Measurement of On-Road Emission
Rates by Laser Remote Sensing:
Westminster Field Study
SEPA
United States
Environmental Protection
Agency
-------
Measurement of On-Road Emission
Rates by Laser Remote Sensing:
Westminster Field Study
Assessment and Standards Division
Office of Transportation and Air Quality
U.S. Environmental Protection Agency
Prepared for EPA by
ERG, Eastern Research Group, Inc.
EPA Contract No. EP-C-17-011
Work Assignment 5-13
NOTICE
This technical report does not necessarily represent final EPA decisions
or positions. It is intended to present technical analysis of issues using
data that are currently available. The purpose in the release of such
reports is to facilitate the exchange of technical information and to
inform the public of technical developments.
A United States
Environmental Protection
^1 Agency
EPA-420-R-23-011
May 2023
-------
Table of Contents
Page
1.0 Executive Summary 1-1
1.1 Research Goals 1-1
1.2 Benefits of the RSD Instrumentation Used in this Study 1-2
1.3 Dispersion of Vehicle Pollutant Releases 1-2
1.4 RSD Data Signal Processing 1-3
1.5 RSD Emission Rate Method Performance 1-4
1.6 Areas for Future Development 1-5
2.0 Testing Description 2-1
2.1 Testing Location and Conditions 2-1
2.2 EDAR Configuration 2-2
2.3 Traffic Flow Monitoring Equipment 2-5
2.4 Test Vehicles 2-6
2.5 Test Vehicle Exhaust Emissions Equipment 2-8
2.6 Test Vehicle Running Loss Emissions Equipment 2-10
2.7 EDAR Data Collection for Mass Emission Rate Method Development 2-14
2.8 Field Data Handling and Storage 2-15
2.9 Westminster Dataset EDAR Quality Flag 2-16
2.10 Westminster Dataset and Analysis Program Locations 2-17
3.0 Field Data Collection Results 3-1
3.1 Test Vehicle Test Conditions 3-1
3.2 Model Years and Gross Vehicle Weight Ratings of Fleet Vehicles 3-4
3.3 Wind Speed and Direction 3-6
4.0 Detailed Data Post-Processing by HEAT 4-1
5.0 RSD Emission Rate Method: Step-by-Step Description 5-1
5.1 Remote Sensing Device 5-2
5.2 Pre-Processing Device 5-6
5.3 Separation/Estimation Device 5-8
Stage 1: Blind Source Separation 5-10
Stage 2: Emission Estimation 5-13
5.4 Vortex Shape Estimation Device 5-15
Step 1. Air Speed Calculation Device 5-16
Step 3. Vehicle Characteristics Device 5-18
in
-------
Table of Contents (Continued)
Page
Step 2. Vortex Entrainment Time Calculation Device 5-19
Step 4. Weights Calculation Device 5-20
5.5 Emission Calculation Device 5-25
Step 5. Pollutant RSD Signal Device 5-25
Step 6. 100% Illumination Speed Device 5-27
Step 7. Mass-in-Vortex Calculation Device 5-27
Step 8. Emission Rate Calculation Device 5-28
6.0 Quantities Needed by the RSD Emission Rate Method 6-1
6.1 Vehicle Air Speed and Direction 6-1
6.2 Vehicle Footprint Length and Width 6-5
6.3 100% Illumination Speed for this Study 6-11
6.4 Vortex Entrainment Time (VET) Functionality 6-12
6.5 Vortex Shape (Weights) Functionality 6-22
6.6 Future Improvement: EvapHC Release Location Detection 6-40
7.0 Demonstration of Signal Analysis of RSD Detailed Data 7-1
7.1 Adjustment and Improvement of RSD Detailed Data 7-1
Adjusting Constant-Level Offsets 7-2
Removing Outliers 7-6
De-Striping via Multi-Tonal Cancellation 7-7
Adaptive Notch Filtering 7-13
Interpolating Measured Pixel Positions to a Rectangular Grid 7-17
7.2 Blind Source Separation by Independent Component Analysis 7-22
7.3 Enhanced Blind Source Separation using Correlation Constraints 7-29
7.4 Estimation of EvapHC and ExhHC from Candidate Plumes 7-34
7.5 Development of Flags to Qualify Processed Detailed Data 7-45
8.0 Exhaust Concentrations Reported by EDAR 8-1
8.1 EDAR Exhaust Concentrations on Test Vehicles 8-1
8.2 EDAR Exhaust Concentrations on Fleet Vehicles 8-9
9.0 Performance of the Emissions Rate Measurement Methodology 9-1
9.1 Exhaust CO, NO, and CO2 Release Rates from Reference EVs 9-2
9.2 Total HC, Evaporative HC, Exhaust HC Release Rates from Reference EVs.... 9-3
9.3 Evaporative HC Release Rates from Reference Gasoline Vehicles 9-13
-------
Table of Contents (Continued)
Page
9.4 Recommendations for Development of the RSD Emission Rate Method 9-24
10.0 Application of the RSD Emission Rate Method to the Fleet Sample 10-1
10.1 Simulation of Vortex Entrainment using PEMS Data 10-1
10.2 Comparison of RSD Emission Rates and MOVES Release Rates 10-10
10.3 Comparison of the Traditional RSD Concentration Method with the RSD
Emission Rate Method 10-22
10.4 Characterization of NO Mass in the Westminster Fleet Sample 10-31
List of Appendices
Appendix A : Test Vehicle EDAR Exhaust Emissions Measurements
Appendix B : Westminster Dataset and Analysis Program Locations
Appendix C : Plots for Signal Adjustment Demonstration
Appendix D : Examples of Blind Source Separation using Standard ICA
Appendix E : Examples of Blind Source Separation with Correlation Constraints (BSScov)
Appendix F : Comparing Westminster and MOVES Release Rates by Class and Age Group
-------
List of Figures
Page
Figure 2-1. Roadways Used for EDAR Testing in Westminster, Colorado 2-1
Figure 2-2. Ambient Temperature at the Test Site During Testing 2-2
Figure 2-3. EDAR Test Set-Up (Looking West) at the Test Site 2-3
Figure 2-4. EDAR Test Site on N. Federal Parkway, Westminster, Colorado 2-4
Figure 2-5. Video Camera Installed Near EDAR System 2-5
Figure 2-6. Fake Tailpipe Location on EV-1 Test Vehicle 2-9
Figure 2-7. Fake Tailpipe Installed on EV-2 Test Vehicle 2-9
Figure 2-8. Rotameters and Diverter Valve for Simulated Running Loss Releases 2-11
Figure 2-9. Fake Fuel Fill Door Release Point on EV-2 Test Vehicle 2-11
Figure 2-10. Under-Hood Release Point on an EV Test Vehicle 2-12
Figure 2-11. Fake Tank Release Point on GMC Test Vehicle 2-12
Figure 3-1. Wind Speed vs. Wind Direction at 6 Meters above Pavement 3-6
Figure 3-2. Wind Direction at 6 Meters above Pavement 3-7
Figure 3-3. Wind Speed at 6 Meters above Pavement 3-7
Figure 5-1. Flow Diagram of Methodology without Separation/Estimation of Emission
Sources 5-2
Figure 5-2. Flow Diagram of Methodology with Separation/Estimation of Emission
Sources 5-2
Figure 5-3. HEAT Remote Sensing Device Test Set-Up 5-4
Figure 5-4. RSD ZigZag Scan Pattern on Pollutants from a Moving Vehicle 5-4
Figure 5-5. Flow Diagram of Separation/Estimation Device 5-9
Figure 5-6. Flow Diagram of Vortex Shape Estimation Device 5-16
Figure 5-7. Time-Decay Factor for Weights 5-21
Figure 5-8. Vehicle Length Factor for Weights 5-21
Figure 5-9. Air Speed Parallel Factor for Weights 5-22
Figure 5-10. Flow Diagram of Emission Calculation Device 5-25
Figure 5-11. Continuously Stirred Tank Analogy 5-29
Figure 6-1. Estimated Wind Speeds at 1 Meter above Pavement 6-1
Figure 6-2. Distribution of Fleet Vehicle Road Speeds 6-2
Figure 6-3. Vehicle Reference Frame: Fleet Air Speed Distribution 6-4
Figure 6-4. Vehicle Reference Frame: Fleet Air Direction Distribution 6-4
Figure 6-5. Vehicle Reference Frame: Fleet Distribution of Air Movement 6-5
Figure 6-6. Image of an EDAR Detailed Data Array 6-5
vi
-------
List of Figures (Continued)
Page
Figure 6-7. CO2 ScanSums and Blank Pixel Counts for a Car (45.5 mph,
Series_Transit=505_000299) 6-7
Figure 6-8. CO2 ScanSums and Blank Pixel Counts for a Vehicle with Trailer (19.8 mph,
Series_Transit=505_000262) 6-8
Figure 6-9. Footprint Size as a Function of GVWR for the Westminster Set 6-10
Figure 6-10. Exhaust CO2 VET vs. AirSpeed Para for EV-1 6-15
Figure 6-11. Exhaust CO2 VET vs. AirSpeed Para for EV-2 6-15
Figure 6-12. Residual VET Trend vs. AirSpeed Perp for EV-1 6-16
Figure 6-13. Residual VET Trend vs. AirSpeed Perp for EV-2 6-16
Figure 6-14. VET vs. EvapHC Release Location and Air Speed for EV-1 6-18
Figure 6-15. VET vs. EvapHC Release Location and Air Speed for F150 6-18
Figure 6-16. VET vs. EvapHC Release Location and Air Speed for GMC 6-19
Figure 6-17. VET vs. EvapHC Release Location and Air Speed for Subaru 6-19
Figure 6-18. ScanSum Traces for Vehicle with 10,913 mg/mile HC at 37.5 mph 6-24
Figure 6-19. Grand Average HC ScanSum Trace for HC Releases 6-25
Figure 6-20. Log of Grand Average HC ScanSum Trace for HC Releases 6-25
Figure 6-21. Log of HC ScanSum Traces Averaged by Vehicle ID 6-26
Figure 6-22. Log of HC ScanSum Traces Averaged by Road Speed 6-26
Figure 6-23. Log of HC ScanSum Traces Averaged by Release Location 6-27
Figure 6-24. Example Showing CO2 ScanSum Zero Adjustment 6-28
Figure 6-25. Distribution of the Parallel Component of the Vehicle Air Speed 6-29
Figure 6-26. Distribution of Vehicle Length for Westminster Fleet Vehicles 6-29
Figure 6-27. Full Distribution of Vehicle Length and Parallel Air Speed 6-31
Figure 6-28. Zoomed Distribution of Vehicle Length and Parallel Air Speed 6-31
Figure 6-29. CO2 Mass Trace Averaged by Vehicle Length 6-32
Figure 6-30. Log CO2 Mass Trace Averaged by Vehicle Length 6-32
Figure 6-31. CO2 Mass Trace Averaged by AirSpeed Para 6-33
Figure 6-32. Log CO2 Mass Trace Averaged by AirSpeed Para 6-33
Figure 6-33. Fit of 30,559 CO2 ScanSum Traces to an Exponential Decay 6-34
Figure 6-34. Average Residual CO2 Scansum Trace for 19-22 ft Length Bin 6-36
Figure 6-35. Average Residual CO2 Scansum Traces for Vehicle Length Bins 6-37
Figure 6-36. Normalized Average Residual CO2 Scansum Traces by Length for 40mph
AirSpeed Para 6-37
Figure 6-37. Average Residual CO2 Scansum Trace for 18-21 mph AirSpeed Para Bin 6-38
Figure 6-38. Average Residual CO2 Scansum Traces for AirSpeed Para Bins 6-39
vii
-------
List of Figures (Continued)
Page
Figure 6-39. Normalized Average Residual CO2 Scansum Traces by AirSpeed 6-39
Figure 6-40. EDAR Scansums v. Scan Number for DOOR Evaporative Releases 6-42
Figure 6-41. EDAR Scansums v. Scan Number for HOOD Evaporative Releases 6-42
Figure 6-42. EDAR Scansums v. Scan Number for TANK Evaporative Releases 6-43
Figure 6-43. EV-2 HC Releases from Different Locations in a Transverse Air Flow 6-43
Figure 6-44. Example Zones for Detecting Releases from Vehicle Locations 6-45
Figure 6-45. EvapHC Hood Release for a Single Transit in a Transverse Air Flow 6-45
Figure 7-1. Detailed Data Patterns for Example Westminster Transit 7-1
Figure 7-2. Histogram of CO2 Pixel Counts for Series=512 Transit=l 188 7-4
Figure 7-3. Example of Linear and Multi-Transit Prediction for HC Channel 7-9
Figure 7-4. Noise Power Spectra for RSD Channels in the Vehicle Direction 7-10
Figure 7-5. De-Striping Evaluation using HC Noise Power Spectra 7-11
Figure 7-6. Spatial Structure of EDAR Pixels for ZigZag Collection Pattern 7-13
Figure 7-7. Example of EDARNO2 Signal Collected for One Transit 7-14
Figure 7-8. Change in Tonal Disturbance Frequency for Test Vehicle Transits 7-15
Figure 7-9. Cartoon Demonstrating Interpolation to Rectangular Grid 7-18
Figure 7-10. Example of Interpolation of CO2 Data to a Rectangular Grid 7-19
Figure 7-11. BSS IC A Separation (p=0) of Example: EV-1, High EvapHC from TANK 7-31
