however. Consider comparing the Federal Short Cycle to the Federal Three-
Mode. At FTP level I, the actual percentages of E are approximately the
same for the high speed mode and the Federal Short Cycle (statistically, they
are equivalent). Now, comparing the percent FF and percent E curves,
percent FF and percent E are both higher on the high-speed mode than the
Federal Short Cycle. This difference is not statistically significant at the
95 percent level, and the two tests would have to be judged as equal. Also,
at the 95 percent level, the high-speed mode is superior to the idle mode.
The differences between laboratory and garage instruments
are quite predictable, based upon the previous results from individual pollu-
tants. For a fixed predicted percent E , on their respective modes,
a. Actual percent E is higher for garage instruments than
for laboratory instruments
b. Actual percent FF is lower for garage instruments than
for laboratory instruments
c. Actual percent E is higher for garage instruments than
for laboratory instruments.
4-96
-------
4.3 REFERENCES FOR SECTION 4
The following references are used in Section 4:
4-1. W. J. Dixon, ed. , BMPD Biomedical Computer Program, University
of California Press, Berkeley (1974).
4-2. D. F. Morrison, Multivariate Statistical Methods, McGraw-Hill
Book Co., Inc., New York (1967).
4-3. F. A. Graybill, An Introduction to Linear Statistical Models, Vol. 1,
McGraw-Hill Book Col, Inc., New York (1961).
4-4. T. W. Anderson, Multivariate Statistical Analysis, John Wiley and
Sons, Inc., New York (1958).
4-5. H. Cramer, Mathematical Methods of Statistics, Princeton University
Press, New Jersey (1971).
4-6. C. R. Rao, Linear Statistical Inference and Its Applications, John Wiley
and Sons, Inc., New York (1965).
4-97
-------
5. IN-USE 1974 MODEL YEAR
VEHICLE FLEET
-------
5. IN-USE 1974 MODEL YEAR VEHICLE FLEET
This section summarizes the results of statistical analyses
conducted to determine the degree of correlation existing between the vari-
ous short tests (STs) and FTP tests conducted on a fleet of in-use 1974
model year vehicles.
Several distinguishing features of the 1974 model year fleet
resulted in variations in focus and scope of the statistical analyses from
those reported for the CEV fleet in Section 4. They include:
a. The 1974 model year fleet was manufactured to known
emission standard values, whereas the CEV fleet was not.
b. The 1974 model year fleet population was stratified by
three inertia test weight groups, whereas the CEV fleet
was at a. single inertia test weight value.
c. There was no substantial number of replicate test obser-
vations for the 1974 model year fleet.
The appropriate 1975 FTP emission standards for the 1974
model year fleet were computed to be:
HC = 3.02 gm/mi
CO = 28.0 gm/mi
NO =3.1 gm/mi
The three inertia test weight groups were designated as:
Group A (4000-Ib class)
Group B (2750-Ib class)
Group C (5500-Ib class)
For analysis purposes, laboratory instrument test data were
available for 147 cars, while garage-type instrument data were available for
144 cars. These test data had been processed by EPA and were received
stored on magnetic tape. Correlation analysis results are summarized in
Sec. 5. 1; the contingency table analysis results are summarized in Sec. 5.2.
5-1
-------
5. 1 CORRELATION ANALYSIS RESULTS
A conventional correlation analysis was made for the 1974
model year fleet. The method was as described in Sec. 4.2. 1. 1. The re-
sulting ST/FTP correlation coefficients are summarized in Table 5-1 for
the individual inertia test weight groups (A, B, C) and for the pooled vehi-
cle population (combined groups A, B, C). For N = 147 cars, a computed
correlation coefficient greater than 0. 16 indicates that the ST and FTP pol-
lutants are statistically correlated with 95% confidence. For N = 48 to 50,
this threshold is approximately 0.29.
In addition, a correlation analysis of FTP composite emis-
sions versus FTP bags 2 and 3 was made (by the method outlined in
Sec. 4.2. 1. 1. 1). The results are shown in Table 5-2.
As can be seen in Table 5-1, no single ST performs con-
sistently well on all three individual groups, or on a pooled basis. Gen-
erally, the STs are unable to track HC and CO emission levels on Group C
(5500-lb Chevrolet vehicles). This is also supported by the FTP composite
versus bags 2 plus 3 correlations of Table 5-2. The low correlation for
NO in Group C in Table 5-2 is the result of a single outlying point and,
thus, does reflect an usually low relatability. However, the HC and CO
correlations for Group C are significantly different (in the sense of a rigor-
ous statistical test) than those of Groups A and B. This would indicate that
"hot" procedures would not perform as well on Group C as on Groups A
and B.
The presence of one ST with good NO correlation across the
Ji
population is missing in the 1974 model year fleet. From a correlation view-
point, the garage analyzers are inferior to the laboratory analyzers. ST
ratings using the scale established in Sec. 4. 2. 1. 4. 4 for the CEV fleet are
given in Table 5-3. As with the CEV fleet (Table 4-8), the bag-type STs
have higher ratings than the volumetric tests. The 2500 rpm Unloaded test
shows substantially higher correlation for the 1974 model year fleet than for
the CEV fleet. The extreme CO tracking deficiency for the CEV fleet data
is not evident for the 1974 model year fleet.
5-2
-------
Table 5-1. Correlation Coefficient Summary:
1974 Model Year Fleet
Short Test
Federal
Short
Cycle
NY/NJ
Composite
Key Mode
(Laboratory)
Key Mode
(Garage)
Vehicle
Group *a)
Pooled
A
B
C
Pooled
A
B
C
Pooled
A
B
C
Pooled
A
Test
Mode
High
Low
Idle
High
Low
Idle
High
Low
Idle
High
Low
Idle
High
Low
Idle
High
Low
Idle
N(b)
147
50
48
49
147
50
48
49
147
50
48
49
145
50
ST/FTP Correlation
Coefficient*0'
HC
0.932
0.933
0.897
0.383
0.906
0.911
0.920
0.513
0.757
0.776
0.793
0.590
0.595
0.723
0.812
0.868
0.825
0.238*
0.228*
0.460
0.528
0.545
0.455
0.228*
0. 151*
0.245*
CO
0.905
0. 972
0.897
0.476
0.890
0.950
0.857
0.498
0.518
0.769
0.739
0.514
0.827
0.704
0.262*
0.738
0.650
-0. 195*
0.435
0.757
0.507
0.472
0.470
0.563
0.652
0.372
NO
X
0.355
0.780
0. 104*
0.674
0.060*
0.733
0.005*
0.611
0. 521
0.419
0.463
0.562
0.495
0.381
0.731
0.635
0.548
0.555
0. 580
0. 571
5-3
-------
Table 5-1. Correlation Coefficient Summary:
1974 Model Year Fleet (Continued)
Short Test
Federal
Three -Mode
(Laboratory)
Federal
Three -Mode
(Garage)
Vehicle
Group'*'
B
C
Pooled
A
B
C
Pooled
A
Test
Mode
High
Low
Idle
High
Low
Idle
High
Low
Idle
High
Low
Idle
High
Low
Idle
High
Low
Idle
High
Low
Idle
High
Low
Idle
N(b)
46
49
147
50
48
49
145
50
ST/FTP Correlation
Coefficient^)
HC
0.478
0.765
0.692
0.191*
0. 198*
0. 100*
0.766
0.771
0.803
0.507
0.523
0.709
0.890
0.859
0.851
0.522
0.533
0.252*
0.474
0.531
0.632
0.138*
0.107*
0.660
CO
0.362
0.540
0.560
-0.221*
-0.091*
0.229*
0.604
0.729
0.734
0.717
0.801
0.724
0.278*
0.737
0.622
0.159*
0.592
0.733
0.387
0.409
0.476
0.533
0.597
0.397
NO
X
0.467
0.453
0.411
0.492
0.664
0.369
0.722
0.611
0.665
0.552
0.707
0.639
5-4
-------
Table 5-1. Correlation Coefficient Summary:
1974 Model Year Fleet (Continued)
Short Test
2500 rpm
Unloaded
(Laboratory)
2500 rpm
Unloaded
(Garage)
Vehicle
Group la>
B
C
Pooled
A
B
C
Pooled
A
B
C
Test
Mode
High
Low
Idle
High
Low
Idle
N(b)
46
49
147
50
48
49
147
50
46
49
ST/FTP Correlation
Coefficient(c)
HC
0.536
0.763
0.717
0.095*
-0.008*
-0.060*
0. 809
0.832
0.865
0. 107*
0. 574
0.487
0.781
-0.064*
CO
0.268*
0.539
0.550
-0.083*
0.239*
0.392
0.740
0.812
0.724
0.350
0.447
0.676
0.684
-0.051*
NO
X
0.447
0.524
0. 577
0.679
(b)
(c)
;A = Chrysler (4000 Ib)
B = Ford (2750 Ib)
C = Chevrolet (5500 Ib)
Pooled = Groups A + B + C
Number of cars in the data set
The correlations are statistically significant at the 95 percent confidence
level except where indicated by an asterisk. ST and FTP uncorrelated
for correlations below 0.28.