Figure 7-12. BSScov Separation (p=0.1) of Example: EV-1, High EvapHC from TANK.... 7-32
Figure 7-13. Evaluation of Plume Outputs while Varying p for Example: EV-1, High
EvapHC from TANK 7-33
Figure 7-14. Use of Weights for Estimation for Example: EV-1, High EvapHC from
DOOR, Low Speed 7-37
Figure 7-15. Use of Weights for Estimation for Example: EV-1, Low EvapHC from
TANK, Low Speed 7-38
Figure 7-16. Use of Weights for Estimation for Example: EV-1, Medium EvapHC from
TANK, Highspeed 7-39
Figure 7-17. Use of Weights for Estimation for Example: EV-1, High EvapHC from
HOOD, Low Speed 7-40
Figure 7-18. 2-Dimensional CO2 Plume Averages for Different Parallel AirSpeed Ranges:
EV-1 and EV-2 7-42
Figure 7-19. 2-Dimensional CO2 Plume Averages for Different Perpendicular AirSpeed
Ranges: EV-1 at Low and High Speeds 7-43
Figure 7-20. 2-Dimensional CO2 Plume Averages for Different Perpendicular AirSpeed
Ranges: EV-2 at Low and High Speeds 7-44
Figure 7-21. Heatmaps for an Example Westminster Transit 7-46
Figure 7-22. Statistics for Adjusted CO and CO2 for the 127-Transit Sample Set 7-47
viii
-------
List of Figures (Continued)
Page
Figure 8-1. EDAR Exhaust Concentration Measurements on EV-1 8-2
Figure 8-2. EDAR Exhaust Concentration Measurements on EV-2 8-3
Figure 8-3. EDAR Exhaust Concentration Measurements on Subaru 8-4
Figure 8-4. EDAR Exhaust Concentration Measurements on Infiniti 8-5
Figure 8-5. Model Year Distribution of Fleet HC Concentration Measurements 8-11
Figure 8-6. Model Year Distribution of Fleet CO Concentration Measurements 8-12
Figure 8-7. Model Year Distribution of Fleet NO Concentration Measurements 8-13
Figure 8-8. Model Year Distribution of Fleet Mean [HC] Measurements 8-15
Figure 8-9. Model Year Distribution of Fleet Median [HC] Measurements 8-15
Figure 8-10. Model Year Distribution of Fleet Mean [CO] Measurements 8-16
Figure 8-11. Model Year Distribution of Fleet Median [CO] Measurements 8-16
Figure 8-12. Model Year Distribution of Fleet Mean [NO] Measurements 8-17
Figure 8-13. Model Year Distribution of Fleet Median [NO] Measurements 8-17
Figure 9-1. HC Performance (Average) for Test Vehicle EV-1 9-8
Figure 9-2. HC Performance (Average) for Test Vehicle EV-2 9-9
Figure 9-3. HC Performance (Details) for Test Vehicle EV-1 9-11
Figure 9-4. HC Performance (Details) for Test Vehicle EV-2 9-12
Figure 9-5. HC Performance (Average) for Test Vehicle F150 9-18
Figure 9-6. HC Performance (Average) for Test Vehicle GMC 9-19
Figure 9-7. HC Performance (Average) for Test Vehicle Subaru 9-20
Figure 9-8. HC Performance (Details) for Test Vehicle F150 9-21
Figure 9-9. HC Performance (Details) for Test Vehicle GMC 9-22
Figure 9-10. HC Performance (Details) for Test Vehicle Subaru 9-23
Figure 10-1. Vortex/RSD Simulation using PEMS Data for the Forester 10-3
Figure 10-2. NO v CO2 Release Rates as Measured by PEMS for Forester Snippet 10-6
Figure 10-3. NO v CO2 Release Rates as Simulated for RSD for Forester Snippet 10-6
Figure 10-4. NO v CO2 Release Rates as Measured by PEMS for Forester 10-7
Figure 10-5. NO v CO2 Release Rates as Simulated for RSD for Forester 10-7
Figure 10-6. NO v CO2 Release Rates as Measured by PEMS for Trail Blazer 10-9
Figure 10-7. NO v CO2 Release Rates as Simulated for RSD for Trail Blazer 10-9
Figure 10-8. LDT Exhaust CO2 Release Rates v. VSP Bin and Age Group 10-12
Figure 10-9. LDV Exhaust CO2 Release Rates v. VSP Bin and Age Group 10-12
Figure 10-10. LDT Exhaust NOx/NO Release Rates v. VSP Bin and Age Group 10-13
Figure 10-11. LDV Exhaust NOx/NO Release Rates v. VSP Bin and Age Group 10-13
Figure 10-12. LDT Exhaust CO Release Rates v. VSP Bin and Age Group 10-14
IX
-------
List of Figures (Continued)
Page
Figure 10-13. LDV Exhaust CO Release Rates v. VSP Bin and Age Group 10-14
Figure 10-14. LDT Exhaust Total HC Release Rates v. VSP Bin and Age Group 10-15
Figure 10-15. LDV Exhaust Total HC Release Rates v. VSP Bin and Age Group 10-15
Figure 10-16. Portions of Pollutant Mass in a 4-Second-VET Vortex 10-17
Figure 10-17. Comparison of NO v CO2 Release Rates by RSD for Westminster 15-to-19-
Year-Old LDVs 10-19
Figure 10-18. Comparison of NO v CO2 Release Rates for Datasets of 15-to-19-Year-Old
LDVs 10-21
Figure 10-19. Diagram of Traditional RSD Method for Calculating Exhaust Concentration
from Detailed Data 10-23
Figure 10-20. Pairwise Optical Mass Plots for the Example Transit (Series=515,
Transit=2469) 10-24
Figure 10-21. Traditional RSD Method Exhaust Concentration Results for the Example
Transit 10-26
Figure 10-22. Re-Calculation of Exhaust Concentrations after separating ExhHC from
EvapHC Plumes for the Example Transit 10-28
Figure 10-23. Diagram of RSD Method for Calculating Mass Emission Rates (g/mile) from
Detailed Data 10-29
Figure 10-24. Calculation of Mass Emission Rates (g/mile) for the Example Transit 10-30
Figure 10-25. Distributions of Number of Vehicles and NO Mass by NO Emission Rate
and Age 10-32
Figure 10-26. Fleet Fraction that Produces Most NO Mass 10-33
x
-------
List of Tables
Page
Table 2-1. Minimum Following Distances for Test Vehicles 2-4
Table 2-2. Descriptions of Test Vehicles 2-7
Table 2-3. Rotameter Settings Required for Designed Propane Releases 2-13
Table 2-4. Reported Variables in Master Dataset 2-16
Table 2-5. Transit Counts by EDAR QC Label and Vehicle Category 2-17
Table 3-1. Vehicles Used in the Test Vehicle Convoy 3-2
Table 3-2. Test Condition Combinations Used by the Test Vehicle Convoy 3-3
Table 3-3. Gross Vehicle Weight Ratings of Vehicles with Colorado Plates 3-4
Table 3-4. Model Years of Vehicles with Colorado Plates 3-5
Table 5-1. Terrain Surface Roughness Length Descriptions 5-17
Table 6-1. Release Location Factors for Test Vehicles 6-20
Table 6-2. VET Proportionality Constant vs. Vehicle and Release Location 6-20
Table 6-3. Fit of VET Proportionality Constant vs. Vehicle and Release Location 6-21
Table 6-4. Relative VET Proportionality vs. Release Location 6-21
Table 6-5. Relative VET by Vehicle Drag Area 6-22
Table 6-6. Exponential Vortex Time-Decay Constants for Various Dataset Strata 6-35
Table 7-1. Application of Improvement Steps to Transit Data 7-2
Table 7-2. Criteria for Selection of Processing Examples 7-23
Table 8-1. Reported Exhaust Concentrations for the Test Vehicles 8-6
Table 8-2. Dry, Artificial Exhaust Zero Performance by EV-1 Test Vehicle 8-7
Table 8-3. Dry, Artificial Exhaust Span Performance by EV-2 Test Vehicle 8-8
Table 8-4. Fleet Vehicle Model-Year-Mean and Median Concentration Measurements and
Confidence Intervals 8-14
Table 9-1. Combinations of Test Vehicles and Pollutant Types 9-1
Table 9-2. Comparison of Metered and RSD-Measured Exhaust CO, NO, and CO2 Release
Rates for EV-1 and EV-2 9-3
Table 9-3. Metered HC and Measured HC Average Responses for Test Vehicle EV-1 9-5
Table 9-4. Metered HC and Measured HC Average Responses for Test Vehicle EV-2 9-6
Table 9-5. Metered HC and Measured HC Average Responses for Test Vehicle F150 9-14
Table 9-6. Metered HC and Measured HC Average Responses for Test Vehicle GMC 9-15
Table 9-7. Metered HC and Measured HC Average Responses for Test Vehicle Subaru 9-16
Table 10-1. Kansas City PEMS Data on Two Vehicles 10-2
XI
-------
Acknowledgments
Colorado Department of Public Health and Environment staff, including Rob Dawson, Jim
Sidebottom, Jim Kemper, and Mike Mallory, provided key local Denver assistance with this
project. This included identifying and evaluating candidate on-road RSD sites, receiving delivery
of electric test vehicles, installation and operation of artificial exhaust gas metering equipment on
electric vehicles, installation and operation of artificial running loss metering equipment on all
test vehicles, providing logistics for the handling of test gas cylinders, and providing and driving
their personal vehicles as gasoline test vehicles during RSD testing. Yolla Hager of Hager
Emissions and Atmospheric Technology (HEAT) applied to the City of Westminster for HEAT
to become an approved contractor and then applied for and received the RSD site permit. James
Ashby, Kevin Stockton, and Andy Rimelman from PG Environmental provided critical
contributions for development, collection, and rendering of continuous videos of vehicle license
plates at the RSD site. Also, they set the pace of the test vehicle convoy by driving the lead
electric test vehicles in traffic to efficiently achieve the required test speeds and vehicle spacings.
Xll
-------
Glossary
100% Illumination Speed (100%IS) - A geometric characteristic of an RSD instrument in
which a vehicle travelling at the 100%IS speed would produce a signal equivalent to the signal
produced if the scan path would be illuminated once and only once by the RSD light beam.
Air Speed - The speed of a vehicle at 1 meter above the roadway with respect to the air
surrounding the vehicle at a substantial distance from the vehicle.
AirSpeed Para - The scalar air speed component parallel to the direction of vehicle motion at 1
meter above the roadway surface. The sign convention is positive for air moving toward the
windshield.
AirSpeed Perp - The scalar air speed component perpendicular to the direction of vehicle
motion at 1 meter above the roadway surface. The sign convention is positive for air moving
toward the left side of the vehicle.
Data Location Index - The scan or pixel in the data stream that corresponds to a transit event,
such as the passing of the front or rear of a vehicle.
Detailed data - The raw data stream of optical mass species measurements collected by an RSD
at individual pixels.
Drag Area - The product of a vehicle's aerodynamic drag coefficient and the vehicle's frontal
area.
EDAR - Emissions Detection and Reporting, which is the remote sensing instrument
manufactured by HEAT and used in this study.
Emission Rate (g/mile) - The distance-based rate at which a species mass is emitted from a
body in a flow field.
Evaporative HC (EvapHC) - Hydrocarbon gas that is produced by release of liquid gasoline or
gasoline vapor, or by the release of other hydrocarbon vapor from other vehicle materials such as
paint solvents. Evaporative HC specifically does not include exhaust hydrocarbon emissions
from the tailpipe.
Exhaust HC (ExhHC) - Hydrocarbon gas that is produced as a pollutant during combustion of
a fossil fuel and is emitted from a vehicle's tailpipe.
Footprint - The contiguous group of RSD pixels with missing detailed data values that are
produced when the outgoing RSD beam is blocked by the vehicle body.
Frontal area - The area of the silhouette of a vehicle as viewed from the front of the vehicle.
Improved data - Raw data that has been improved through processing to reduce noise, artifacts,
outliers, distortion, and baseline offsets.
Xlll
-------
HEAT - Hager Environmental and Atmospheric Technology, which is the company that
manufactures the EDAR remote sensing device that was used in this study.
Location (Locn) - An emission point on a body in a flow field.
Optical mass (mole/m2) - An RSD's fundamental measured quantity that is proportional to a
species' mass per cross-sectional area of the RSD light beam.
Pixel - A small location where an RSD makes a detailed optical measurement. A scan is made
up of pixels.
Plume - The region in a flow field that contains a material released from a body in a flow field.
Pollutant Conversion Factor - A factor that is used to convert RSD optical values to mass, for
example, molecular weight for gases, or extinction coefficient for particulate material.
Raster-scanning process - A method of spatial scanning that uses motion of a beam across a
field of view.
Release Location Factor - A factor that reflects the Vortex Entrainment Time of an emission
release location relative to the Vortex Entrainment Time of a release from the tailpipe.
Release Rate (g/hr) - The time-based rate at which a species mass is released from a body in a
flow field.
Remote Sensing Device (RSD) - an instrument for measuring pollutants in the air around a
vehicle without touching the vehicle or notifying the vehicle operator.
Remote Sensing Device (RSD) system - an RSD instrument plus associated instrumentation for
defining vehicle, vehicle operation, and ambient conditions including determining vehicle license
plate, road speed, and wind velocity.
Retro-reflective tape - A surface-applied tape that reflects a substantial portion of incident light
back toward the source of the light regardless of the angle of incidence.
Road Direction - The direction (with respect to north) that a vehicle or traffic is moving on a
road.
Road Speed - The speed of a vehicle with respect to the road surface.
RSD Signal (g) - The mass of a species reported by an RSD in the sample of a vortex that the
RSD illuminates.
Scan - A sequential series of pixels produced by an RSD when its light beam is moving in one
direction.
ScanSum - The sum of a pollutant's pixel values for a given scan.
xiv
-------
Species - a material measured by an RSD including gases, mixtures of gases, and particulate
material.
Stripping Rate - The rate at which species are removed from a vortex by air that is passing over
the surface of the vortex.
Surface Roughness - A measure of the irregularities of the terrain in the vicinity of a roadway.
Surface roughness influences the wind speed profile at different heights above a roadway.
Transit - An event in which a vehicle passes by an RSD.
Vehicle reference frame - The coordinate system in which a vehicle is stationary and all things
that are moving with respect to the vehicle are not stationary.
Vortex - the low-pressure zone downstream of a body in a gaseous flow field. The vortex can
act as a temporary storage region for species released from the body.
Vortex Entrainment Time (VET) - A proportionality constant that expresses the ratio of the
mass (g) of a released species in the vortex to the release rate (g/hr) from a body in a flow field.
Wake - The region downstream of the vortex formed by a body in a flow field.
Weights - Spatial factors assigned to spatial coordinates in a vortex that describe the anticipated
relative mass distribution of species emitted from a body in a flow field.
Wind speed and direction - The speed and direction (with respect to north) of wind as
measured by an RSD system.
xv
-------
1.0 Executive Summary
1.1 Research Goals
The broad goal of this research was to find a method to measure on-road evaporative emissions,
known as running losses, of gasoline-fueled vehicles. Running losses are hydrocarbon (HC)
gaseous emissions generated primarily from unintended releases of gasoline vapor or gasoline
liquid from vehicles. For many years, researchers have wanted to quantify running losses, in part
to answer the question: Which on-road fleet emissions are larger, on-road exhaust hydrocarbon
emissions (ExhHC) or on-road evaporative hydrocarbon emissions (EvapHC)? This report
documents the recent in-depth effort to develop a remote sensing device (RSD) method to
measure the running loss emissions of vehicles as they drive in traffic. But the work has
importantly produced an unanticipated capability: an RSD method that can be used to measure
on-road release rates (g/hr) and emission rates (g/mile) of any type of emissions (exhaust,
evaporative, fugitive) for gases and particulate material.
The emission rate (g/mile) of running losses from an individual vehicle is affected by many
vehicle, fuel, operational, and environmental factors including fuel tank capacity and volume,
fuel tank thermal shielding, emission control system malfunctions, evaporative control canister
state, canister purge schedule, fuel level, fuel volatility, fuel oxygenate content, recent driving
pattern, recent soak time, ambient temperature, atmospheric pressure, and the presence of
gasoline liquid and vapor leaks. Because of the numerous factors affecting running loss emission
rate and because evaporative emission control systems operate in a non-linear manner, modeling
running losses is challenging. And even if they could be modelled, without real-world on-road
running loss data, there would not be any real-world data to validate a model.
Nevertheless, for decades running losses could be measured on vehicles in the laboratory. Using
lab methods, the Environmental Protection Agency started certifying light-duty vehicles to meet
a running loss specification beginning with the 1996 model year. The running loss certification
uses a well-defined test condition to determine if a prototype vehicle can meet the standard of 50
mg/mile.
All of the current traditional on-road vehicle RSD technologies (University of Denver, Opus, and
Hager Emissions and Atmospheric Technology) report estimates of exhaust pollutant
concentrations (ppm) or mass of pollutant per mass of fuel (g/gFuel). The instruments use light
beams or lasers to collect detailed optical data around a vehicle as it drives by. The RSDs process
the detailed data using procedures that were developed over 30 years ago. However, vehicle
emissions researchers really want mass emission rates - not concentrations or fuel-based mass
rates. It was thought that release rates and emission rates could not be measured by RSDs. In this
report we show that, using an alternative data processing procedure, the same detailed data that is
used to calculate exhaust concentrations can be used to calculate release rates (g/hr) and emission
rates (g/mile).
For this study, the Colorado Department of Public Health and Environment (CDPHE), the U.S.
Environmental Protection Agency (EPA), and Eastern Research Group (ERG) collected RSD
detailed data on over 30,000 transits in October 2019 in Westminster, Colorado. We embedded a
convoy of electric and gasoline test vehicles in the local traffic while we metered artificial
1-1
-------
evaporative and exhaust emissions. The RSD emission rate method was developed on the test
vehicle detailed data. Then, the method was used to begin characterization of the fleet vehicle
emissions.
1.2 Benefits of the RSD Instrumentation Used in this Study
We chose the Hager Environmental and Atmospheric Technology (HEAT) RSD instrument,
known as Emissions Detection and Reporting (EDAR), to collect the detailed data for this study.
The EDAR instrument is described in more detail in Section 2.2. EDAR has several advantages
for this effort.
EDAR scans the width of a lane of traffic from above the pavement. For each vehicle transit,
EDAR generates thousands of individual detailed optical measurements (pixels). The large
number of individual measurements benefits the signal processing algorithms. Because EDAR
scans from above the vehicle, it sees pollutants in front of, to the side of, as well as, behind the
vehicle. Evaporative emissions originating under the hood are typically first seen at the side of a
moving vehicle.
EDAR uses lasers as the light source. EDAR uses Differential Absorption LIDAR (DIAL) to
collect high signal-to-noise (S/N) detailed optical data at each pixel for pure compounds CO2,
CO, NO, and NO2. EDAR uses a non-DIAL technique, which has a lower S/N, to collect HC
detailed data on the mixture of HC compounds present in vehicle emissions.
For each detailed data pixel measurement, EDAR reports the mass (mole/m2) of the pollutant
illuminated by the laser beam between the RSD instrument and the pavement. Thus, the
combination of the laser beam scanning the full width of the lane from above and the individual
detailed data measurements for all pollutant channels reported as mass provides a reasonably
good optical sample of the mass of emissions released from a moving vehicle.
1.3 Dispersion of Vehicle Pollutant Releases
One key realization made the determination of release rate (g/hr) and emission rate (g/mile) from
RSD detailed data possible: a practical way to quantitatively relate the mass of a pollutant in the
vortex to the release rate of the pollutant from the moving vehicle.
The RSD instrument gets its signal by scanning the laser beam through the air and pollutants
around the moving vehicle. RSD instrument designers have known for years that the largest RSD
signals come from the first few scans behind the vehicle rear. When vehicles move through air,
they create a low-pressure zone just behind the vehicle. The water mist swirling behind the trailer
of an 18-wheeler driving on wet pavement is a vivid demonstration of this process. In this report,
we call this ill-defined, swirling low-pressure zone the vortex.
Any current RSD gets a large part of its signal from pollutants caught up, or temporarily stored,
in the vortex. When pollutants are released from anywhere on a vehicle, they have a good chance
of getting into the vortex, where the RSD can measure a sample of them. Then, as surrounding
air passes over the surface of the vortex, some of the pollutant in the vortex will be stripped away
to be laid down in the air over the roadway.
1-2
-------
The major conceptual problem to be resolved is the units of the measured quantities. The release
rate has units of grams/hour, but the RSD-measured mass of pollutant in the vortex is in grams.
We resolve this seeming units incompatibility by thinking about what happens to the emissions
as they leave the release point on the vehicle. For a given vehicle moving at a constant speed and
having a constant pollutant release rate, a dynamic equilibrium is set up between the release rate
and the mass in the vortex. The rate of pollutant going into the vortex tends to be equal to the
rate that the pollutant is stripped from the vortex by surrounding air. Thus, the pollutant mass in
the vortex (g) will tend to be proportional to the pollutant release rate (g/hr). For example, if the
release rate is zero, there will be zero mass in the vortex. If the release rate is high, the mass in
the vortex will be high. The proportionality constant has units of time. We call it the Vortex
Entrainment Time (VET) with units of hours.
In the study, we released metered flows of different gases (artificial EvapHC and artificial
exhaust mixtures) at various release locations on different light-duty test vehicles while we drove
them under the RSD instrument at different speeds. For each RSD transit of a test vehicle, we
calculated the effective VET by simply dividing the RSD-measured Mass in Vortex (g) by the
metered release rate (g/hr). The results indicated that for light-duty vehicles the VET was
typically around 4 seconds. We also found that the VET was relatively well behaved. The VET
was approximately proportional to the inverse square root of the vehicle's air speed in the
direction of motion, to the one-third root of the vehicle drag area, and to the release location
relative to the rear bumper of the vehicle. The VET is independent of pollutant.
Calculating the emission rate (g/mile) for a transit begins by determining the VET from the
vehicle's air speed (from the vehicle velocity and the wind velocity) in the direction of vehicle
motion, estimating the vehicle drag area, and estimating the relative front-to-back release
location (under the hood = 0.3, rear tailpipe = 1). Then, the RSD-measured Mass in Vortex (g) is
used with the VET and an easily calculated RSD-instrument geometry factor to produce the
release rate (g/hr). Finally, the emission rate (g/mile) is calculated by dividing the release rate
(g/hr) by the vehicle road speed (mile/hr).
1.4 RSD Data Signal Processing
The HEAT EDAR instrument uses the traditional RSD method to calculate its reported exhaust
concentrations. The method compares a transit's detailed optical data for the pollutant of interest
with the corresponding detailed optical data for CO2 taken at the same time along the same
optical path. HEAT has worked out its own signal conditioning methods to de-noise, adjust, and
improve the raw optical data before the calculation of exhaust concentrations.
The new RSD emission rate method, which is described in this report, uses the same detailed
optical data to produce release rates and emission rates - instead of concentrations. Also, instead
of using two pollutant "channels" to calculate each pollutant result, the new method uses only the
pollutant's detailed dataset from the pollutant's one channel. These differences call for the
detailed optical data to be conditioned differently.
For this study we applied standard signal conditioning methods to adjust the constant-level
offset, remove outliers, de-stripe via multi-tonal cancellation, perform adaptive notch filtering,
and interpolate the scanning laser's optical mass measurements to a rectangular grid. See Section
7.1.
1-3
-------
Evaporative hydrocarbon emissions (EvapHC) and exhaust hydrocarbon emissions (ExhHC)
have different chemical compositions and are always released from different locations on a
vehicle. To separately quantify EvapHC and ExhHC emission rates, we applied a signal
processing technique generally known as Blind Source Separation (BSS) to the conditioned
detailed data. The specific technique that we used for the analysis of the Westminster data is
known as Independent Component Analysis1 (ICA). We found that when the EvapHC plume and
the ExhHC plume are not overlapping, as produced by EvapHC from the fuel fill door and
ExhHC from the tailpipe, ICA provides very good results.
However, if plumes overlap substantially, as produced by EvapHC from the fuel tank and
ExhHC from the tailpipe, ICA produces only satisfactory separations. Therefore, for this project,
we began development of a new type of BSS called BSScov, which produces good separations
even if plumes overlap substantially.