5-5
-------
Table 5-2. FTP Composite Versus Bag 2 + 3 Correlation
Coefficients: 1974 Model Year Fleet
Vehicle Group
Pooled
A
B
C
N(a)
147
50
48
49
FTP/FTP Bag 2 + 3
Correlation Coefficient^'
HC
0.992
0.987
0.998
0.965
CO
0.994
0.993
0.996
0.987
NOX
0.925
0.976
0.996
0.761
'
Number of cars in data set
The correlations are statistically significant at the 95% confidence level
5-6
-------
Table 5-3. ST Ratings: 1974 Model Year Fleet
Short Test
Federal Short Cycle
NY/NJ Composite
Key Mode
(Laboratory)
Key Mode
(Garage)
Federal Three -
Mode (Laboratory)
Federal Three -
Mode (Garage)
Vehicle
Group (a)
Pooled
A
B
C
Pooled
A
B
C
Pooled
A
B
C
Pooled
A
B
C
Pooled
A
B
C
Pooled
A
B
C
Ratings
HC
E
E
G
P
E
E
E
P
G (I)(c)
G (I)
G (L)
P (D
P (L)
U
G (L)
U
G (I)
G (I)
G (H)
P (L)
F (I)
F (I)
G (L)
U
CO
E
E
E
P
G
E
G
P
G (L)
G (L)
G (L)
G (I)
P (H)
F (L)
P (L)
U
G (I)
G (L)
G (I)
G (I)
P (D
P (D
P (D
P (D
N0x
P
G
U
F
U
G
U
F
P (H)
P (H)
G (H)
P (L)
P (H)
F (L)
G (H)
G (L)
5-7
-------
Table 5-3. ST Ratings: 1974 Model Year Fleet
(Continued)
Short Test
2500 rpm Unloaded
(Laboratory)
2500 rpm Unloaded
(Garage)
Vehicle
Group la'
Pooled
A
B
C
Pooled
A
B
C
Ratings (b)
HC
G
G
G
U
P
P
G
U
CO
G
G
G
P
P
F
F
U
NO
X
P
P
P
F
(b)
;A = Chrysler (4000 Ib)
B = Ford (2750 Ib)
C = Chevrolet (5500 Ib)
Pooled = Groups A + B + C
Rating scale as in Sec. 1. 1.1. 8
{C) I = idle
L = low speed mode
H = high speed mode
5-8
-------
5.2 CONTINGENCY TABLE ANALYSIS RESULTS
5. 2. 1 Maximum Correlation Method
Using the method as defined in Sec. 4. 2. 2. 1, a maximum
correlation analysis was made for the pooled sample population of the 1974
model year fleet. Table 5-4 summarizes the analysis results for the pre-
dicted population. For N = 147, a computed table correlation coefficient
greater than 0.16 indicates that the ST and FTP pollutants are statistically
correlated with 95% confidence.
Examination of these results indicates that the correlation
indices (Table 5-4) are quite similar to the relative ST ratings developed
in Table 5-3. NO tracking difficulty is indicated by a high percentage of
E relative to percent FF. CO is the dominant variable in that it has the
highest percent FTP failure rate. (For the CEV fleet, the dominant variable
was HC.)
5. 2. 2 Bounded Errors of Commission Method
A contingency table analysis for the 1974 model year fleet
was made using the methods described in Sec. 4. 2. 2. 1. 2 and 4. 2. 2. 2. For
this analysis the bound on percent E was varied from 5 percent to 1 percent
in 1 percent increments, with the values 0. 5 percent and 0. 1 percent included.
The results of the analysis are summarized below. The data shown are for
the predicted 1974 model year fleet population.
5.2.2.1 Single-Constituent Tests
5.2.2.1.1 Hydrocarbon Emission
The variations of E , E , and FF as a function of HC cut-point
o c
are shown in Figures 5-1 through 5-8 for each ST examined. The graphical
displays indicate the general nature of the tradeoffs available for policy
formulation. Reducing the errors of commission (Ec) increases the errors
of omission (E ) and decreases the correct failures (FF). To illustrate
specific values and trends among the STs, Table 5-5 summarizes data from
the figures for the E value of 5 percent.
5-9
-------
Table 5-4. Maximum Correlation Summary; 1974 Model Year
Fleet, Predicted Population
ST
Federal Short Cycle
NY/NJ Composite
Key Mode
(Laboratory)
Key Mode
(Garage)
Federal Three -Mode
(Laboratory)
Federal Three-Mode
(Garage)
2500-rpm Unloaded
I Laboratory)
2500-rpm Unloaded
(Garage)
N
147
147
147
144
147
144
147
144
Test Mode
High speed
Low speed
Idle
High speed
Low speed
Idle
High speed
Low speed
Idle
High speed
Low speed
Idle
Pollutant
HC
CO
NO
X
HC
CO
NO
X
HC
CO
NO
X
HC
CO
HC
CO
NO
X
HC
CO
HC
CO
HC
CO
HC
CO
NO
X
HC
CO
NO
X
HC
CO
NO
X
HC
CO
HC
CO
HC
CO
HC
CO
NO
X
HC
CO
% E
C
6.05
5.02
26.8
7. 20
5.37
38.56
12. 2
9.99
20.4
11.7
7.35
11. 17
7.74
22. 63
18. 1
10. 1
17. 4
10. 4
11 4
10.4
11.9
9.20
22. 5
11.8
7.86
23.0
6.80
7.80
24.6
18.9
11.1
17.7
10.«
12.2
10.3
10.7
7.72
23.3
16.5
10.6
% E
5.44
6.94
7.56
6.36
7. 60
8.77
10.0
20.3
6.77
9.64
12. 13
9.29
13. 16
7.06
13.9
20. 6
13. 5
21.7
14. 8
21.7
9. 83
17.5
7.04
9.74
13. 5
7. 11
4.08
13. 33
7.31
14.4
24.3
13.7
23.7
15. 4
21.4
8.94
13. 1
7. 14
12.9
22.3
% FF
35.4
65.5
10. 5
34.5
64.8
9.24
30.8
52. 1
11.24
31.2
60.3
31.5
59.3
10.9
27.2
51.7
27.6
50. 62
26.3
50.7
31.0
54.8
11.0
31. 1
58.9
10.1
28. 6
59. 1
10.70
26.7
48.0
27.41
48.7
28.9
50.9
31.9
59.3
10.9
28.2
50.0
Table
Correlation
Index* a>
0.763
0.708
0.201
0.720
0.684
0.033*
0. 544
0.330
0.310
0.563
0.641
0. 580
0. 511
0. 270
0.351
0.322
0.371
0.297
0.305
0. 298
0. 553
0. 394
0.273
0. 558
0.502
0.263
0.759
0. 506
0. 236
0.326
0. 239
0.363
0. 254
0. 437
0.303
0. 597
0.513
0.259
0.402
0.283
The correlation is statistically significant at the 45*o confidence level except where
indicated by an asterisk.
5-10
-------
30
25
t 20
a
o
OC
UJ
°- 10
\
*
'.HC
\
\
\
\/
)(
/ \
/HC
0
FF
NO,
0
HC AND NOX CUT-POINT,
Figure 5-1. Variation of Ec, Eo» and FF with HC
and NOX Cut-Point; 1974 Model Year Fleet; Fed-
eral Short Cycle Test; Bounded Errors of Com-
mission Method
5-11
-------
I
t-*
IS)
35
30
25
20
j°
"o
O
oc
15
10
0
FF
^ NO,
\
\
\
A
/
/HC
\
*\HC
100
125 150 175
HC AND NOX CUT-POINT, ppm
225
250
Figure 5-2. Variation of Ec, Eo, and FF with HC and NOX Cut-Point; 1974 Model
Year Fleet; NY/NJ Composite Test; Bounded Errors of Commission Method
-------
(Jl
I
(JO
30 -
25
20
o
Q£
LU
Q_
10
1500 2000 2500 3000 3500 4000
HC CUT-POINT, ppm
4500
H = HIGH SPEED
L = LOW SPEED
I - IDLE
5000
5500
Figure 5-3. Variation of Ec, Eo, and FF with HC Cut-Point;
1974 Model Year Fleet; Key Mode Test; Bounded Errors of
Commission Method
-------
Ol
I
40
30
u°
-., 20
O
10
~~
H
E
o
FF
H « HIGH SPEED
L - LOW SPEED
I - IDLE
60
100
120 140 160
HC CUT-POINT, ppm
Figure 5-4. Variation of Ec, Eo, and FF with HC Cut-Point;
1974 Model Year Fleet; Key Mode Test; Garage Instruments;
Bounded Errors of Commission Method
-------
35
30 -
25
< 20
. .o
o±
LU
D_
15
10
/
.'
1 V '
H / L >C /
\ ' \ / \ /
A
H7 \
V v
\ X * En
f \ / \ 0
\ X^ ^ f-r
*
\
«
\
*
\ Xs
\
H
1500 2000 2500 3000 3500 4000
HC CUT-POINT, ppm
4500
H = HIGH SPEED
L = LOW SPEED
I = IDLE
5000
5500
Figure 5-5. Variation of Ec, Eo, and FF with HC Cut-Point;
1974 Model Year Fleet; Federal Three-Mode Test; Bounded
Errors of Commission Method
-------
40
30
J°
J° 20
Ul
I
o
oc
LU
OU
10
H
o
FF
H - HIGH SPEED
L - LOW SPEED
I - IDLE
60
100
120 140 160
HC CUT-POINT, ppm
180
200
220
Figure 5-6. Variation of Ec, Eo, and FF with HC Cut-Point;
1974 Model Year Fleet; Federal Three-Mode Test; Garage
Instruments; Bounded Errors of Commission Method
-------
01
I
25
fc 20
1
L«J^
o15
LU
UJ
O
s 10
t
n
tc /
rL /
E /
0 y
FF HC /
\ / 40
\ /
X t 30
XN°x / \ 1
/ / \ _
/ / LU°
* f \ ULJ
1 ' \ - ?n
/ HC ^ upa
\ %
\ 0
\ § 10
\
\
NOX VHC X
\ x 0
~ /
/
_ /
/ c
E
w
FF
\
_ \
\
\A , , ,
NOA V 0 100 200 300 400
V ^^^^ HC CUT-POINT, ppm
V i i i i -^ i
"0 500 1000 1500 2000 2500 3000
HC AND NOX CUT-POINT, ppm
Figure 5-7. Variation of Ec, Eo, and FF with HC
and NOX Cut-Point; 1974 Model Year Fleet;
Unloaded 2500 rpm Test; Bounded Errors of Com-
mission Method
Figure 5-8. Variation of Ec, E0,
and FF with HC Cut-Point; 1974
Model Year Fleet; Unloaded
2500 rpm Test; Garage Instru-
ments; Bounded Errors of Com-
mission Method
-------
Table 5-5. Comparison of ST Hydrocarbon Results: 1974 Model
Year Fleet, Bounded Errors of Commission Analysis
(E = constant = 5%)
Short Test
Federal Short Cycle
NY/NJ Composite
Clayton Key Mode (Laboratory)
Idle
Low Speed
High Speed
Clayton Key Mode (Garage)
Idle
Low Speed
High Speed
Federal Three -Mode (Laboratory)
Idle
Low Speed
High Speed
Federal Three -Mode (Garage)
Idle
Low Speed
High Speed
2500 -rpm Unloaded (Laboratory)
2500 -rpm Unloaded (Garage)
Parameter, %
E
o
6.5
8.5
16
17
18
11.5
14
13
15.5
17.5
18
17
14
12
16
16
FF
34.5
32
24.5
23. 6
22.5
29
27
28
> 25
23
23
24
27
29
26
26
Figure No.
5-1
5-2
5-3
5-3
5-3
5-4
5-4
5-4
5-5
5-5
5-5
5-6
5-6
5-6
5-7
5-8
5-18
-------
The bag tests (Federal Short Cycle and NY/NJ Composite)
have lower EQ and higher FF at the fixed E = 5 percent condition than do
the volumetric tests. There is little difference shown between the various
volumetric STs.
5.2.2. 1.2 Carbon Monoxide Emission
The variations of E , E , and FF as a function of CO cut-
o c
point are shown in Figures 5-9 through 5-16 for each ST examined. As in the
preceding area of hydrocarbon emission, these figures indicate the possible
tradeoffs between EC, EQ, and FF. To illustrate specific values and trends
among the STs, Table 5-6 summarizes data from the figures for the E
value of 5 percent.
The bag-type STs (Federal Short Cycle and NY/NJ Composite)
exhibit excellent CO tracking characteristics; the E values are considerably
better (lower) than the volumetric tests, and the FF values are the highest.
When garage-type instruments are used, the E values are essentially doubled
(over laboratory instrument values) and FF values are significantly reduced.
5.2.2.1.3 Oxides of Nitrogen Emission
The variations of E , E and FF as a function of NO cut-point
O C 3C
are shown in Figures 5-1, 5-2, 5-7, 5-17, and 5-18 for each ST examined.
The significant results at the E level of 5 percent are summarized in Table
5-7 for comparative purposes. As can be noted, all STs identified very low
percentages of correctly failed vehicles (FF), <5 percent, while having
significant errors of omission, ~15 percent.
5.2.2.1.4 Variance Estimates
The general variance trends discussed in Sec. 4. 2. 2. 5. 1. 5
for the CEV fleet are also applicable in this case. However, the actual
magnitude of the standard deviation is different for the 1974 model year
fleet. For the example illustrated in Sec. 4. 2. 2. 5.1.5, the Federal Short
Cycle HC cut-point is 2. 6 gm/mi at \ = 0.045 (Figure 5-1). PF is 0.048 and
5-19
-------
I
M
O
65
60
55
u_
u.
H
8«
g50
0.
45
40
yj
r~ 30
25
o
UJ
-Q 20
*0
z
rAjl_ ^|J «t
ce
UJ
CL
10
5
n
v
*N
X /
%v /
- \ /
\
\ /
ft jf
\ /
A
/ \
mLrri j ^
\
\
_ / \
s'
s
s
x.^
1 1 l^^T 4-
W15 20 25 30 35 4t
0
FF
«.
\
\
45
50
CO CUT-POINT, gm/mi
Figure 5-9. Variation of Ec, Eo, and FF with CO Cut-Point;
1974 Model Year Fleet; Federal Short Cycle Test; Bounded
Errors of Commission Method
-------
351-
I
t\>
60 f~
3(
0
600
700 800 900 1000 1100 1200 1300 1400 1500
CO CUT-POINT, ppm
Figure 5-10. Variation of Ec, Eo, and FF with CO Cut-Point;
1974 Model Year Fleet; NY/NJ Composite Test; Bounded
Errors of Commission Method
-------
fc
Q
70 r-
60
50
40
20
10
H
\
\ /
>< /
X \
' \ \
\
E.
H - HIGH SPEED
L = LOW SPEED
I = IDLE
3000 10.000 15.000 20.000
25.000 30.000 35.000
CO CUT-POINT, ppm
40.000 45.000 50,000 55.000
Figure 5-11. Variation of Ec, Eo, and FF with CO Cut-Point;
1974 Model Year Fleet; Key Mode Test; Bounded Errors of
Commission Method
-------
UJ
60 -
50 -
40
30
20
10
H HIGH SPEED
L LOW SPEED
I IDLE
0
0.5
1.0
1.5
2.0 2.5
CO CUT- POINT, %
Figure 5-12. Variation of Ec, Eo, and FF with CO Cut-Point;
1974 Model Year Fleet; Key Mode Test; Garage Instruments;
Bounded Errors of Commission Method
4.5
-------
U1
70
60
50
40
g 30
o
20
10
Lx /
\ /
H ,X
\ / \ /
V v
A ,/\
* \ / '»
H y \
\
0
FF
H = HIGH SPEED
L - LOW SPEED
I IDLE
5000 10.000 15.000 20.000 25.000 30.000 35.000 40.000
CO CUT-POINT, ppm
45,000 50,000 55.000
Figure 5-13. Variation of Ec, E0. and FF with CO Cut-Point;
1974 Model Year Fleet; Federal Three-Mode Test; Bounded
Errors of Commission Method
-------
Ul
I
60
50-
40
o30
20
10
0
H'
L/
- "I L\
N
H HIGH SPEED
L - LOW SPEED
1 IDLE
o
FF
0 0.5 1.0 1.5 2.0 2.5 3.0
CO CUT-POINT, %
3.5
4.0
Figure 5-14. Variation of Ec, Eo, and FF with CO Cut-Point;
1974 Model Year Fleet; Federal Three-Mode Test; Garage
Instruments; Bounded Errors of Commission Method
4.5
-------
60
50
fc 40
171
i
ro
S
S 20
10
FF
6000
7000
8000
9000 10.000
CO CUT-POINT,
11.000
ppm
12.000 13.000 14.000 13,000
Figure 5-15. Variation of Ec, Eo> and FF with CO Cut-Point;
1974 Model Year Fleet; Unloaded 2500 rpm Test; Bounded
Errors of Commission Method
-------
01
I
o
oc
60 -
50
40
30
20
10
c
FF
0
0.5
0.6 0.7 0.8 0.9 1.0
CO CUT-POINT, %
1.1
1.2
1.3
Figure 5-16. Variation of Ec, Eo, and FF with CO Cut-Point;
1974 Model Year Fleet; Unloaded 2500 rpm Test; Garage
Instruments; Bounded Errors of Commission Method
-------
Table 5-6. Comparison of ST Carbon Monoxide Results: 1974
Model Year Fleet, Bounded Errors of Commission
Analysis (E = constant = 5%)
Short Test
Federal Short Cycle
NY/NJ Composite
Clayton Key Mode (Laboratory)
Idle
Low Speed
High Speed
Clayton Key Mode (Garage)
Idle
Low Speed
High Speed
Federal Three -Mode (Laboratory)
Idle
Low Speed
High Speed
Federal Three -Mode (Garage)
Idle
Low Speed
High Speed
2500 -rpm Unloaded (Laboratory)
2500 -rpm Unloaded (Garage)
Parameter, %
E
0
7
8
19
18
35
35
35
37
20
20
29
35
31
30
19
33
FF
65
64
53
54
38
38
38
35
53
52
43
37
41
42
53
40
Figure No.
5-9
5-10
5-11
5-11
5-11
5-12
5-12
5-12
5-13
5-13
5-13
5-14
5-14
5-14
5-15
5-16
5-28
-------
25 -
20
FF < 5% ON ALL MODES
15
ro
10
H HIGH SPEED
L LOW SPEED
I IDLE
H L
500 1000 1500 2000
NOX CUT-POINT, ppm
2500
3000
3500
Figure 5-17. Variation of Ec, Eo, and FF with NOX Cut-Point;
1974 Model Year Fleet; Key Mode Test; Bounded Errors of
Commission Method
-------
30
25
ht 20
o
u>
o
o
az
10
0
H
L
I
HIGH SPEED
LOW SPEED
IDLE
FF < 5% ON ALL MODES
500
1000
1500
2000
2500
3000
3500
NOX CUT-POINT, ppm
Figure 5-18. Variation of EC, Eo, and FF with NOX Cut-Point;
1974 Model Year Fleet; Federal Three-Mode Test; Bounded
Errors of Commission Method
-------
Table 5-7. Comparison of ST NOX Results: 1974 Model Year
Fleet, Bounded Errors of Commission Analysis
(E = constant = 5%)
Short Test
Federal Short Cycle
NY/NJ Composite
Clayton Key Mode (Laboratory)
Idle
Low Speed
High Speed
Federal Three -Mode (Laboratory)
Idle
Low Speed
High Speed
2500 -rpm Unloaded (Laboratory)
Parameter, %
E
0
14. 5
16.5
13.5
14
13.5
14
14
14
14
FF
3
1.5
<5
<5
<5
<5
<5
<5
4
Figure No.
5-1
5-2
5-17
5-17
5-17
5-18
5-18
5-18
5-7
5-31
-------
dp(y) *0.o2
*y|y=2.6
With N = 147, Eq. (4-1) yields a standard deviation of 0.35 gm/mi. At
v = 0.005, the standard deviation is about 0.95 gm/mi where the cut-point
is 3. 2 gm/mi. The standard deviation on CO (Figure 5-9) is estimated at
2. 4 gm/mi for the cut-point at 21. 5 gm/mi. For CO at v = 0.005, the cut-
point is 38 gm/mi and the standard deviation is approximately 4.9 gm/mi.
These estimates show that the standard deviation is on the
order of 10 percent to 15 percent of the estimated cut-point.
5.2.2.1.5 Instrument Comparisons
For comparing the instruments used in the test program, plots
of the type shown in Figures 5-19 through 5-32 are informative. Here per-
cent E and percent FF have been plotted against percent E for HC and CO
with each modal ST. If a policy decision is given in terms of percent E
allowable, then the percent FF and percent E can be compared. To illustrate, , ,
suppose percent E is fixed at 3 percent. For CO on the low speed Key Mode
(Figure 5-22), the laboratory instruments (dashed lines) give 48 percent FF
and 24 percent E , while the garage instruments give 26 percent FF and
46 percent E .
If policy is stipulated in terms of percent rejected by the ST,
then percent E can be compared. For the CO low-speed mode on the Key
Mode test, suppose ST percentage rejection is to be approximately 30 percent.
Then, for percent FF equal to 30 percent (percent E equals 42 percent),
percent E is 0.6 percent for the laboratory instruments and 3.9 percent
for the garage instruments.