1.5 RSD Emission Rate Method Performance
The collection of field data for this project was designed primarily to collect EDAR detailed data
on light-duty test vehicles with metered natural and artificial running losses and exhaust
emissions. Using the test condition variables identified from prior staged data collection efforts,
the test design was planned with wide variations in test conditions so that the test vehicle data
could be used to create an RSD method that could connect metered running loss emission rates
and RSD detailed data measurements. We embedded the test vehicle operation in real traffic (in
Westminster) so that if we could actually develop a method, we could apply it to a sample of a
real-world fleet.
This report documents the fundamentally sound RSD emission rate method that we developed.
However, we found that our ability to fully characterize Westminster's fleet emission rates was
hampered by some EDAR data characteristics that we were unaware of until we analyzed the
Westminster data. We believe that during preparations for the next field data collection effort, we
can work with HEAT to address these issues.
Accordingly, this report is heavy on method development and description and light on fleet
emissions characteristics.
Section 9 compares the method's results for the test vehicles with their metered release rates
(g/hr). Overall, the method's measured exhaust release rates for CO, NO, and CO2 on the test
vehicles had recovery rates between 66 and 87%. Recovery of artificial running losses was linear
with the metered EvapHC release rates and varied between 0 and 100% depending on test
vehicle and release location. The detection limit for a single RSD measurement for the RSD
release rate appears to be about 75 g/hr. Turbulence of the vortex and noise in the RSD HC
detailed data both contribute to this rather high detection limit. We expect that improved noise
reduction techniques will be able to lower the EvapHC detection limit. We would expect that the
1 Jonathon Shlens, "A Tutorial on Independent Component Analysis," https://arxiv.org/abs/1404.2986,
April 14, 2014. This freely downloadable article provides an excellent, intermediate-level discussion of
independent component analysis.
1-4
-------
average release rates of major fractions of fleets could be determined with small errors if the
uncertainties of individual transits are randomly distributed.
Section 10 compares the method's results with model predictions from MOVES as set up for
October 2019 in Westminster. Overall, the method's exhaust release rates were in the same range
as MOVES release rates for CO2, CO, and NO, but HC release rates were high compared to
MOVES values. The relative effects of vehicle age were quite similar between the method's and
MOVES values. The method's low-VSP average values tended to be higher than MOVES
values, and the method's high-VSP average values tended to be lower than MOVES values.
We provide an example transit of a 2001 pick-up truck that happened to drive by the EDAR at
Westminster. The traditional RSD calculation indicated that the vehicle had a low ExhHC
concentration. The RSD emission rate method indicated that the vehicle had massive evaporative
emissions of 8.6 g/mile.
1.6 Areas for Future Development
The RSD emission rate method was developed primarily on the test vehicle EDAR data collected
in Westminster. The method development and its application to the real-world data from the
Westminster fleet sample has pointed out areas where the method needs improvement.
Subsection 9.4 describes sixteen suggested areas to improve the RSD emission rate method in
general and to extend it to medium- and heavy-duty on-road vehicles:
1. Poor Correlations among Exhaust Pollutant Detailed Data
2. RSD Signal Dependence on Laser Pathlength
3. Vortex Entrainment Time
4. Release Location Detection of Light-Duty EvapHC
5. Vortex Shape
6. RSD Signal Accuracy
7. RSD Signal Attenuation
8. Evaporative Plume Signal-to-Noise Improvement
9. Drag Area
10. Enhanced Blind Source Separation
11. Release Location Detection of Medium- and Heavy-Duty Exhaust
12. Diesel Engine Load
13. Particulate Material Pollutant Correction Factor
14. Trailer Configuration Detection
15. Emissions of Vehicles with Trailers
16. Interfering Plumes
1-5
-------
2.0 Testing Description
2.1 Testing Location and Conditions
The testing was conducted on public roadways in Westminster, Colorado, a suburb northwest of
Denver. A Google Map view of the test site is shown in Figure 2-1. The EDAR instrument was
set up on the northeast-bound lane of N. Federal Parkway at approximately 39.91776 N,
105 .02018 W. The 1.48 mile test loop consisted of a rectangle formed by N. Federal Parkway on
the west to the EDAR instalment (0.39 mile), N. Federal Parkway on the north from the EDAR
instrument to N. Zuni Street (0.26 mile), N. Zuni Street on the east (0.33 mile), and W. 120th
Avenue on the south (0.50 mile).
Figure 2-1. Roadways Used for EDAR Testing in Westminster, Colorado
As shown in the figure, speed limits were 50 mph on W. 120th Avenue and 45 mph on N. Federal
Parkway. Test vehicles were driven on the loop in a clockwise direction. At the EDAR
instrument, the roadway has a moderate upward grade.
Colorado routinely monitors the gasoline sold in the state. Tests of gasoline samples collected in
Westminster between 10/8/2019 and 10/24/2019 had an average volatility of 12.0 psi RVP and
9.4 vol% ethanol.
2-1
-------
Figure 2-2 shows the ambient temperatures recorded by the EDAR instrument during vehicle
emissions measurements for the 6 days of testing. Snow caused wet pavement from about 6:00
pm of Wednesday, October 23 through about 11:00 am on Thursday, October 24. EDAR cannot
make emissions measurements with wet pavement.
Figure 2-2. Ambient Temperature at the Test Site During Testing
70
65
U-
0
60
a:
<
Q
111
55
+-!
ra
a)
50
3
ro
<5
45
Q.
E
C
N
1
1
J
•
1
Jv-i
n
1
1
i
<
t'
^ \l
I
1
t
1
¦il
L
i
tf
u-
$
* '•.' \S
\ :
V
/'J
¦ 1
A
1 1
1
f\
?
c
<
*
«
!
f
;:
*
v.': »
%
<
IS
\
1
10/20
10/21
10/22
10/23
2019 Date
10/24
10/25
10/26
2.2 EDAR Configuration
Figure 2-3 shows the HEAT EDAR and associated instrumentation set up on N. Federal Parkway
during the testing in Westminster. The EDAR laser instrument is the box hanging from the
horizontal gantry boom. The approximate infrared laser scanning curtain is drawn in the figure in
red. The instrument scans a 20mm diameter infrared laser at 20 scans per second onto a retro-
reflective tape that is attached to the pavement perpendicular to the direction of traffic flow. The
laser light returns to the instrument for analysis as gases emitted from vehicles absorb a portion
of the light. The instrument used in this study provided HC, CO, NO, NO2, and CO2 optical mass
(moles/m2) measurements for 256 pixels across the 3.66 meter (12 feet) long tape.
The optical window at the bottom of the EDAR box was measured to be 5.3 meters directly
above a point on the reflective tape 0.56 meters from the white line and 3.53 meters from the
closest yellow line.
2-2
-------
Figure 2-3. EDAR Test Set-Up (Looking West) at the Test Site
The horizontal gantry boom is also equipped with a license plate reader that can operate at night
as well as during the day, a weather sensor that measures wind speed and direction, ambient
temperature, relative humidity, and barometric pressure at about 6 meters above the pavement,
and a sensor bar that measures vehicle speed and acceleration when a vehicle passes under the
instrument. The EDAR system instruments and data storage computer were powered by a set of
12-volt deep-cycle batteries, which allowed the EDAR system to be operated unattended for 12-
hour periods.
The EDAR instrument saves scanned optical mass data for a specified number of scans after the
rear bumper of a target vehicle clears the laser beam. As a default, the data from 30 scans after
the rear bumper is saved. If another vehicle is following the target vehicle too closely, the second
vehicle will prevent any of the scans following the target vehicle to be saved. Instead, the scans
behind the second vehicle would be saved. We wanted to collect most fleet vehicle data and half
of the test vehicle test runs with the default 30 scans, but we also wanted to collect some data
with 60 lines to explore the potential benefits of more scans behind vehicles. Figure 2-4 shows
the location of orange traffic cones that we put on the sidewalk to help test vehicle drivers judge
convoy test vehicle following distances.
2-3
-------
Figure 2-4. EDAR Test Site on N. Federal Parkway, Westminster, Colorado
The minimum following distances are given in Table 2-1 for the EDAR instrument scan rate of
20 times per second, the nominal test vehicle operating speeds of 22.5 and 45 mph, and the
minimum scans behind each vehicle of 30 and 60 scans. Because all distances in Table 2-1 are
approximately multiples of 50 feet, we put traffic cones at the 0-, 50-, 100-, and 200-foot
distances as shown in Figure 2-4. The convoy vehicle drivers were able to conveniently judge
their following distances by using the traffic cones as a visual gauge. When the vehicle they were
following was at the 0-foot cone, they had to be no closer than the cone whose distance is given
in Table 2-1.
Table 2-1. Minimum Following Distances for Test Vehicles
Number of After-Rear-Bumper
Scan Lines Set in EDAR
30 scan lines
60 scan lines
Vehicle
Speed
(inph)
22,5
49.5 feet
99 feet
45
99 feet
198 feet
2-4
-------
2.3 Traffic Flow Monitoring Equipment
A camera was used for taking continuous video of cars passing through the monitoring location.
The camera was set up with its focus on vehicle license plates. The camera model was a GoPro
Hero 5 mounted inside a security camera housing to protect the camera and then bolted to a sign
post near the roadside as shown in Figure 2-5. The camera housing can also be seen at the left
margin of Figure 2-3. There was space inside the security housings for a USB battery to power
both the camera and a cooling exhaust fan. The battery powered the camera for approximately 8
hours.
Batteries were charged overnight, and extra batteries were always in reserve for backup or if a
battery could not be charged in time overnight. Two identical cameras and housings were
available to be able to easily swap units, if needed. The GoPro camera could be controlled
wirelessly within 20-30 feet of the unit using an iPad application that allowed for a live view of
what the camera saw, adjustment of settings, and the starting and stopping of image storage. This
application was used to confirm the correct field of view for the security camera housing when
positioning it on its mount to the sign post, and to periodically check the units and ensure that
they were still running and not out of storage space.
Figure 2-5. Video Camera Installed Near EDAR System
2-5
-------
The GoPro Hero 5 model is capable of 12 MegaPixel still images and up to 4K resolution live
video. For this project, live video was made using a resolution slightly below the maximum of
4K to maximize storage space and reduce the heat created by the camera when continuously run.
More importantly was the frames-per-second (fps) rate of filming, which was selected at 90 fps
to capture the highest number of frames possible as vehicles moved past the camera at 20 to 60
miles per hour to reduce blurring and make license plates legible. Very large 500 GB microSD
storage cards were used to store the video as the selected high-resolution, high-fps format
consumed considerable space for each minute recorded. About 500 GB was needed for an 8-hour
period. It was discovered that if the highest possible resolution and an ever faster 120 fps speed
was used then the resulting data writing operation to the storage card would overheat the camera
and cause it to shut down when run continuously for multiple hours. Slightly reducing the filmed
resolution and frame-acquisition rate (fps) solved this problem.
Post-processing of the videos was performed with the "DVMP Pro 7" software product to extract
timestamp metadata from the video files and burn it into each frame of the video. This provided a
running clock and date at the bottom of each video for reviewers to easily reference and find
specific times and vehicles when needed. The DVMP Pro 7 software was relatively slow,
hindered both by the technical limitations of the available computer hardware to run the software
and as well simply by the nature of the process, which is to open and write a timestamp to each
single frame of a high-definition, high-fps video and then write that new frame back to a new
high-definition, high-fps video file, frame by individual frame. Processing all of the captured
video took multiple weeks of continuously running the software. Future project work with
cameras should ensure that a timestamp can be written directly to the video at the time of
filming, which the GoPro Hero 5 model camera used here was not capable of doing.
2.4 Test Vehicles
The EDAR instrument was set up to collect measurements on fleet vehicles. However, at this
point in the development of EDAR for measuring running loss emissions, the connection
between EDAR measurements and an individual vehicle's running loss emission rate (g/mile)
was unknown. Therefore, to help establish that connection, we operated a set of test vehicles
with metered flows of artificial and real exhaust and/or running loss emissions in the traffic on
N. Federal Parkway during the six days of remote sensing testing. The EDAR data collected on
the test vehicles will be used for two purposes:
establishing a connection between EDAR internal-instrument measurements and the metered
running loss emission rate (g/mile) as modified by test conditions such as vehicle speed
and wind speed and direction, and
applying that connection to the test conditions and EDAR instrument-internal measurements
taken on fleet vehicles to estimate the running loss emission rates (g/mile) of the fleet
vehicles.
Table 2-2 gives descriptions of the six test vehicles used in the study. We used two all-electric
vehicles and four light-duty gasoline vehicles. Because the all-electric vehicles carry no gasoline
on board and have no combustion source, their inherent running loss and exhaust emissions are
2-6
-------
zero2. Therefore, by metering artificial running loss and/or exhaust gases, we can know the
composition, concentrations, and release rates of all emissions from these vehicles down to very
low levels. Disadvantages of using all-electric test vehicles are that they are somewhat difficult
to procure, must be regularly charged, which can be time-consuming, must be fitted with a fake
tailpipe that can only approximate the exit position and flow rates of a real exhaust system, and
simulated exhaust gases are most easily released in a dry state, which is different from
combustion emissions, which contain water of combustion. The connection between EDAR
internal-instrument measurements and the metered running loss emission rate (g/mile) would be
developed on data obtained from the all-electric test vehicles.
Table 2-2. Descriptions of Test Vehicles
Vehicle
Desc.
Plate
VIN
Certification
Fuel
Group
Evap
Equipment
Position 1:
2017
Chevrolet
Bolt
Ohio
J595030
1G1FW6S01H4190705
n/a
Electric
n/a
n/a
n/a
Position 2:
2017
Chevrolet
Bolt
Ohio
J595031
1G1FW6S03H4190771
n/a
Electric
n/a
n/a
n/a
Position 3:
2019
Subaru
Outback
Colorado
ABWD21
4S4BSAFC4K3376269
EPA: T3B70
LDV/LDT2
CA:
SULEV30
PC/LDT2
Gasoline
KFJXJ02.5HRV
2.5L
KFJXR01485DX
TWC(2)/
WR-H02S/
H02S/
SFI/
EGR/
EGRC
Position 3:
2019
GMC
1500
Colorado
BXS510
3GTU9DEL2KG154600
EPA: LDT/
Tier3
CA: LDT /
ULEV125
Gasoline
KGMXT06.2375
6.2L
KGMXR017350D
DFI/
H02S/
TWC
Position 3:
2016
Ford
F150
Colorado
ZQO710
1FTFX1EG5GKF11400
EPA: T2B4
LDT4
CA: Not for
sale in CA
Gasoline
GFMXT03.54JG
3.5L
GFMXR023 5NBC
TWC/
DFI/
WR-H02S/
H02S/
TC /
CAC
Position 4:
2015
Infiniti
Q50
Colorado
582ZHP
JN1B V7AR8FM415300
EPA: T2B5
LDV
CA: LEV2-
ULEVPC
Gasoline
FNSXV03.7GAA
3.7L
FNSXR0120MBA
2TWC(2)/
2H02S/
2WR-H02S/
SFI
2 We acknowledge that off-gas evaporative emissions from vehicle materials such as elastomers, paints,
and lubricants are not zero. We assume that those emissions are negligible compared to the levels of gases
that we are artificially releasing from the all-electric reference vehicles.
2-7
-------
The three gasoline test vehicles listed for Position 3 in Table 2-2 were used to overcome the
disadvantages of the all-electric test vehicles, and their results can be used to evaluate algorithms
that predict the running loss emissions of gasoline vehicles. That is, if the application of
connections developed between EDAR internal-instrument measurements and the metered
running loss emission rate (g/mile) of the all-electric test vehicles produce accurate predictions
of the artificial running loss emissions from the gasoline test vehicles, then we would be even
more confident that running loss estimates of fleet vehicles would be accurate.
The two electric vehicles, which were obtained in Aurora, Colorado, were rented from Mike
Albert Rental (mikealbertrental.com) of Cincinnati, Ohio. The Subaru was also a rental vehicle.
The GMC, Ford, and Infiniti test vehicles were personal vehicles of CDPHE staff.
2.5 Test Vehicle Exhaust Emissions Equipment
For the four gasoline test vehicles, the normal exhaust was emitted through the as-equipped
exhaust system. The GMC and Infiniti had dual tailpipes exiting at the bottom edge near the ends
of the rear bumper. The Subaru had a single tailpipe exiting at the left rear. The Ford F150 had a
single tailpipe exiting at the right rear, but it was aimed to the side just behind the right rear
wheel.
Since the EDAR instrument uses signals from the exhaust CO2 to trace the exhaust plume and
calculate exhaust emissions concentrations, we wanted to release artificial CO2 from fake
tailpipes attached to the rear of the two all-electric test vehicles. We attached short pieces of
PVC tubing to the EVs at locations that might be used if those vehicles had gasoline engines. To
be able to distinguish the two EVs from each other in EDAR's infrared plume images, EV-l's
fake tailpipe was installed under the bumper on the left rear end, as shown in Figure 2-6, and
EV-2's was installed under the right end of the rear bumper, as shown in Figure 2-7. Simulated
exhaust gas was routed from a gas cylinder inside each vehicle, through the regulator, an on-off
valve, and finally via H-inch Teflon tubing to the forward end of the PVC tubing, as shown in
Figure 2-7.
We also wanted to use exhaust compositions and release rates that might be observed for
gasoline combustion vehicles. We ordered 7 cylinders for each of two different stoichiometric
exhaust gas mixtures - a clean mixture and a dirty mixture - that came close to satisfying this
equation:
[C02] = 150537.66 - 0.7168 * [CO] - 0.3011 * [HC] - 0.3584 * [NO]
where:
[C02] is the CO2 concentration in ppm,
[CO] is the CO concentration in ppm,
[HC] is the HC concentration in ppmC3, i.e., ppm Propane, and
[NO] is the NO concentration in ppm.
2-8
-------
Figure 2-6, Fake Tailpipe Location on EV-1 Test Vehicle
2-9
-------
The clean and dirty mixtures were ordered in aluminum cylinder size 150A using standard part
numbers used by Envirotest for Colorado I/M testing. The cylinders received had the following
labelled concentrations:
Clean mixture (AirGas Part Number X02NI84T15AC004 ± 2% blend tolerance):
15.05 % CO2, balance N2.
Dirty mixture (AirGas Part Number X05NI84T15AC004 ± 2% blend tolerance):
14.76% CO2, 402 ppm C3H8, 5043 ppm CO, 996 ppm NO, balance N2
The clean mixture was released from test vehicle EV-1. The dirty mixture was released from test
vehicle EV-2.
2.6 Test Vehicle Running Loss Emissions Equipment
The Position 1, 2, and 3 test vehicles, which were the EV-1, the EV-2, and either the Subaru,
GMC, or F150, were set up with equipment to release metered flows of 100% consumer-grade
propane to simulate running loss emissions. After the propane tank and regulator, propane was
routed to a series of three rotameters piped in parallel and then to a 4-way diverter valve. An
example set-up is shown in Figure 2-8. The rotameters were sized for low, medium, and high
flow capacities to cover the wide range of flows needed for propane releases:
Dwyer RMA-150-SSV, 10 to 100 cc/min air (0.021 to 0.21 scfh)
Dwyer RMA-3-SSV, 0.2 to 2.0 scfh air
Dwyer RMA-6-SSV, 2 to 20 scfh air
Teflon tubing from the three outlets of each test vehicle's diverter valve routed the flow of
metered artificial running loss propane to a location at either the fuel fill door (DOOR), the top
of fuel tank (TANK), or under the hood (HOOD). Figure 2-9 shows the tubing outlet on the left
quarter panel of EV-2 to simulate a fuel fill door release location on the opposite side of the
vehicle from the fake tailpipe. Figure 2-6 shows the corresponding simulated fuel fill door
release point for EV-1. Figure 2-10 shows the under-hood release point used for both EV-1 and
EV-2. Figure 2-11 shows the top of tank release point used for the GMC test vehicle.
2-10
-------
Figure 2-8. Rotameters and Diverter Valve for Simulated Running Loss Releases
v \mMWclm * <
DOOR 1/ HOOD
Figure 2-9. Fake Fuel Fill Door Release Point on EV-2 Test Vehicle
2-11
-------
Figure 2-10. Under-Hood Release Point on an EV Test Vehicle
Figure 2-11. Fake Tank Release Point on GMC Test Vehicle
2-12
-------
The eight propane emissions rates (mg/mile) were produced by paired combinations of two test
speeds and four propane release rates (scfh), as shown in Table 2-3.
Table 2-3. Rotameter Settings Required for Designed Propane Releases
Propane Emission
Rate (mg/mile)
Vehicle Speed
(mph)
Propane Release
Rate (scfh)
Propane Rotameter Setting
(air basis)
6400
22.5
2.78
3.43 scfh
3200
45
2.78
3.43 scfh
1600
22.5
0.70
0.86 scfh
800
45
0.70
0.86 scfh
400
22.5
0.174
100 cc/min (0.214 scfh)
200
45
0.174
100 cc/min (0.214 scfh)
100
22.5
0.043
25 cc/min (0.054 scfh)
50
45
0.043
25 cc/min (0.054 scfh)
The propane emissions rate was calculated from the speed and propane release rate using this
equation, with scfh defined at 70 F:
mg Propane = scfh Propane * (460+32) * 28.32 L * 1 mole Propane * 44.10 g Propane * 1000 mg
mile mph Speed (460+70) ft3 22.4 L Propane 1 mole Propane g
Additionally, each test condition had zero-propane release tests interspersed. The purpose of
frequently interlacing zero running loss tests was to collect data to help distinguish non-zero
running loss data streams from those of zero running loss emissions. This will especially assist
the analysis of the data collected at low propane emissions rates (mg/mile) where EDAR's
running loss signal may be hidden in a noisy background.
The propane release rates in the third column of Table 2-3 needed to be converted to settings for
the rotameters, which are calibrated on air. The correction for the specific gravity of the gas
flowing through the rotameters is given by Dwyer (https://www.dwyer-
inst.com/Products/FlowmeterCurves.cfm), whose rotameters were used:
Q2 = Q1 * SQRT( 1 / S.G.)
where:
Q1 = Observed flowmeter reading
Q2 = Actual flow of test gas corrected for specific gravity of test gas
1 = Specific gravity of air, which was used to calibrate flowmeter
S.G. = Specific gravity of test gas used in flowmeter
For example, for propane, which has a specific gravity of 1.52 (=44.1/29), a rotameter setting of
1.23 scfh would be required to produce a propane flow of 1.0 scfh.
2-13
-------
2.7 EDAR Data Collection for Mass Emission Rate Method Development
Here we provide a brief description of how EDAR obtains optical measurements from pollutants
of vehicles operating on the road. While many aspects of how EDAR works are proprietary,
others have been discussed publicly by HEAT.
The EDAR instrument uses laser-based open-path infrared photometry to collect optical
measurements of the gases that surround a vehicle as it passes under the instrument. EDAR scans
a 12-foot-long retro-reflective tape attached to the pavement with a collimated 20mm-diameter
beam of laser light. The laser beam scans the tape back and forth at 10 Hz and therefore obtains
20 scans each second. In this study, the instrument was set up to make measurements at 256
individual points (pixels) during each scan. By comparing the intensity of outgoing light with the
intensity of returning light, the instrument determines the amount of light that is absorbed
between the instrument and the pavement. By selecting appropriate infrared frequencies, the
instrument can make measurements for a variety of gaseous compounds. The EDAR instrument
used in this study collected optical data for measuring the four pure compound pollutants, CO,
NO, NO2, and CO2, and a mixture of pure hydrocarbon (HC) compounds.
A laser technique known as differential absorption LIDAR (DIAL) can be used to get high
signal-to-noise ratio (SNR) infrared absorption signals for small molecules that have small
moments of inertia. Such compounds have infrared rotation-vibration spectra with many sharp
absorption peaks separated by nearby zero-absorption valleys. The DIAL technique uses a single
laser to rapidly oscillate between the peak frequency and the adjacent valley frequency. DIAL
thereby produces a signal that is directly proportional to the amount of the pure compound
present in the optical path. In addition, because the peak and valley frequencies are close, any
interference or noise generally affects absorptions at both frequencies. This makes DIAL, by its
nature, able to reject substantial amounts of noise.
Examples of compounds that can be measured with DIAL include CO, NO, NO2, CO2, methane
(CH4), ethane (C2H6), and ethylene (C2H4). However, larger molecules, like butane (C4H10) and
ethanol (C2H5OH), have larger moments of inertia and have so many possible modes of rotations
and vibrations that the infrared spectra are generally continuous with no or few distinct sharp
peaks and valleys. Thus, DIAL cannot generally be used to obtain signals from larger molecules.
For larger molecules, regular non-DIAL absorption techniques can be used, but the SNRs of such
measurements can be hundreds of times poorer than DIAL techniques. Thus, the EDAR
instrument uses the DIAL technique for CO, NO, NO2, and CO2, but the less sensitive standard
absorption technique for the mixture of HC compounds.
The EDAR instrument uses five lasers to measure optical absorptions and store five channels of
data - in this study, one channel for each of CO, NO, NO2, CO2, and the HC mixture. The data
for each channel typically consists of an array of individual optical mass measurements at each
pixel from 10 scans (0.5 s) in front of the front bumper to 30 scans (1.5 s) behind the rear bumper
of a vehicle. The total number of scans in the array depends on the speed of the vehicle. For
example, a 15-foot vehicle moving at 30 ft/s would cause the array to have an additional 5 scans.