5.2.2.1.6 Discussion of Results
On the average, the bag-type tests have lower E and higher
FF for a fixed rate of E than do the volumetric tests. However, FF rates
c
in the 30 percent range can be achieved with any of the tests. For a fixed
percent FF, the percent E is determined since the sum of FF and E is the
5-32
-------
OO
0
GARAGE
INSTRUMENT
LABORATORY
INSTRUMENT
MAXIMUM DIFFERENCE
Eo
FF
10%
10%
0
10
20
30
40
50
60
70
PERCENT EQ AND FF
Figure 5-19- Variation of Ec, Eo, and FF with Instrument Type;
HC; 1974 Model Year Fleet; Key Mode Test; High Speed Mode;
Bounded Errors of Commission Method
-------
en
i
o
UJ o
o_ 2
0
0
GARAGE
INSTRUMENT
LABORATORY
INSTRUMENT
MAXIMUM DIFFERENCE
E0<1%
FF <
10
20
30
40
50
60
PERCENT EQ AND FF
70
Figure 5-20. Variation of Ec, Eo, and FF with Instrument Type;
CO; 1974 Model Year Fleet; Key Mode Test; High Speed Mode;
Bounded Errors of Commission Method
-------
(Jl
I
(Jl
0
GARAGE
INSTRUMENT
LABORATORY
INSTRUMENT
MAXIMUM DIFFERENCE
EQ ~ 10%
FF ~ 10%
0
10
20
30
40
50
60
70
PERCENT EQ AND FF
Figure 5-21. Variation of Ec, Eo, and FF with Instrument Type;
HC; 1974 Model Year Fleet; Key Mode Test; Low Speed Mode;
Bounded Errors of Commission Method
-------
Ul
I
OJ
0
GARAGE
INSTRUMENT
LABORATORY
INSTRUMENT
MAXIMUM DIFFERENCE
E -17%
o
FF~17%
0
10
20
30
40
50
60
70
PERCENT EQ AND FF
Figure 5-22. Variation of Ec, Eo, and FF with Instrument Type;
CO; 1974 Model Year Fleet; Key Mode Test; Low Speed Mode;
Bounded Errors of Commission Method
-------
Ul
I
OO
o_ 2
0
GARAGE
INSTRUMENT
LABORATORY
INSTRUMENT
MAXIMUM DIFFERENCE
EQ~13%
FF ~ 13%
0
10
20
30
40
50
60
70
PERCENT EQ AND FF
Figure 5-23. Variation of Ec, Eo, and FF with Instrument Type;
HC; 1974 Model Year Fleet; Key Mode Test; Idle Mode; Bounded
Errors of Commission Method
-------
i
OJ
00
0
0
GARAGE
INSTRUMENT
LABORATORY
INSTRUMENT
MAXIMUM DIFFERENCE
FF ~ 18%
10
20
30
40
50
60
70
PERCENT EQ AND FF
Figure 5-24. Variation of Ec, E0, and FF with Instrument Type;
CO; 1974 Model Year Fleet, Key Mode Test; Idle Mode; Bounded
Errors of Commission Method
-------
UJ
O
UJ
0.
U)
0
GARAGE
INSTRUMENT
LABORATORY
INSTRUMENT
MAXIMUM DIFFERENCE
0
10
20
30
40
50
60
70
PERCENT EQ AND FF
Figure 5-25. Variation of Ec, Eo, and FF with Instrument Type;
HC; 1974 Model Year Fleet; Federal Three-Mode Test; High
Speed Mode; Bounded Errors of Commission Method
-------
Lul
0
GARAGE
INSTRUMENT
LABORATORY
INSTRUMENT
MAXIMUM DIFFERENCE
EQ~13%
FF~13%
10
20
30
40
50
60
70
PERCENT EQ AND FF
Figure 5-26. Variation of Ec, Eo» and FF with Instrument Type;
CO; 1974 Model Year Fleet; Federal Three-Mode Test; High
Speed Mode; Bounded Errors of Commission Method
-------
0
GARAGE
INSTRUMENT
LABORATORY
INSTRUMENT
MAXIMUM DIFFERENCE
E
0
FF ~ 9%
0
10
20
30
40
50
60
70
PERCENT EQ AND FF
Figure 5-27. Variation of Ec, Eo, and FF with Instrument Type;
HC; 1974 Model Year Fleet; Federal Three-Mode Test; Low
Speed Mode; Bounded Errors of Commission Method
-------
Ul
I
*.
is)
0
GARAGE
INSTRUMENT
LABORATORY
INSTRUMENT
MAXIMUM DIFFERENCE
EQ~21%
FF~21%
10
20
30
40
50
60
70
PERCENT EQ AND FF
Figure 5-28. Variation of Ec, Eo, and FF with Instrument Type;
CO; 1974 Model Year Fleet; Federal Three-Mode Test; Low
Speed Mode; Bounded Errors of Commission Method
-------
5r-
ui
i
oo
o
UJ
°- 2
0
GARAGE
INSTRUMENT
LABORATORY
INSTRUMENT
MAXIMUM DIFFERENCE
FF ~ 8%
0
10
20
30
40
50
60
70
PERCENT EQ AND FF
Figure 5-29. Variation of Ec, Eo, and FF with Instrument Type;
HC; 1974 Model Year Fleet; Federal Three-Mode Test; Idle
Mode; Bounded Errors of Commission Method
-------
o
a:
Ul
i
*»
f
GARAGE
INSTRUMENT
LABORATORY
INSTRUMENT
MAXIMUM DIFFERENCE
EQ~17%
FF~17%
10
20
30
40
50
60
70
PERCENT EQ AND FF
Figure 5-30. Variation of Ec, E0> and FF with Instrument Type;
CO; 1974 Model Year Fleet; Federal Three-Mode Test; Idle
Mode; Bounded Errors of Commission Method
-------
Ul
*
\jn
o
UJ
0-
0
GARAGE
INSTRUMENT
LABORATORY
INSTRUMENT
MAXIMUM DIFFERENCE
10%
FF ~ 10%
Eo~
0
10
20
30
40
50
60
70
PERCENT EQ AND FF
Figure 5-31. Variation of Ec, Eo, and FF with Instrument Type;
HC; 1974 Model Year Fleet; Unloaded 2500 rpm Test; Bounded
Errors of Commission Method
-------
O
UI
I
I
GARAGE
INSTRUMENT
LABORATORY
INSTRUMENT
MAXIMUM DIFFERENCE
EQ -19%
FF ~19%
0
10
20
30
40
50
60
70
PERCENT EQ AND FF
Figure 5-32. Variation of Ec, E0i and FF with Instrument Type;
CO; 1974 Model Year Fleet; Unloaded 2500 rpm Test; Bounded
Errors of Commission Method
-------
FTP rejection rate. Thus, the "best" test for fixed percent FF is the one
with the lowest percent E . In general, the bag-type STs are better in this
respect. However, the actual level of percent E on the volumetric tests is
still quite low. For example, at 30 percent FF on the CO Federal Short
Cycle (Figure 5-9), the percent EC is essentially zero. For CO on the Key
Mode low-speed mode, percent EC is 0. 65 percent for laboratory instruments
(Figure 5-11) and 3.85 percent for garage instruments (Figure 5-12).
5.2.2.2 Multiple-Constituent Tests
In addition to analyzing each pollutant individually, an analysis
was made for multiple-constituent tests. The method of analysis and com-
putational procedures were the same as for the CEV fleet, as discussed in
Sec. 4.2.2.5.2.
5. 2. 2. 2. 1 Bounded Errors of Commission Results
Three-constituent test results for the Federal Short Cycle
and the Federal Three-Mode (high-speed and idle modes only) are displayed
in Figures 5-33 through 5-35. Both laboratory and garage instrument results
are included. The data plotted are the parametric results. For a detailed
discussion of the plot presentation, see Sec. 4. 2. 2. 5. 2. 1.
5. 2. 2. 2. 2 Variance Estimates
Table 5-8 shows the approximate cell standard deviation for a
range of cell percentages, assuming N = 147. See Sec. 4. 2. 2. 5. 2. 2 for a
detailed discussion of the approximation procedures.
5.2.2.2.3 Discussion of Results
A comparison of modes on the Federal Three-Mode ST indi-
cates that the idle mode may be more favorable. Using laboratory instru-
ments, the idle mode has fewer errors of commission while maintaining a
superior percent FF and percent E relation over the high-speed mode for
most of the range of predicted percent EC shown in Figure 5-34. Using
garage instruments (Figure 5-35), no statistical difference between the modes
is observed.
5-47
-------
8
o
a:
<
o
0
I
I
80i-
60
0 tt
a: "-
o
<
20
0
FF
o
PREDICTED PERCENTE
Figure 5-33. Variation of Actual Ec, Eo, and FF with
Predicted Ec; Federal Short Cycle; Three-Constituent
Test; Bounded Errors of Commission Method; 1974
Model Year Fleet
5-48
-------
,» 4
o
ce.
tt
0
ACTUAL PERCENT Er - 0 FOR IDLE MODE
* c
80
60
40
o
20
0
FF
HIGH SPEED
1
0
2 3
PREDICTED PERCENT
Figure 5-34. Variation of Actual Ec, E0, and FF with
Predicted Ec; Federal Three-Mode; Three-Constituent
Test; Laboratory Instruments; Bounded Errors of Com-
mission Method; 1974 Model Year Fleet
5-49
-------
ACTUAL PERCENT EQ - 0 FOR IDlf MODE
ACTUAL PERCENT EQ = 0.68 FOR HIGH SPEED MODE ±0.67
FF
801-
60
O U_
of Q
I <
l-<
O
<
-40
20
0
OHIGH SPEED
AIDlf
HIGH SPEED
AIDL£
0
1 2 3
PREDICTED PERCENT E
Variation of Actual Ec, Eo, and FF with
Figure 5-35.
Predicted Ec; Federal Three-Mode; Three-Constituent
Test; Garage Instruments; Bounded Errors of Commis-
sion Method; 1974 Model Year Fleet
5-50
-------
Table 5-8. Standard Deviation for Three-Constituent
Tests: 1974 Model Year Fleet, N = 147
Cell
Percentage
60
50
40
30
20
10
5
3.5
Cell
Standard
Deviation, %
4.04
4.12
4.04
3.78
3.30
2.45
1.80
1.52
Comparison of the Federal Short Cycle and the Federal Three-
Mode can be made over a limited range of the results. For the actual per-
cent E less than 2 percent, the laboratory results of the Federal Three-
c
Mode and the Federal Short Cycle are comparable. Table 5-9 indicates the
minimum and maximum for percent FF and percent EQ, while percent EC is
less than 2 percent. There is little difference between the idle mode and the
Federal Short Cycle. Over this range of percent EC> the idle mode would
appear favorable to the Federal Short Cycle due to the low value of percent
E on the idle mode.
Q
A comparison of instrument types shows that the laboratory
instruments are generally preferable.
5-51
-------
Table 5-9. ST Comparison: 1974 Model Year Fleet;
Multiple Constituent Tests (E S 2%)
Short Test
Federal Short Cycle
Federal Three -Mode:
Idle
High
% FF
Min
25
22
5
Max
36
38
42
%EO
Min
44
42
38
Max
55
58
75
5-52
-------
5.3 RELATIVE IMPACT ON AIR QUALITY
5.3.1 By Individual Pollutant
The FTP standards, or cut-points, can be interpreted as
establishing the desired impact on air quality in that the FTP cut-points fix
the percent of the population classified as high- polluting vehicles. If the
FTP were used as the test procedure in an inspection/ maintenance program
which tested all vehicles (i. e. , as the ST), the relative impact on air quality
would ideally be 100 percent; that is, all the vehicles that are failures are
in fact identified as such.
Similarly, the effectiveness of the various STs can also be
used as a measure of impact on air quality, where "ST effectiveness" is
defined as :
c_ ,, .. __ % FF for the short test . .
ST effectiveness =
FTP faiiures in same population
% FF
~ % FF + % E
o
Thus, on this basis, the ST is always less effective than the FTP, in
proportion to the percent of errors of omission (E ) associated with a given
ST. Table 5-10 shows the ST effectiveness values for the 1974 model year
fleet for an E rate of 5 percent. These values indicate the relative impact
on air quality of the ST as compared with the impact of the FTP on air quality,
for the E conditions shown.
c
Actual benefit or impact is dependent upon the user's needs
and constraints. One measure of benefit would be the tons of pollutant re-
moved from the atmosphere on an annual basis in a given region by the use
of an ST in an inspection/maintenance program. This can be approximated
by the relationship:
Tons removed = ST effectiveness X A pollutant to be removed
in population X % population sampled (5-2)
5-53
-------
Table 5-10. Short Test Effectiveness; E = 5%
1974 Model Year Fleet
Short Test
Federal Short Cycle
NY/NJ Composite
Key Mode
Laboratory
Garage
Federal Three -Mode
Laboratory
Garage
2500 rpm Unloaded
Laboratory
Garage
ST Effectiveness**'
HC
0.83
0.78
0.58
0.34
0.61
0.41
0.61
0.39
CO
0.90
0.88
0.76
0.51
0.72
0.48
0.73
0.47
NOX
0. 17
0.06
0.28
0.22
0.22
%FF
HC
34
32
24 (I)(b)
14 (L)
25 (I)
17 (I)
25
16
CO
65
64
55 (L)
37 (H)
52 (I)
35 (I)
53
34
NOX
3
1
5 (I)
4(H)
4
(a)
ST Effectiveness =
where
FF
FTP Fails
(b)
FTP HC Fails = 41. 09%
FTP CO Fails = 72. 35%
FTP NOX Fails = 17. 8%
I = idle mode
L = low speed mode
H = high speed mode
5-54
-------
where
ST effectiveness . ^ r £
o
and
A pollutant to be removed in population = average value for the
population of HC, CO,
or NOX, in tons/year,
in excess of that per-
mitted by the FTP
standard; it is based
on the FTP failures
and corresponding
emission values ob-
served in the popula-
tion, and vehicle-miles-
traveled characteristics
This relationship ignores those additional benefits likely to occur if the failed
vehicles were repaired and achieved emission levels below the FTP standards
after repair.
Equation (5-2) indicates areas of tradeoff that should be ex-
amined prior to the implementation of a specific inspection/maintenance
program. Figure 5-36 depicts one aspect of such tradeoffs. This figure is
an illustrative plot of Eq. (5-2) for two different ST (Federal Short Cycle,
and Unloaded 2500 rpm with garage instruments) as used for CO emissions.
As indicated in Table 5-10, their effectiveness values are 0.90 and 0.47,
respectively; i. e. , as compared with the CO discrimination capability of the
FTP procedure, they are 90 and 47 percent as effective as the FTP in iden-
tifying vehicles which fail the FTP test on CO. Thus, to achieve the same
benefit in total CO pollutant removal, the percentage of the population that
must be sampled by the Unloaded 2500 rpm ST is approximately double that
which must be sampled with the Federal Short Cycle ST. Alternatively stated,
for any given percent sampling of the population, the use of the Federal Short
Cycle ST would result in approximately double the amount of CO removed.
The complexity of program implementation can be measured
in annual cost. The cost components would include such items as annual
5-55
-------
o
o
FEDERAL
SHORT
CYClf
2500 rpm
UNLOADED
PERCENT VEHICLE POPULATION SAMPLED
Figure 5-36. Impact of Percent Population Sampled on CO
Removed (Illustrative Example Only)
5-56
-------
operating expenses, maintenance expenses, and amortized initial development
and installation expenses. The ST requiring laboratory instrumentation would
have substantial initial procurement costs, and higher annual maintenance and
operating expenses than those using garage instruments. The bag-type ST
requires more skilled personnel and a CVS station. The bag ST and multi-
mode tests also require a dynamometer. Thus, the ST can be ranked
according to cost as follows:
Federal Short Cycle, NJ/NY Composite
Three-Mode volumetric with laboratory instruments
Three-Mode volumetric with garage instruments
2500 rpm Unloaded with laboratory instruments
2500 rpm Unloaded with garage instruments
For those inspection/maintenance programs targeted to 100 percent inspection
of all vehicles, the above ranking of ST by cost would appear valid. However,
if less than 100 percent inspection is envisioned for some reason, then addi-
tional factors should be considered. For example, the unit cost of a program
(per vehicle) would be expected to decrease as the percent of the population
sampled increases. Thus, in the example of Figure 5-36, if the program were
targeted to a defined level of CO removal, a cost-benefit analysis might be an
appropriate method to select the ST and the percentage sampled for minimum
cost purposes. The type of constraint normally imposed on a tradeoff study
would typically be total annual cost; however, additional constraints on per-
cent E or percent rejected (E plus FF) are also admissible under this
approach. Other areas of consideration are effective sampling and site
selection, importance of the pollution source as a function of geographic
location, social impact, etc.
5.3.2 Multiple Constituent Tests
Short test effectiveness is also a useful measure of test quality
for the multiple-constituent test, although the pollutant removal implications
of Eq. (5-2) must apply on an individual pollutant basis. Shown in Table 5-11
5-57
-------
Table 5-11. Short Test Effectiveness Values for Multipl
Constituent Tests; 1974 Model Year Fleet
Pie
la)
Short Test
Federal Short Cycle
Federal Three-Mode
(Laboratory Instruments)
Idle
High
Federal Three-Mode
(Garage Instruments)
Idle
High
ST Effectiveness
0.77
0.373
0.314
0.483
0. 568
0.330
0.374
Percent EC
Predicted(b)
5
0.05
0.01
5
5
5
5
Actual
8.84
2.04
0.68
0.00
2.72
0.00
0.69
(a)
(b)
FTP failures = 80%
Using bounded errors of commission method of analysis
are the effectiveness values for the Federal Short Cycle and the Federal
Three-Mode. Comparison of the test-to-test effectiveness values should,
of course, be made at points where the actual percent E is equal; however,
this can be only approximated with the existing data.
The technical favorability of the Federal Short Cycle is
diminished when comparing on the basis of equivalent percent E . Although
the Federal Short Cycle effectiveness is 0.77 at actual percent E equal to
8. 84, it is reduced to 0. 373 and 0. 314 for actual percent E values of 2. 04
and 0.68, respectively. However, as shown in Table 5-11, the effectiveness
values of the high-speed mode of the Federal Three-Mode ST with laboratory
5-58
-------
and garage instruments are 0. 568 (actual percent E = 2. 72) and 0. 374 (actual
percent E = 0. 69), respectively. Comparable effectiveness values for the
idle mode with laboratory and garage instruments are 0. 483 and 0. 330, re-
spectively, both with actual percent E equal to 0. Thus, in the actual per-
cent E range below approximately 3, the Federal Three-Mode ST with
garage instruments (idle or high-speed mode) is essentially equivalent to the
Federal Short Cycle in effectiveness while the Federal Three-Mode ST with
laboratory instruments has a higher effectiveness than the Federal Short
Cycle.
Although the favorability of the laboratory instruments over
the garage instruments persists under this method of comparison, considera-
tion of program complexity could bias test desirability in favor of the Federal
Three-Mode with garage instruments.
5-59
-------
6. DEFECT DATA FROM CATALYST-EQUIPPED
EXPERIMENTAL VEHICLE FLEET
-------
6. DEFECT DATA FROM CATALYST-EQUIPPED
EXPERIMENTAL VEHICLE FLEET
Upon completion of the FTP and ST tests performed on the
CEV fleet as described in Sections 3 and 4, 95 defect tests were performed
on 5 of the vehicles from the 40-vehicle CEV fleet.
The 95 defect tests simulated a wide variety of malfunctions
that could occur in typical passenger cars. The general categories of
defects are defective ignition components, changes in ignition timing, dwell,
and spark advance, faulty carburetion, defective valves, clogged air filter,
and faulty emission control components. The defects were introduced
individually and mixed. The Appendix lists the defect test runs on the five
cars. These test data were analyzed to (a) determine the statistical char-
acter of the defect tests, and (b) to examine the ability of the STs to detect
defective vehicles of this nature. The results are discussed below.
6.1 STATISTICAL ANALYSIS OF DEFECT TESTS
Listed in Table 6-1 are the estimated ST/FTP correlation
coefficients for the ungrouped defect data and the original 40-car catalyst-
equipped fleet (first good data only), using the method defined in Sec.
4.2.1.1. The HC correlations are consistently higher, over 0.9, among
the defect data than the previous 40-car CEV fleet. Addition of all defect
data to the original CEV fleet data will significantly distort the population
characteristics with regard to HC. CO and NO distortion will also occur,
3C
although not as pronounced as with HC.
This distortion is also evident when examining elementary-
statistics. Table 6-2 compares statistics on the FTP data for the two
groups. Clearly the data are different and need to be analyzed as distinct
groups since the proportion of defect cars to normally operating cars in the
true population is unknown.