In that case, the complete array would have 45 scans made up of 10 scans before, 5 scans during,
and 15 scans after the vehicle transit. Since each scan has 256 pixels, the complete array for one
channel would have 11,520 optical mass measurements. The optical mass measured for each
pixel is reported in units of mole/m2, which means moles of the compound being measured per
2-14
-------
square meter of the laser beam. It is important to recognize that the optical mass measurements
are not for the entire emissions plume but only for the 20mm wide zigzag swath of the plume
that the laser beam illuminates as the vehicle transits.
In usual EDAR operation while measuring specifically for exhaust emissions, EDAR does not
routinely save the arrays for the five channels. EDAR just uses the data in the arrays to calculate,
save, and output the exhaust emissions concentrations (ppm) or fuel-based emission rate (g/kg
fuel). The arrays are not routinely saved. But for this study, we asked HEAT to save all arrays so
that we could analyze them. Basically, we want to find an algorithm that uses the same array
data, which is used to calculate exhaust emissions, to additionally calculate Release Rates (g/hr)
and Emission Rates (g/mile) of exhaust emissions and evaporative emissions.
2.8 Field Data Handling and Storage
After the end of field data collection, HEAT provided ERG and CDPHE data for each transit that
EDAR had recorded during the field deployment. The data included the variables on the left side
of Table 2-4 and still photographs3 of vehicle license plates. In addition, HEAT used their
license plate transcriptions to look up variables for each Colorado-registered vehicle in a
snapshot of the Colorado registration database. Those variables are shown in the top right of
Table 2-4. ERG transcribed the hand-written notes from the paper data packets that were filled
out by the personnel in each of the test vehicles corresponding to each of the convoy vehicle test
conditions. Those variables, which are specific to the test vehicle test runs, are shown in the
lower right of Table 2-4.
We then time-aligned the data provided by HEAT with the data from the transcribed data packets
so that the EDAR results for each test vehicle transit could be easily found for analysis.
Additional flag variables were added to the final spreadsheet4 for sorting and analysis purposes.
For analysis of the data by SAS, a CSV version of the spreadsheet was read by a SAS program5
and merged with decoded information6 from the ERG VIN decoder to create a final SAS
dataset7.
HEAT also provided ERG with the EDAR data arrays8 for each transit and for each of the five
EDAR pollutant channels (HC, CO, NO, NO2, CO2).
3 P:\EDARinDenver-OCT2019\HEATphotos\All_33074_EDAR_snapshots-OCT2019/* .jpg
4 P:\EDARinDenver-OCT2019\EDARpngs_Denver_20_24OCT2019-200120/
Wcstminstcr_OCT2019Rcsults_200124Rcproccss-200203_\\RclcascRatcs_gph.xlsx
5 P:\EDARinDenver-OCT2019\Analysis/ read_SS_VIN.sas
6 P:\EDARinDenver-OCT2019\Analysis/ vin_output.csv
7 P:\EDARinDenver-OCT2019\Analysis/ Westminster_ss_vin.sas7bdat
8 P :\EDARinDenver-OCT2019\EDAR_CSVs\OriginalCSVs-received200124\
7_2019102?_*_0005??_Denver_2019/ *_array.csv
2-15
-------
Table 2-4. Reported Variables in Master Dataset
Source
Variable
Source
Variable
HEAT
EDAR Date MTN (mm/dd/yyyy)
CO Regis
VIN
HEAT
EDAR Time MTN (hh:mm:ss AM/PM)
CO Regis
Model Year
HEAT
EDAR License Plate Number
CO Regis
Make
HEAT
EDAR License Plate State
CO Regis
Model
HEAT
EDAR Vehicle Speed (mph)
CO Regis
Fuel Type
HEAT
EDAR Road Grade (rise/run)
CO Regis
Body Type
HEAT
EDAR Latitude (deg)
CO Regis
Vehicle Type
HEAT
EDAR Longitude (deg)
CO Regis
Emissions Expiration Date
HEAT
EDAR Ambient Temperature (F)
CO Regis
Emissions Area
HEAT
EDAR Relative Humidity (%)
CO Regis
Registration Date
HEAT
EDAR Barometric Pressure (inch Hg)
CO Regis
Registration County
HEAT
EDAR Wind Speed @ 6m (mph)
HEAT
EDAR Wind Direction (degN)
ERG
Test Vehicle ID
HEAT
EDAR Vehicle Acceleration (mph/s)
ERG
GMC TailGate
HEAT
EDAR Vehicle Specific Power (kW/Mg)
ERG
Run No.
HEAT
EDAR Epoch Car Time (micro s)
ERG
Evap Location
HEAT
EDAR Car Name
ERG
Nominal Speed (mph)
HEAT
EDAR HC Mole Ratio (moleC6/moleC02)
ERG
Nominal Propane Emission Rate (mg/mile)
HEAT
EDAR NO Mole Ratio (moleNO/moleC02)
ERG
Measured Release Rate (g/hr)
HEAT
EDAR CO Mole Ratio (moleCO/moleC02)
ERG
Exhaust Gas Release Volume (scf)
HEAT
EDAR HC (ppmC6)
ERG
Labeled Exhaust Cylinder HC (ppmC3)
HEAT
EDAR CO (%)
ERG
Labeled Exhaust Cylinder CO (ppm)
HEAT
EDAR NO (ppm)
ERG
Labeled Exhaust Cylinder NO (ppm)
HEAT
EDAR C02 (%)
ERG
Labeled Exhaust Cylinder C02 (%)
HEAT
EDAR Clean Screened?
ERG
Field Notes QC
HEAT
EDAR QC
ERG
Test Vehicle Run Quality Flag
2.9 Westminster Dataset EDAR Quality Flag
The EDAR instrument produces the EDAR QC flag, which is listed at the bottom of the left
column of Table 2-4. The EDAR QC flag assigns one of four values to each vehicle transit.
"Interfering plume" is assigned if the instrument detects substantial amounts of pollutants in
front of the vehicle. The source for an interfering plume could be from emissions of a vehicle
driving in front of the target vehicle or from a vehicle in the oncoming lane. "Low C02" is
assigned when the size of the CO2 plume is small. This can occur if the driver takes his foot off
the accelerator while passing under the EDAR instrument. "No plate" is assigned if the vehicle
has no discernable license plate. Otherwise, the EDAR QC flag is set to "valid."
Table 2-5 shows the counts of the test vehicle transits that met the planned test condition criteria
and for the fleet vehicles as a whole. Because the test vehicle convoy scrupulously maintained
2-16
-------
specified minimum following distances, interfering plumes should only be generated by fleet
vehicles in the oncoming lane. On the other hand, tailgating was common for fleet vehicles. Test
vehicles EV-1 and EV-2 show relatively high counts of Low CO2 flags compared to the other
test vehicles. We believe that this is a consequence of the 30 scfm release rate of artificial
exhaust from EV-1 and EV-2 which is low compared to the likely higher release rate of the other
four test vehicles that had natural exhaust emissions releases. The fleet vehicle counts of Low
CO2 flags probably occurred when some drivers saw the RSD equipment and took pressure off
the accelerator pedal.
Table 2-5. Transit Counts by EDAR QC Label and Vehicle Category
EDAR QC
Label
Test Vehicles
Fleet
Vehicles
EV-1
EV-2
Subaru
F150
GMC
Infiniti
Valid
236
282
103
86
15
290
25544
Interfering Plume
8
6
14
5
0
5
2302
Low C02
59
19
0
0
0
2
2833
No Plate
0
2
1
0
44
2
1084
Total
303
309
118
91
59
299
31763
2.10 Westminster Dataset and Analysis Program Locations
Appendix B gives the locations of the Westminster datasets and analysis programs with details of
the inputs and outputs of each program. This information can be used to help the analysis of
future RSD data.
2-17
-------
3.0 Field Data Collection Results
During the field event, EDAR collected data on on-road private vehicles, on the imbedded test
vehicles that we drove as a convoy interspersed in normal traffic, and on weather for each EDAR
vehicle transit while the emissions measurements were being taken. In the following subsections,
we discuss the test vehicle test conditions, the characteristics of the weather during the study, and
the optical data collected by the EDAR instrument.
3.1 Test Vehicle Test Conditions
We drove a group of study test vehicles repeatedly past the RSD instrument on each day of field
testing. The test vehicles, which served as RSD measurement controls in the study, were
imbedded in normal traffic flow as a four-vehicle9 "convoy." Table 3-1 gives the purpose,
characteristics, and emissions releases of the vehicles that were convoy members. The test crew
drove the convoy under the EDAR instrument at the designed test conditions while trying to
prevent any public vehicles from getting between the individual test vehicles. Test vehicle
drivers did not allow "interlopers" to sneak in to maintain adequate following distances so that
the EDAR instrument would properly "trigger" on test vehicles in Position 1, 2, and 3. Vehicle 4
was used solely as a "blocker" to prevent public vehicles from tailgating Vehicle 3.
Occasionally, aggressive interlopers did force into the convoy. In those instances, the entire data
collection run was aborted, all data was marked for deletion, and the test condition was repeated
on the next transit.
On any given test run, all vehicles in the convoy drove past the EDAR instrument at the same
nominal speed - either 22.5 or 45 mph. Vehicle 1 released "clean" artificial exhaust, while
Vehicle 2 released "dirty" artificial exhaust, as shown in the table. The artificial exhaust was
released at 30 scfm10 for about 10 seconds before and during each test vehicle's transit under the
EDAR instrument. The exhaust from vehicles in Positions 3 and 4 were their natural exhaust.
On any given test run, Vehicles 1, 2, and 3 released propane, as the artificial running loss
emissions, at the same artificial running loss release rate and release location. Of course, since
Vehicles 1 and 2 were EVs, their propane releases were their only running losses. On the other
hand, since Vehicle 3 was always a gasoline-fueled vehicle, its propane releases were in addition
to any natural running losses that Vehicle 3 might have.
As shown in Table 3-1, nine non-zero propane release rates (288, 144, 72, 36, 18, 9, 4.5, 2.25,
and 1.125 g/hr) were used. Some tests were also performed with no propane released (0 g/hr).
With the two nominal speeds (22.5 and 45 mph), these ten release rates produced nine nominal
running loss emission rates (6400, 3200, 1600, 800, 400, 200, 100, 50, and 0 mg/mile).
9 The third vehicle position was not filled on the 10/24/2019 test day. Therefore, the convoy had only
three vehicles on that day.
10 Standard (70°F, 760 Torr) cubic feet per minute
3-1
-------
Table 3-1. Vehicles Used in the Test Vehicle Convoy
Vehicle Position
(Test Dates) Purpose
Vehicle
Description
Nominal
Vehicle
Speed
(mile/hr)
Exhaust
Mixture and
Flow
Propane:
Artificial
Running Loss
Release Rate
(g/hr)
Propane:
Artificial
Running Loss
Release
Location
1
(10/20-24/2019)
Clean exhaust,
0 natural running loss
EV-1: 2017
Chevrolet Bolt
22.5 mph,
45 mph
Artificial:
15.05% C02
balance N2
at 30 scfm
(left exhaust)
Artificial only:
288, 144,
72, 36,
18, 9,
5,3,
Fuel Fill
DOOR,
Top of TANK,
Under HOOD
2
(10/20-24/2019)
Dirty exhaust,
0 natural running loss
EV-2: 2017
Chevrolet Bolt
Artificial:
402 ppm
C3H8
5043 ppm CO
996 ppm NO
14.76% C02
balance N2
at 30 scfm
(right exhaust)
1,0
3
(10/20-21/2019)
Natural exhaust
2019
Subaru Outback
Natural
(left exhaust)
Natural +
Artificial:
288, 144,
3
(10/22/2019)
Natural exhaust
2019
GMC 1500
Natural
(dual exhaust)
72, 36,
18, 9,
5,3,
1,0
3
(10/23/2019)
Natural exhaust
2016
Ford F150
Natural
(right exhaust)
4
(10/20-24/2019)
Blocker
2015
Infiniti Q50
Natural
(dual exhaust)
Natural
Natural
3-2
-------
The total 298 realized combinations of metered artificial propane release rates and nominal
speeds for the test vehicle convoy are described in Table 3-2. During each transit, the propane
releases of Vehicles 1, 2, and 3 were all set to the same release location and rate.