6-1
-------
Table 6-1. ST/FTP Correlation Coefficient Comparison:
Defect Test Vehicles vs Original CEV Fleet
(laboratory instruments)
Test
Federal
Short Cycle
NY/NJ
Composite
Key Mode
Federal
Three -Mode
2500 rpm
Unloaded
N
-------
Table 6-2. Elementary FTP Statistics: Defect Test Vehicles
vs Original CEV Fleet (gm/mi)
Pollutant
HC
CO
NO
X
Defect
Mean
4.35
10.04
3.23
Standard
Deviation
6.00
11.81
1.42
Original
Mean
0.64
2.86
2.48
Standard
Deviation
0.54
1.52
0.59
Many of the defect tests are either replications or produce
similar data. The defect tests for each car were grouped according to
similarity of defect (see Appendix, under the column denoted Group No.).
Group No. 1 is the baseline group and represents the normally operating
vehicle. A test for a significant difference in the FTP average values of
the defect group and the base group was made for each defect group on
each car. Defect groups that have no significant difference cannot be
statistically distinguished, on the basis of their FTP values, from the
baseline group. The defect group contains at least one test distinguishable
from normal operation, if there is a significant difference. The distinguish-
able defect groups were further analyzed for similarity among themselves.
The result of this analysis is a smaller set of defect tests,
on each car, that are statistically different from one another. These test
data are then taken to represent observations on independent vehicles.
Thus, the 95 tests on 5 cars were reduced to approximately 24 defect test
observations representing 24 distinct vehicles each with a defect. The
results of the analysis are shown in Table 6-3.
6-3
-------
Table 6-3. Groups Distinguishable from Baseline
Operation: Defect Test Fleet
Di stingui shable
Car ID Group No.
Description of Defect
162
4 Lean main fuel system
6 EGR circuit reduced flow
8 Valves defective (exhaust)
9 Valves defective (intake)
Groups 4 and 9 are statistically similar for Car 162
164
6
7
8
9
Inefficient catalyst
Inefficient catalyst and 10% misfire
Inefficient catalyst and 5% CO idle
Baseline after leaded fuel use
165
3 Early power circuit activation
4 No secondary air injection
6 Rich idle and 10% misfire
7 No EGR and 6° timing advance
8 Reduced secondary air and oversize fuel jets
Groups 4 and 8 are statistically similar for Car 165
169
2
3
8
9
10
Timing under-advanced
Timing over-advanced
Rich idle and no secondary air
Rich idle and PCV closed
Defective spark plug
170
3 Rich idle 8% CO
4 10% intermittent misfire
5 3% intermittent misfire
6 No EGR
8 10% misfire and rich idle
9 10% misfire and lean idle
10 10% misfire and no EGR
11 Rich idle and no EGR
14 Rich idle and rich main
Groups 6 and 11 are statistically similar for Car 170
6-4
-------
6.1.1 Data Selection Procedures
The statistical procedure used to test for differences between
ife
groups was a multivariate linear hypothesis test. The likelihood ratio
statistic which has an equivalent F-statistic was used to make the test of
significance at the 95% level. The analysis was conducted on the FTP data,
as these are most representative of the true state of the vehicle. The
conclusion of this analysis is shown in Table 6-3.
To establish a data base for further analysis, actual data
from the individual groups were selected according to the following rules:
a. One run (testing sequence) may be selected from each
distinguishable group. If distinguishable groups are
similar, only one run may be selected from the
similar groups.
b. Run preferences are:
1. More acceptable ST data
2. Less ambiguity in the run
3. Lowest run number
As the assumption of independence of the observations is
crucial to contingency table analysis, the 95 defect tests were statistically
pruned to 24 tests representing 24 independent defective vehicles. These
data are considered to represent a population distinct from the original
40-car population. Of these 24, 6 have no Federal Three-Mode (laboratory)
data and 5 have no Key Mode (laboratory) data.
*T. W. Anderson, An Introduction to Multivariate Statistical Analysis,
John Wiley and Sons, Inc., New York (1958).
6-5
-------
6.2 CONTINGENCY TABLE ANALYSIS OF DEFECT
DATA
The analysis proceeded in two stages. The original CEV
fleet population was first analyzed, using first good data. The analysis
method was the bounded errors of commission procedure, which established
the ST cut-points (see Sec. 4.2.2.1.2). Percent E was varied from 10%
to 1% in 1% increments, with the addition of points at 0.5% and 0. 1%.
Immediately following analysis of the original CEV fleet, the defect popula-
tion was analyzed. The contingency table results were calculated for this
population, using the cut-points previously determined from the original
CEV fleet population. The computations were performed at each of the E
settings. Thus the analysis is merely an assessment of how well a test
constructed using an unknown mix of normal and defect operation data will
perform on a population of defective vehicles known to represent extreme
departures from normal operation.
A summary of the analysis on each constituent is given in
Table 6-4. The ST cut-points were established for E less than or equal to
5%, and the FTP level was level I (HC = 0.41 gm/mi, CO = 3.4 gm/mi,
NO = 3. 1 gm/mi).
3t
Sample plots are shown in Figures 6-1 through 6-6 for the
Clayton Key Mode (laboratory data). Comparing Figures 6-1 and 6-2, which
represent the analysis for HC, at E equal to 0. 1%, the original fleet has
approximately 33% E , and 35% FF. The defect data show E at 5%, E at
8%, and FF at 66%. As the loci of Figure 6-2 are relatively flat, the defect
discrimination qualities of the Key Mode on HC appear virtually insensitive
to policy decisions of 10% E or less.
The results of three- and nine-constituent tests for the Key
Mode (laboratory) are shown in Table 6-5. These results are typical for
all the multi-constituent tests.
6-6
-------
Table 6-4. Defect Analysis Comparison Summary:
Predicted Population [% E = 5^a),
FTP Level I
-------
i
00
70
60
it 50
o
\40
LU
~O
LU
z 30
LU
fc 20
10
0
H * HIGH SPEED
L LOW SPEED
IDLE
0 100 200 300 400 500 600 700
HC CUT-POINT, ppm
it
0
«c
"o
LU
0
LU
^_
LLJ
O
0£
LLJ
CL
80
70
60
50
40
30
20
10
0
- .^JBO**,*^
1 H L
r ii - HIPH Trrn
Fc L LOW SPEED
fco 1 IDLE
FF
i "i 1 r 1 1-
100 200 300 400 500
HC CUT-POINT, ppm
600 700
Figure 6-1. Variation of Ec, EO, and FF
with HC Cut-point; Original CEV Fleet;
Key Mode Test; 1975 FTP Level = 0.41
gm/mi; Bounded Errors of Commission
Method ^
Figure 6-2. Variation of Ec, Eo, and FF
with HC Cut-point; Defect Tests Only; Key
Mode Test; 1975 FTP Level = 0.41 gm/mi;
Bounded Errors of Commission Method
-------
0
FF
H = HIGH SPEED
L = LOW SPEED
I = IDLE
0
50 ~
40 -
30
»
20
10
70
60
50
40
30
LU
O
a:
£ 20
10
Ec
Eo
FF
H = HIGH SPEED
L = LOW SPEED
I = IDLE
\
\
400 600 800
CO CUT-POINT, ppm
1000 1200
200 400 600 800
CO CUT-POINT, ppm
1000 1200
Figure 6-3. Variation of EC, E0, and FF
with CO Cut-point; Original CEV Fleet;
Key Mode Test; 1975 FTP Level = 3.4
gm/mi; Bounded Errors of Commission
Method
Figure 6-4. Variation of EC, EQ, and FF
with CO Cut-point; Defect Tests Only; Key
Mode Test; 1975 FTP Level = 3.4 gm/mi;
Bounded Errors of Commission Method
-------
I
»-*
o
70
60
50
. 40
o
30
20
10
0
c
Eo
FF
H HIGH SPEED
L = LOW SPEED
I = IDLE
200 400 600 800 1000 1200 1400 1600 1800 2000 2200
NO CUT-POINT, ppm
A
Figure 6-5. Variation of Ec, Eo, and FF with NOX Cut-point; Original CEV
Fleet; Key Mode Test; 1975 FTP Level = 3. 1 gm/mi; Bounded
Errors of Commission Method
-------
70
60
50
O
o:
30
20
10
0
FF
H = HIGH SPEED
L = LOW SPEED
I = IDLE
200 400 600 800 1000 1200 1400 1600 1800 2000
NO CUT- POINT, ppm
A
2200
Figure 6-6. Variation of Ec, Eo, and FF with NOX Cut-point; Defect Tests
Only; Key Mode Test; 1975 FTP Level - 3. 1 gm/mi; Bounded
Errors of Commission Method
-------
Table 6-5. Key Mode Composite Test*a' (laboratory data)
Test Type
Three- constituent:
High Speed
Low Speed
Idle
Nine - constituent
Original
CEV Fleet
% FF
27.5
22.5
60.0
62.5
%Ec
5.00
5.00
5.00
12.50
%Eo
37.5
42.5
5.00
2.50
Defect
Fleet
% FF
89.5
73.7
89.5
94.7
%tr
c
0
0
0
0
%TT*
Eo
10.5
26.3
10.5
5.26
(a)
% Ec < 5; FTP Level I (HC = 0.41 gm/mi, CO = 3.4 gm/mi, NO =
3. 1 gm/mi) x
6.3
CONCLUSIONS
A review of the typical results illustrates that the short tests
perform well at isolating a population of defective cars. This is noted by
the general tendency for percent FF to increase and percent E to decrease
in the defect population. Although percent E decreased for HC, this was
not generally true for CO and NO .
JL
The sources of the errors of commission and omission are
two-fold. The first and usual source is that of the test procedures, i.e.,
measurement errors. The second source is due to mixing of defects. An
observation was classified as a defective car if any component of this vehicle
was defective. Hence, all the NO data analyzed are not representative of
j£
NO defects, for example. The multiple-constituent tests (which tend to
eliminate mixing errors), show a very high probability, greater than 70%,
of detecting defect vehicles (note that all the defective cars failed the FTP
at Level I).
In conclusion, the ST/FTP tracking of defective vehicles is
very good.
6-12
-------
APPENDIX
DEFECT TEST DESCRIPTIONS
-------
Table A-l. Defect Vehicle Test Schedule and Defect Description
Step
No.