Table 3-2. Test Condition Combinations Used by the Test Vehicle Convoy11
Propane: Metered
Artificial
Running Loss
Release Rate
(g/hr)
Nominal
Vehicle
Speed
(mile/hr)
Propane:
Nominal
Artificial
Running Loss
Emission Rate
(mg/mile)
Number of Convoy Transits
Release
Location:
DOOR
Release
Location:
TANK
Release
Location:
HOOD
Total
288
45
6400
5
5
5
15
144
22.5
6400
5
6
5
16
144
45
3200
5
6
5
16
72
22.5
3200
5
5
5
15
72
45
1600
5
5
5
15
36
22.5
1600
5
6
5
16
36
45
800
5
6
5
16
18
22.5
800
5
5
5
15
18
45
400
5
5
5
15
9
22.5
400
5
6
5
16
9
45
200
5
5
5
15
4.5
22.5
200
5
5
5
15
4.5
45
100
5
5
5
15
2.25
22.5
100
5
6
5
16
2.25
45
50
5
6
5
16
1.125
22.5
50
5
5
5
15
0
45
0
n/a
n/a
n/a
25
0
22.5
0
n/a
n/a
n/a
26
As the table shows, during the study at least five replicates were obtained for each test condition.
The convoy was driven to achieve this minimum number of replicates. That is, if some feature of
a convoy run did not meet quality assurance criteria, for example, if a following distance was too
short or a gas valve was not in the proper position on one vehicle, then the transit for the entire
convoy for the needed test condition was repeated. In general, each of the five or six replicates
was performed on a different day of the testing to reduce the risk of an imbalanced dataset as a
11 C:\Documents\EPA WA3-13 (MAR20-FEB21)\8_Reports/Table_ConvoyTestConditions.xlsx
3-3
-------
consequence of a pause in or termination of data collection caused by inclement weather or
equipment malfunctions. Thus, at least one replicate of the entire set of test conditions in Table
3-2 was completed during each test day.
In addition to the 298 convoy transits described by Table 3-2, eleven more convoy transits were
made with EV-1 and EV-2 not releasing either simulated exhaust gas or simulated running loss
gas. These runs provided EDAR instrument-internal data that could be used to characterize
instrument noise when no vehicle emissions are present. During these "blank" runs, the test
vehicles in Positions 3 and 4 were operated with their natural exhaust emissions and no artificial
running losses released.
3.2 Model Years and Gross Vehicle Weight Ratings of Fleet Vehicles
The characteristics of the fleet vehicles were examined12 using the license plates, Colorado
vehicle registration database, and the ERG VIN decoder. Because of various idiosyncrasies, the
counts of transits and vehicles presented in this subsection should be regarded as approximate.
In general, during the October 2019 Westminster field study, the HEAT RSD instrument and its
license plate reader operated day and night. 30,590 RSD transits of fleet vehicles (i.e., not test
vehicle transits) were obtained by the RSD instrument. After eliminating missing license plates
(N=973) and "NOREAD" plates (N=l 1), the license plate reader had recorded 18,547 unique
plates. Because the plate reader occasionally reads plates improperly, the actual number of
unique vehicles would be slightly lower.
Of those 18,547 unique plates, 17,927 vehicles had Colorado plates. We used the ERG VIN
Decoder to determine the GVWRs of the vehicles, where possible. The GVWR distribution is
shown in Table 3-3. We also used the ERG VIN Decoder and a snapshot of the Colorado
registration database to determine, where possible, the consensus model years of the vehicles
with Colorado plates. The model year distribution is shown in Table 3-4.
Table 3-3. Gross Vehicle Weight Ratings of Vehicles with Colorado Plates
GVWR Bin
(pounds)
Frequency
Percent
Cumulative
Frequency
Cumulative
Percent
LDV
8113
45.26
8113
45.26
0-3,750
212
1.18
8325
46.44
3,751-6,000
5029
28.05
13354
74.49
6,001-8,500
2719
15.17
16073
89.66
8,501-10,000
326
1.82
16399
91.48
10,001-14,000
74
0.41
16473
91.89
14,001-19,500
5
0.03
16478
91.92
Unknown
1449
8.08
17927
100
12 P:\EDARinDenver-OCT2019\Analysis_MLout\220817\Anal_MLout\FleetVehs/ OCT19_FleetStats.sas
3-4
-------
Table 3-4. Model Years of Vehicles with Colorado Plates
Consensus Model
Frequency
Percent
Cumulative
Cumulative
Year
Frequency
Percent
1947
1
0.01
1
0.01
1966
2
0.01
3
0.02
1967
1
0.01
4
0.03
1970
1
0.01
5
0.03
1973
2
0.01
7
0.04
1974
1
0.01
8
0.05
1978
2
0.01
10
0.06
1981
1
0.01
11
0.07
1983
1
0.01
12
0.08
1984
6
0.04
18
0.11
1985
2
0.01
20
0.13
1986
4
0.03
24
0.15
1987
6
0.04
30
0.19
1988
4
0.03
34
0.22
1989
6
0.04
40
0.25
1990
11
0.07
51
0.32
1991
21
0.13
72
0.46
1992
17
0.11
89
0.56
1993
20
0.13
109
0.69
1994
55
0.35
164
1.04
1995
62
0.39
226
1.43
1996
61
0.39
287
1.82
1997
114
0.72
401
2.54
1998
137
0.87
538
3.4
1999
204
1.29
742
4.7
2000
265
1.68
1007
6.37
2001
298
1.89
1305
8.26
2002
346
2.19
1651
10.45
2003
401
2.54
2052
12.99
2004
512
3.24
2564
16.23
2005
545
3.45
3109
19.68
2006
586
3.71
3695
23.38
2007
723
4.58
4418
27.96
2008
748
4.73
5166
32.69
2009
498
3.15
5664
35.85
2010
648
4.1
6312
39.95
2011
799
5.06
7111
45
2012
885
5.6
7996
50.6
2013
1027
6.5
9023
57.1
2014
1194
7.56
10217
64.66
2015
1395
8.83
11612
73.49
2016
1308
8.28
12920
81.77
2017
1246
7.89
14166
89.65
2018
1148
7.27
15314
96.92
2019
482
3.05
15796
99.97
2020
5
0.03
15801
100
missing
2126
3-5
-------
3.3 Wind Speed and Direction
Wind speed and direction measurements are important for the determination of vehicle Release
Rates (g/hr) and Emission Rates (g/mile) from RSD measurements. The reason for this is that, in
addition to vehicle road speed, wind speed and direction influence the dispersion of emissions
from vehicles driving on the road and thereby influence the dimensions and location of the
vortex behind each moving vehicle. Because RSDs get their largest signals from pollutants in the
vortex, the factors that influence the vortex affect the optical pathlengths and size of emissions
plumes sampled by RSDs.
The EDAR instrument was equipped with a weather sensor that measured wind speed and
direction, ambient temperature, relative humidity, and barometric pressure at about 6 meters
above the pavement. At each vehicle transit, the EDAR system recorded those variables.
Accordingly, because of diurnal differences in traffic flow, weather measurements are frequent
during the day and infrequent at night. Figure 3-1 shows a plot of the EDAR-reported wind
speed and direction as measured at the 6-meter height of the weather sensor above the pavement.
Each point represents a measurement taken at a vehicle transit. Figures 3-2 and 3-3 show
separate wind direction and wind speed histograms of the same dataset. The winds tended to be
high-speed and gusty during the week.
Figure 3-1. Wind Speed vs. Wind Direction at 6 Meters above Pavement13
0 15 30 45 60 75 90 105 120 135 150 165 180 195 210 225 240 255 270 285 300 315 330 345 360
Wind Direction (degN)
/projl/EDARinDenver-OCT2019/Analysis/analyze_FIeet_Wind_1.sas 06MAY2017:28
13 C:\Documents\EPAWA3-13 (MAR20-FEB21)\8_Reports/
Wcstminstcr_OCT2019Rcsults_200124Rcproccss-200203_\\ Release Ratcs_gph.xlsx. Wind Plots tab
3-6
-------
Figure 3-2. Wind Direction at 6 Meters above Pavement
105
113
120
128
135
143
150
158
165
173
180
188
195
203
210
218
225
233
240
248
255
263
270
278
285
293
300
308
315
323
330
338
345
353
CUM.
CUM.
FR§ft
FR§ft
P1%
P1%
346
672
1.09
2.12
375
1047
1.18
3.30
416
1463
1.31
4.61
432
1895
1.36
5.97
538
2433
1.69
7.66
1017
3450
3.20
10.86
552
4002
1.74
12.60
294
4296
0.93
13.53
294
4590
0.93
14.45
403
4993
1.27
15.72
348
5341
1.10
16.82
398
5739
1.25
18.07
504
6243
1.59
19.65
847
7090
2.67
22.32
893
7983
2.81
25.13
700
8683
2.20
27.34
580
9263
1.83
29.16
674
9937
2.12
31.28
657
10594
2.07
33.35
533
11127
1.68
35.03
451
11578
1.42
36.45
424
12002
1.33
37.79
466
12468
1.47
39.25
370
12838
1.16
40.42
739
13577
2.33
42.74
864
14441
2.72
45.46
863
15304
2.72
48.18
645
15949
2.03
50.21
590
16539
1.86
52.07
667
17206
2.10
54.17
847
18053
2.67
56.84
1312
19365
4.13
60.97
1754
21119
5.52
66.49
2187
23306
6.89
73.37
1699
25005
5.35
78.72
1229
26234
3.87
82.59
1183
27417
3.72
86.32
941
28358
2.96
89.28
870
29228
2.74
92.02
705
29933
2.22
94.24
614
30547
1.93
96.17
389
30936
1.22
97.40
283
31219
0.89
98.29
162
31381
0.51
98.80
147
31528
0.46
99.26
235
31763
0.74
100.00
FREQUENCY
Figure 3-3. Wind Speed at 6 Meters above Pavement
CUM.
CUM.
FB
P6%
P6%
2440
4496
7.68
14.15
1980
6476
6,23
20.39
2647
9123
8.33
28.72
2233
11356
7.03
35.75
2299
13655
7,24
42.99
1686
15341
5,31
48.30
1275
16616
4.01
52.31
1648
18264
5.19
57.50
1208
19472
3.80
61.30
1543
21015
4.86
66.16
1060
22075
3.34
69.50
1401
23476
4.41
73.91
1124
24600
3.54
77.45
1249
25849
3.93
81.38
757
26606
2.38
83.76
1023
27629
3.22
86.98
676
28305
2.13
89.11
735
29040
2.31
91.43
407
29447
1.28
92.71
481
29928
1,51
94.22
299
30227
0.94
95.16
341
30568
1.07
96.24
203
30771
0.64
96.88
198
30969
0,62
97.50
193
31162
0.61
98.11
118
31280
0.37
98.48
106
31
386
0.33
98.81
103
31
489
0.32
99.14
82
31
571
0.26
99.40
47
31
618
0.15
99.54
31
31
649
0.10
99.64
31
31
680
0.10
99.74
22
31
702
0.07
99.81
15
31
717
0.05
99,86
12
31
729
0.04
99.89
8
31
737
0.03
99.92
6
31
743
0.02
99.94
5
31
748
0.02
99.95
2
31
750
0.01
99.96
3
31
753
0.01
99.97
4
31
757
0.01
99.98
3
31
760
0,01
99.99
1
31
761
0.00
99.99
0
31
761
0.00
99.99
1
31
762
0.00
00.00
0
31
762
0.00
00.00
1
31
763
0.00
00.00
0
31
763
0.00
00.00
0
31763
0.00 100.00
FREQUENCY
/proj1/EDARinDenver-OCT2019/Analysis/analyze_Fleet_Wind_1 .sas 06MAY20 17:28
3-7
-------
4.0 Detailed Data Post-Processing by HEAT
During planning of the field testing, ERG anticipated using raw, unprocessed data - just as it was
obtained by download from the instrument. The reason was that we expected that the signal
analysis methods that we would develop for maximizing the signal-to-noise ratio for running loss
emissions would be different from those used routinely by HEAT for reporting exhaust
emissions. Also, we wanted to have the raw data so that any signal processing that we developed
could be used to write a processing algorithm that could be installed on the instrument for in-
real-time processing in future studies.
We had discussions with HEAT to identify the different post-processing actions that HEAT used
to determine the actions that were acceptable. Our main concern was that we did not want to lose
data or to do something that would harm or mask information in the data. On the other hand, we
did want HEAT to post-process for things that we did not want to spend a substantial amount of
time to "re-invent." The following paragraphs describe the raw data post-processing activities
that HEAT performed before conveying the data files to ERG.
Background Corrections - HEAT routinely corrects the background of the CO2 arrays for
absorption by the ambient 400 ppm CO2 levels. This correction amounts to about 20% of the
total CO2 absorption from vehicle exhaust. Additionally, because the EDAR pathlength varies
with scan angle, the absorption varies with scan angle, as can be seen in Figure 2-3. HEAT has a
proprietary method to make the correction. Therefore, we decided to have HEAT continue to
make the correction to the CO2 arrays. We believe that HEAT makes corrections only for the
global ambient CO2 and not for CO2 or other vehicle pollutants that are above the roadway from
previous vehicles driving past the instrument.