1
2
3
4
5
6
7
8
9
10
-
11
12
Type of Defect
Baseline
Idle system lean
Baseline
Idle system low
rpm
Idle
Baseline
Lean main fuel
system
Baseline
Carburetor
power circuit
Baseline
EGR circuit
reduced flow
Baseline
Car Setup Procedure
Car 1754162
Check CO, timing, dwell, etc., and record. Perform one
baseline test on the vehicle.
Lean idle system to either 0. 5% CO before catalyst with
secondary air disconnected or 100-rpm drop lean from
lean best idle. Do not allow excessive misfire, however.
Return idle setting to original setting.
Decrease idle rpm 75 to 100 rpm while holding all other
parameters at manufacturer's specifications.
Decrease idle rpm by 150 rpm, providing misfire is not
encountered.
Return car to original setting.
Install main fuel jets that are two sizes (0. 002 in. )
smaller than original fuel jets. Fuel float level remains
same as with original jets.
Remove jets and reinstall original jets. Run one baseline
test.
Disable carburetor power circuit so that the vehicle
receives no power circuit operation.
Return vehicle to original condition by reactivating power
circuit.
Reduce EGR flow in EGR circuit by approximately 50% by
blocking EGR tube to carburetor baseplate.
Restore full EGR flow and return vehicle to original
condition.
Number of
Tests This
Step
1
2
0
2
1*
0
2
1
2
0
2
0
Group
No.
1
2
3
3
4
1
5
6
Olson Labs
Run
Numbe r
(A07752)1
A07905
A07947
A07961
A07974
A07984
A08125X
A08141
A08172
A08191
A08242
A08254
A08260
A08264X
1 = Baseline replicate
* = Tests that require temperature and flow measurements
X = Runs with temperature and flow measurements
-------
Table A-l. Defect Vehicle Test Schedule and Defect Description (Continued)
ro
Step
No.
13
14
15
16
17
1
2
3
4
Type of Defect
Fuel pump low
Baseline
Valves defective
Valves defective
intake
Baseline
Baseline
Advanced basic
ignition timing
Baseline
Insufficient
secondary air
Car Setup Procedure
Car 1754162 (Continued)
Reduce fuel pump pressure by 25% and test vehicle once.
Restore full fuel pump pressure and run one baseline test.
Remove cylinder head from vehicle. Obtain one replace-
ment exhaust valve from a Ford dealer and cut a wedge
in the face of the valve which has an area removed cor-
responding to 5 to 10% of the total valve face area. In-
stall valve in the front cylinder and reinstall head. Main-
tain the same valve lash as for the original valve
removed.
Remove cylinder head and defective exhaust valve. Obtain
the corresponding intake valve for this vehicle and also
take a wedge of 5 to 10% of the total valve face from the
intake valve. Install the front cylinder. Install original
nondefective exhaust-valve.
Remove cylinder head and defective valve. Reinstall
original valve. Run one baseline test.
Car 2104164
Check CO, timing, dwell, etc., and record. Perform one
baseline test on the vehicle.
Using a distributor with vacuum and centrifugal advance
characteristics representative of the five cars under test,
advance the idle timing by 6° (not to exceed audible knock
during first large acceleration on FTP when engine is hot).
Return timing to original setting.
Modify the secondary air supply system (larger pulley,
air leak, etc.) so as to obtain approximately a 50% reduc-
tion in secondary air injection.
Number of
Tests This
Step
1*
1
2
2
1*
1
2
0
2
Group
No.
7
1
8
9
1
1
2
3
Olson Labs
Run
Number
A08278X
A08293
A08371
A08377
A08431
A08445
A08477X
(A07751)1
A07812
A07950
A07960
A07972
A07983
1 = Baseline replicate
* = Tests that require temperature and flow measurements
X = Runs with temperature and flow measurements
-------
Table A-l. Defect Vehicle Test Schedule and Defect Description (Continued)
>
(JO
Step
No.
5
6
7
8
9
10
11
12
13
Type of Defect
Baseline
Over-rich main
fuel system
Baseline
High rpm idle
High rpm idle
Baseline
Inefficient
catalyst
Inefficient cata-
lyst plus inter-
mittent misfire
Inefficient cata-
lyst plus rich
idle
Car Setup Procedure
Car 2104164 (Continued)
Return secondary air injection system to normal.
Install main fuel jets that are three sizes (0.003 in.)
larger than original fuel jets, e. g. , 47F to 50F jet
sizes. Fuel float level remains as previously set.
Return main fuel jets to original size.
Increase engine idle speed by 150 rpm to approximately
800 rpm. All other parameters remain as at lower idle
speed.
Increase engine idle speed by 75 to 100 rpm to between
725 and 750 rpm.
Set all parameters to original baseline levels and test.
Drain the zero -lead fuel from the vehicle and refuel with
leaded regular gasoline. Operate the vehicle so as to
consume the tank of gasoline. Replenish the gasoline sup-
ply and test the vehicle once. Remove the leaded fuel and
replace with unleaded (30% of tank volume). Repeat the
test. Fill the vehicle with leaded fuel. Test again.
Note: The following tests contain two or more defects:
With the catalyst operating inefficiently, as in step No. 11,
introduce a 10% intermittent misfire rate and test on
leaded fuel.
Set idle CO at 5% (without secondary air). Ignition sys-
tem operating normally. Test using leaded fuel. Return
all components to normal and operate the car on unleaded
fuel at high loads and speed so as to reactivate the
catalyst.
Number of
Tests This
Step
0
2
0
1
2
1
3
2*
1*
Group
No.
4
5
5
1
6
7
8
Olson Labs
Run
Number
A07918
A08051
A08066
A08101X
A08110X
A08128
A08155
A08170
A08183
A08214X
A08231X
A08253X
* = Tests that require temperature and flow measurements
X = Runs with temperature and flow measurements
-------
Table A-l. Defect Vehicle Test Schedule and Defect Description (Continued)
Step
No.
14
1
2
3
4
5
6
7
8
9
Type of Defect
Baseline
Baseline
Retarded tim-
ing (basic)
Baseline
Early power
circuit
activation
Baseline
No secondary
air injection
Baseline
Timing over-
advancing
(vacuum)
Baseline
Car Setup Procedure
Car 2104164 (Continued)
Test the car on unleaded fuel. If the emissions have
returned to the original baseline level, proceed with the
next step. If the emissions have not returned to "nor-
mal," operate for one additional tank of unleaded fuel.
If the emissions have still not normalized, the remainder
of this vehicle's tests will be performed on another
vehicle.
Car 2364165
Check CO, timing, dwell, etc. , and record. Perform one
baseline test on the vehicle.
Using a distributor with vacuum and centrifugal advance
characteristics representative of the five cars under
test, retard the idle timing by 6°.
Return car to original condition.
Search the Ford Motor Company parts specifications and
determine the power value part number that is designed to
"come in" soonest, i.e., about 10 in. Install this part in
the carburetor.
Return car to original condition.
Deactivate the secondary air injection system.
Return car to original condition.
Modify the vacuum advance mechanism so as to give early
advancing without impacting the maximum advance ob-
tained. Modify so as to obtain the same advance at 10 in.
as would normally be obtained at 15 in.
Return the car to original condition.
Number of
Tests This
Step
2
1
2
0
2
0
2*
0
2
1
Group
No.
9
1
2
3
4
5
1
Olson Labs
Run
Numbe r
A08259
A08279
(A07906)1
A07934
A07948
A07963
A08003
A08052
A08100X2
A08180X
A08193
A08215
A08230
1 = Baseline replicate
2 = No cat bed roll
* = Tests that require temperature and flow measurements
X = Runs with temperature and flow measurements
-------
Table A-l. Defect Vehicle Test Schedule and Defect Description (Continued)
Step
No.
10
11
1Z
13
14
15
16
17
18
Type of Defect
Rich idle plus
intermittent
misfire of
spark plugs
Baseline
EGR not work-
ing plus ignition
timing advanced
Baseline
Reduced flow
from secondary
air system plus
over -rich main
fuel system
Reduced second-
ary air flow plus
lean main fuel
system
Baseline
Retarded igni-
tion timing plus
high idle rpm
Baseline
Car Setup Procedure
Car 2364165 (Continued)
Note: The following tests contain two or more common
defects:
Richen idle system to either 5% CO before catalyst with
secondary air disconnected or 100 rpm drop rich from
lean best idle plus introduce intermittent misfire at a
10% misfire rate.
Return the car to original condition.
Deactivate EGR system plus advance the idle timing by 6°
(no audible knocks).
Return the car to original condition. Run one baseline
test.
Modify secondary air supply system to obtain approxi-
mately a 50% reduction in secondary air injection plus
install main fuel jets that are three sizes larger than
original fuel jets.
Remove oversize jets and install undersize jets (two sizes
smaller) and retest with reduced secondary air flow
(reduction same as step No. 14).
Return the car to original condition.
Increase idle by 100 rpm and retard idle basic timing
by 6°.
Return the car to original condition. Run one baseline
test.
Number of
Tests This
Step
1
0
2
1
2
1
0
1
1
Group
No.
6
7
1
8
9
10
1
Olson Labs
Run
Numbe r
A08240X
A08256
A08258
A08267
A08295
A08307
A08320
A08432
A08444
X = Runs with temperature and flow measurements
-------
Table A-l. Defect Vehicle Test Schedule and Defect Description (Continued)
Step
No.
1
2
3
4
5
6
7
8
9
10
Type of Defect
Baseline
Timing under -
advancing
(vacuum)
Baseline
Timing over-
ad vane ing
(centrifugal)
Baseline
Timing under -
advancing
(centrifugal)
Baseline
Vacuum line
leaking
Baseline
PCV valve
stuck closed
Car Setup Procedure
Car 2544169
Check CO, timing, dwell, etc., and record. Perform one
baseline test on the vehicle.
Modify the vacuum advance mechanism so as to give late
advancing without impacting the maximum advance ob-
tained. Modify so as to obtain the same advance at 10 in.
as would be obtained at 5 in.
Return car to original condition.
Modify the centrifugal advance mechanics so as to give
early advancing without impacting the vacuum advance
circuit and without increasing the maximum centrifugal
advance possible. Modify so as to obtain the same
advance at 1500 rpm (distributor) as would be obtained
at 2000 rpm normally.
Return car to original condition.
Modify the centrifugal advance mechanism so as to give
late advancing without impacting the vacuum advance cir-
cuit or the maximum amount of centrifugal advance.
Modify so as to obtain the same advance at 2000 rpm
(distributor) as would be obtained at 1500 rpm normally.
Return car to original condition. Perform one baseline
test.
Remove one of the non-emission control device vacuum
lines from the "Christmas tree." Meter if necessary to
prevent excessive lean misfire which could cause engine
stalling.
Return car to original condition.
Remove PCV valve and plug PCV line, allowing no positive
crankcase ventilation.
Number of
Tests This
Step
1
2
0
2
0
2
1
2
0
1
Group
No.
1
2
3
4
1
5
6
Olson Labs
Run
Number
(A07922)1
A07935
A07973
A07987
A08020
A08050
A08065
A08083
A08124
A08132
A08140
A08182
1 = Baseline replicate
-------
Table A-l. Defect Vehicle Test Schedule and Defect Description (Continued)
Step
No.
11
12
13
14
15
16
17
18
19
20
Type of Defect
PCV valve
stuck open
Baseline
Vacuum spark
disconnect not
working
Baseline
Idle system too
rich plus sec-
ondary air
disconnected
Idle system too
rich plus vac-
uum spark dis -
connect not
working
Idle system too
rich plus PCV
valve blocked
Baseline
One defective
sparkplug
Baseline
Car Setup Procedure
Car 2544169 (Continued)
Remove blockage in PCV line and reconnect with PCV
valve in circuit but locked open.
Return to original condition by reinstalling good PCV
valve .
If the vehicle is equipped with a vacuum spark disconnect
circuit, render it inoperative.
Restore VSD circuit and return to original condition.
Perform one baseline test.
Note: The following tests (steps 15 through 18) contain
two or more defects:
Richen idle system to 5% CO before catalyst with second-
ary air disconnected.
With idle CO at 5% CO, disconnect vacuum spark discon-
nect circuit (secondary air system in operation during
testing).
With idle CO at 5%, plug PCV system so that there is no
flow into the intake manifold.
Return vehicle to original condition. Perform one baseline
test.
Disconnect the high tension lead to one spark plug to simu-
latfe a bridged plug or failed lead.
Perform one baseline test.
Number of
Tests This
Step
1
0
Defect i
1
1*
Defect i
1*
1
1
1*
Group
No.
7
Olson Labs
Run
Number
A08192
lot available
1
8
A08217
A08241X
ot available
9
1
10
1
A08266X
A08294
A08321
A08357X
* = Tests that require temperature and flow measurements
X = Runs with temperature and flow measurements
-------
Table A-l. Defect Vehicle Test Schedule and Defect Description (Continued)
00
Step
No.
1
2
3
4
5
5A
6
7
8
9
10
Type of Defect
Baseline
Rich idle
Rich idle
Baseline
Intermittent
misfire
Intermittent
misfire
Intermittent
misfire
Baseline
No EGR
Baseline
Clogged air
filter
Car Setup Procedure
Car 1614170
Check CO at idle with secondary air disconnected
upstream of the catalyst. Reconnect secondary air.
Richen idle system to either 5% CO before catalyst with
secondary air disconnected or 100 rpm drop due to en-
richment from lean best idle. Reconnect secondary air.
Richen idle system to 8% CO before catalyst with second-
ary air disconnected. Reconnect secondary air.
Return idle mixture to original setting.
Introduce intermittent misfire (electronically short cylin-
ders at random) at 10% misfire rate.
Introduce intermittent misfire (electronically short cylin-
ders at random) at 10% misfire rate.
Introduce intermittent misfire at 3% misfire rate.
Return ignition system to original condition and setting.
Deactivate EGR system.
Set all parameters (CO, ignition, and EGR) to original
baseline values and test.
Using a new air filter element, mask 95% of its flow area
or sufficient to cause a 10-fold increase in Ap and then
test vehicle. Leave the open zone of the element in two
quadrants of the circumference. Ap to be read at 50-mph
Key Mode loading. (Ap to be measured across element
only do not include Ap across air horn).
Number of
Tests This
Step
1
Z
1
0
1*
1*
Z*
0
Z
1
1*
Group
No.
1
Z
3
4
4
5
6
1
7
Olson Labs
Run
Number
(A07907)1
A07933
A07949
A0796Z
A08037
A08156X
A08190X
A08Z3ZX
A08478X2
A08243
A08255
A08257
A08265X
1 = Baseline replicate
2 = Run was made out of order; just prior to run A08504
* = Tests that require temperature and flow measurements
X = Runs with temperature and flow measurements
-------
Table A-l. Defect Vehicle Test Schedule and Defect Description (Continued)
i
vD
Step
No.
11
12
13
14
15
16
17
18
Type of Defect
Clogged air
filter
Baseline
Intermittent
misfire plus
idle system too
rich
Intermittent
misfire plus
idle system too
lean
Intermittent
misfire plus
EGR plugged
Baseline
Idle system too
rich plus EGR
not working
Idle system too
rich plus igni-
tion timing
advanced
Car Setup Procedure
Car 1614170 (Continued)
Mask or otherwise chock the flow of air through the air
filter element so as to obtain a 5-fold increase in Ap
across the air filter at 50-mph Key Mode loading. (Ap
to be measured across element only do not include Ap
across air horn).
Return the car to the original condition.
Note: The following tests contain two or more defects:
Introduce intermittent misfire at 10% misfire rate as in
step No. 5 plus richen up the idle system to 5% CO before
catalyst with secondary air disconnected.
Introduce intermittent misfire at 10% misfire rate as
in step No. 5 plus lean out the idle system to 0. 5% CO
(or lowest CO level possible without misfire) before
catalyst with secondary air disconnected.
Deactivate the EGR system plus introduce intermittent
misfire at 10% rate as in step No. 5.
Return the vehicle to original condition. Run one base-
line test.
Deactivate EGR system plus richen idle system to 5%
before catalyst with secondary air disconnected.
With 5% idle CO, advance basic idle timing 6°. EGR
system operating normally.
Number of
Tests This
Step
2
0
1*
1*
1
1
1
2*
Group
No.
7
8
9
10
1
11
12
Olson Labs
Run
Number
A08280
A08292
A08306X
A08319X
A08343
A08376
A08430
A08443X
A08446X
* = Tests that require temperature and flow measurements
X = Runs with temperature and flow measurements
-------
Table A-l. Defect Vehicle Test Schedule and Defect Description (Concluded)
Step
No.
19
20
11
Type of Defect
Idle system too
rich plus igni-
tion timing
retarded
Idle system too
rich plus main
fuel system too
rich
Baseline
Car Setup Procedure
Car 1614170 (Continued)
With 5% idle CO, retard basic idle timing by 6*.
Install main fuel jets that are three sizes too large as
per car No. Z, step 6 and set idle CO at 5% level with
secondary air disconnected.
Return the vehicle to original condition. Run one base-
line test.
Number of
Tests This
Step
1
1
1*
G roup
No.
13
14
1
Olson Labs
Run
Number
A08457
A08470
A08504X
I
N^,
O
* = Tests that require temperature and flow measurements
X = Runs with temperature and flow measurements
-------
TECHNICAL REPORT DATA
(Please read Instructions on the reverse before completing)
. REPORT NO.
EPA-460/3-76-011
2.
3. RECIPIENT'S ACCESSION NO.
4. TITLE AND SUBTITLE
Federal Test Procedure and Short Test
Correlation Analyses
5. REPORT DATE
April 1976
6. PERFORMING ORGANIZATION CODE
7. AUTHOR(S)
M. G. Hinton, J. C. Thacker, and W. B. Lee
8. PERFORMING ORGANIZATION REPORT NO.
ATR-76(7353)-l
9. PERFORMING ORGANIZATION NAME AND ADDRESS
The Mobile Systems Group
Environment & Energy Conservation Division
The Aerospace Corporation
El Segundo, California 90245
10. PROGRAM ELEMENT NO.
11. CONTRACT/GRANT NO.
68-01-0417
12. SPONSORING AGENCY NAME AND ADDRESS
EPA Office of Air and Waste Management
Office of Mobile Source Air Pollution Control
Emission Control Technology Division
Ann Arbor, Michigan 48105
13. TYPE OF REPORT AND PERIOD COVERED
Final
14. SPONSORING AGENCY CODE
15. SUPPLEMENTARY NOTES
16. ABSTRACT
A series of statistical analyses was performed to determine the degree of "correla-
lation" that exists between five specific short tests (STs) and the federal emission
certification test procedure (FTP) for new vehicles. This work was performed to
determine if "reasonable correlation with certification test procedures" exists; this
is a condition precedent to the promulgation of regulations that impose the in-use
warranty provisions of Sec. Z07 (b) of the Clean Air Act of 1970 upon the motor
vehicle manufacturers.
The basis for the analyses was ST and FTP test data from three vehicle fleets:
(a) a catalyst-equipped experimental vehicle fleet (40 vehicles), (b) an in-use 1974
model year vehicle fleet (147 vehicles), and (c) a catalyst-equipped defect test fleet
(5 vehicles). Each of the vehicles in these fleets was tested by the FTP and the
following STs: (a) Federal Short Cycle, (b) New York/New Jersey (NY/NJ) Com-
posite, (c) Clayton Key Mode, (d) Federal Three-Mode, and (e) Unloaded 2500 rpm.
Hydrocarbon (HC) and carbon monoxide (CO) measurements were recorded with
both laboratory analyzers and garage-type instruments for most of the volumetric
tests. All oxides of nitrogen (NOX) measurements were made with laboratory ana-
lyzers. Two different statistical analysis methods were used to assess "correla-
tion"--a conventional correlation analysis and a contingency table analysis.
17.
KEY WORDS AND DOCUMENT ANALYSIS
DESCRIPTORS
b.lDENTIFIERS/OPEN ENDED TERMS
COS AT I Field/Group
Air Pollution
Emission Testing
Short Test Procedures
Test Correlations
Air Pollution Control
Conventional Correlation
Analysis
Contingency Table
Analysis
Laboratory and Garage
Instruments
13 B
14 B
8. DISTRIBUTION STATEMENT
Unlimited
19. SECURITY CLASS (ThisReport)
Unclassified
21. NO. OF PAGES
276
20. SECURITY CLASS (Thispage)
Unclassified
22. PRICE
EPA Form 2220-1 (9-73)
------- |