Scan Position Corrections - The EDAR instrument used in this study had 256 different scan
positions along each scan. An optical measurement for each channel is taken at each scan
position. For a variety of reasons, including wear on the retro-reflective strip, the recorded
absorption among the scan positions can vary systematically. HEAT uses the average optical
measurement for each scan position in the last several scans to adjust all of the optical
measurements for the corresponding scan position for the entire transit's array. Since this is a
relatively simple calculation, we judged that the risk of damaging the underlying optical
measurements was low. Therefore, we agreed to have HEAT make these corrections.
Vehicle Footprint Blanking - During a vehicle transit, EDAR's outgoing beam is not reflected
back to the instrument from scan positions during scans when the vehicle is covering the retro-
reflective tape. Without correction, the instrument would normally calculate a 100% absorption
for those pixels, even though those absorptions would not represent high pollutant
concentrations. Also, when the 20mm diameter beam is partially obscured by the edge of the
vehicle, the resulting measurement also does not correspond to an elevated pollutant
concentration. Therefore, for both of these situations, which occur on the transit of every vehicle,
HEAT must process the raw optical measurements to blank out those array pixels that
completely or partially blocked by the vehicle.
To make its blanking code simple and reliable, HEAT wrote an algorithm that circumscribes a
perfect rectangle around the vehicle footprint. The disadvantage of circumscribing is that the
4-1
-------
optical measurements for some pixels - particularly near the corners of vehicles - are not output
to the final processed array. This loss of data is not a problem for calculating exhaust emissions,
but we did not want to lose any data - especially since the largest running loss absorptions are
likely near the edges and corners of the vehicle footprint. Therefore, we asked HEAT to process
the raw data without circumscribing the vehicle footprint.
Pixel Glitches and Footprint Glints - Occasionally, arrays contain isolated pixels with optical
mass values that are substantially different from the values of adjacent pixels. The optical mass
values are more different than can be expected from the usual background noise. If the pixels are
adjacent to the vehicle footprint, the values might be caused by glints as the laser beam scans the
edge of the vehicle body. Isolated, spurious optical mass values can also occur anywhere in an
array. We refer to these as pixel glitches. Whether these abnormal values are glints or glitches,
neither HEAT nor ERG believe that they represent good values. HEAT has methods to identify,
remove, smooth, or otherwise handle such suspect optical mass values. We asked HEAT to
convey all suspect values to us unchanged so that we could develop methods to identify them
and prevent them from adversely influencing running loss calculations. This approach gives us
multiple opportunities to develop ever-improving signal analysis techniques.
4-2
-------
5.0 RSD Emission Rate Method: Step-by-Step Description
The measurement of pollutant emissions rates by remote sensing devices (RSDs) is based on
three concepts observed in our studies:
• A portion of pollutants released from a moving vehicle become temporarily entrained in
the vehicle's low-pressure zone, or vortex, which is the dominant source of RSD signals.
• Under steady-state vehicle air velocity and pollutant release rate, the pollutant Mass in
Vortex (g) is relatively constant and is proportional to pollutant Release Rate (g/hr).
• The proportionality constant, which we call the Vortex Entrainment Time (hr), depends
only mildly on vehicle air speed, pollutant release location, and light-duty vehicle shape.
The pollutant Mass in Vortex (g) is calculated from the RSD signal with corrections for road
speed and a geometrical factor characteristic of the RSD configuration. The Vortex Entrainment
Time (VET) is estimated from the vehicle air velocity, the estimated vehicle drag area, and the
estimated source location on the vehicle. Then, the pollutant Release Rate (g/hr) is calculated as
the Mass in Vortex (g) divided by the VET (hr). Finally, the pollutant Emission Rate (g/mile) is
just the Release Rate (g/hr) divided by the vehicle Road Speed (mile/hr).
The calculations also use the discovered dependence of vortex shape on vehicle air velocity and
vehicle length, as well as standard signal analysis techniques, to improve RSD signal-to-noise
ratio. Finally, Blind Source Separation is used to apportion the RSD HC signal into an Exhaust
HC signal and an Evaporative HC signal, from which their separate release rates and emission
rates are determined.
All of the above are presented in this section using a recipe-like description of the calculations.
Following sections detail the analyses for the determination of VET functionality and vortex
shape functionality.
Overview of the Method's Flow of Calculations - The method for calculating vehicle
Emission Rates (g/mile) and Release Rates (g/hr) from RSD measurements flows from left to
right using the elements shown in Figure 5-1:
• A Remote Sensing Device (RSD), described in Section 5.1, collects measurements of the
species emitted from a moving vehicle as it passes by the RSD system,
• A Pre-Processing Device (PPD), described in Section 5.2, adjusts these measurements to
improve their overall quality and reduce noise effects,
• A Vortex Shape Estimation Device (VSED), described in Section 5.4, calculates
Weights, which describe the vortex shape, and the Vortex Entrainment Time (VET),
which is critical to the estimation of emission rates, and
• An Emission Calculation Device (ECD), described in Section 5.5, uses one or more of
the outputs of the Pre-Processing Device to calculate the Release Rate and the Emission
Rate of one or more emitted species from the vehicle.
5-1
-------
An alternative method is shown in Figure 5-2. This method contains the same elements as in
Figure 5-1 and adds an optional new element: a combined Separation/Estimation (SED) Device,
described in Section 5.3. The calculation extracts components of the measured species signals to
associate them with one or more different plume sources: exhaust plume and evaporative plume.
This allows the estimation of emission rates of specific species that are associated with specific
spatial locations around the vehicle.
Figure 5-1. Flow Diagram of Methodology
without Separation/Estimation of Emission Sources
Measured
Data
Improved
Data
Emission
Rates
Figure 5-2. Flow Diagram of Methodology
with Separation/Estimation of Emission Sources
Remote
Sensing
Device
Measured
Data
HC
CQ2
CO
NO
NQ2
Improved
Data
Pre-
processing
Device
HC
CQ2.
CO
NO
NQ2,
Environmental Measurements
RSD Properties
Emission
Components
Separation/
Estimation
Device
t Weig
ExhHC.
EvapHC,
HxhCQ2.
ExhNO.
its / Locns
Vortex Shape
Calculation
Device
lT
Emission
Calculation
Device
:FTT
VET
Emission
Rates
ExhHC
EvapHC
ExhCQ2
ExhNO
5.1 Remote Sensing Device
The Remote Sensing Device (RSD) is a system that collects spatial and/or temporal
representations of one or more species that are potentially emitted from a moving vehicle as it
passes by the RSD without touching the vehicle.
5-2
-------
Figure 5-3 shows the Hager Environmental and Atmospheric Technology (HEAT) RSD and
associated instrumentation that was used to collect data. The RSD laser instrument (labelled
EDAR) is the box hanging from the horizontal gantry boom. The approximate location of the
path of the scanning infrared laser is drawn in the figure in red. The instrument scans a 20mm
diameter infrared laser beam at 20 scans per second onto a retro-reflective tape that is attached to
the pavement perpendicular to the direction of traffic flow. The laser light returns to the
instrument for analysis as gases emitted from vehicles absorb a portion of the light. The
instrument shown provides HC, CO, NO, NO2, and CO2 optical mass (moles/m2) measurements
for 256 pixels across the 3.66 meter (12 feet) long tape.
The horizontal gantry boom is also equipped with a license plate reader, a weather sensor that
measures wind speed and direction at about 6 meters above the pavement, and a sensor that
measures the road speed of each vehicle that passes under the instrument.
When a vehicle moves through the air, a low-pressure zone typically forms behind the vehicle. In
this description, the low-pressure zone is called the vortex. As the vehicle drives down the road,
the vortex follows the vehicle at the same speed as the vehicle. The vortex is a dynamic,
swirling, mass of gases and particles with ill-defined boundaries that exchanges material with the
surrounding air moving past it.
RSD instruments obtain their signals using open-path photometric measurements of pollutants.
While pollutants can be anywhere around the moving vehicle, the highest levels of pollutant
mass are usually found in the vortex because the vortex temporarily stores pollutants. The vortex
is approximately as tall and as wide as the rear of the vehicle. In general, current RSD
instruments do not have enough sensitivity to quantify emissions in the wake behind the vortex.
Thus, an RSD's signal is dominated by the mass of a pollutant in the vortex.
Figure 5-4 shows a representation of a vehicle (the black square) driving to the left in the
vehicle's reference frame from above the roadway. If a pollutant is released from the vehicle (the
left red arrow), a portion of the mass of the release is temporarily stored in the vortex (the pink
triangular area behind the vehicle). At the same time, air moving around the vehicle and the
vortex strips off a portion of the pollutant mass from the vortex (the right red arrow). Under
steady-state conditions, the release rate, the mass in the vortex, and the stripping rate are in a
dynamic equilibrium. Consequently, on average, the release rate from the vehicle equals the
stripping rate from the vortex, and the mass in the vortex tends to be constant and proportional to
the release rate. So, for example, if the pollutant release rate from the vehicle is zero, the
pollutant mass in the vortex will tend to be zero.
5-3
-------
Figure 5-3. HEAT Remote Sensing Device Test Set-Up
EDAR
License Reader
Wind sensor
- Vehicle Speed
Figure 5-4. RSD ZigZag Scan Pattern on Pollutants from a Moving Vehicle
5-4
-------
Figure 5-4 also shows a representation of the RSD's measurement collection pattern across the
vehicle reference frame using blue dots. In this figure, the horizontal dimension is elapsed time,
and the vertical dimension is the spatial position across the roadway. The motion of the vehicle
through this scanning device causes a two-dimensional image-like measurement to be collected
for each type of pollutant that is being assessed. The blue dots in Figure 5-4 show how the
RSD's light beam scans the vehicle and its vortex from above as they pass under the RSD.
Because the vehicle and its vortex are moving but the RSD is stationary, the light beam tends to
make a zigzag in the vehicle/vortex reference frame. The blue dots represent the pixels, which
are the spots where the RSD makes each detailed data measurement. At each pixel, the RSD
records the optical mass (mole/m2) of a pollutant between the instrument and the pavement.
The RSD in this study measured the masses of HC, CO, NO, NO2, and CO2 at each pixel. Note
that the RSD does not collect any detailed data while the vehicle is covering the retro-reflective
tape on the pavement. This is shown in the figure as the lack of blue dots on top of the vehicle.
Measurements collected for the vehicle emissions task are assumed to be multichannel in nature.
The data for the vehicle emissions task consists of sets of multiple registered images, collected
simultaneously in a raster-scanning process using a laser-based measurement system over a
roadway. Each image is a collection of pixels in which the horizontal dimension represents time
and the vertical dimension represents position across the roadway. As a vehicle passes through
the measurement system, the system measures the amount of a particular species present in the
reflection of the laser beam at the sensor system. The species being monitored by this RSD
measurement system are hydrocarbons (HC), carbon dioxide (CO2), carbon monoxide (CO), and
nitrogen oxides (NO, NO2), which are associated with five unique RSD measurement channels,
respectively.
The focus of the measurement task is on understanding the hydrocarbon emissions of a moving
vehicle. The location of these emissions is an important key to understanding various parameters
of the vehicle's operation, including the possibility of leaks or other performance-limiting
behaviors of the combustion engine system. In addition, certain emission locations, such as the
vehicle's tailpipe, will emit multiple gases, and thus the spatial extent of one emission type, such
as CO2, may be highly correlated with that of another emission type, such as HC. For processing
purposes, gases emitted from the same location typically have the same spatial signature.
Moreover, due to the optical measurement process, these spatial signatures are linear and
additive where the plume emissions overlap. Thus, we have the following linear model,
x(i) = As (t) + n (t)
Equation 5-1
where
x(t) = [xi(t) x2(t) x3(t) x4(t) x5(t)]T
denotes the vector of five measurements of HC, CO2, CO, NO, and NO2 collected at time t,
s(t) = [si(t) s2(t) s3(t)]r
denotes the vector of three different spatial signatures indexed by emission location at time t, and
5-5
-------
an
0-12
«13
fl21
«22
«23
031
032
033
<141
a42
«43
051
152
^53
denotes the matrix of coefficients that map the amount of each spatial signature to the resulting
data channel collected by the measurement system. The coefficient atj corresponds to the
proportionality constant associated with the spatial signature emitted from Location j as
measured in the 7th channel at measurement time t. Note that l corresponds to measurement
position in the collected images, not the location where a particular gas has been emitted. Finally,
there is additive noise and artifacts in each channel, which is denoted by the noise vector
n(f) = [ni(t) n2(t) n3(t) n4(t) n5(t)]T
where "T" denotes the transpose of a vector. For purposes of the model, the spatial signatures
have some normalization associated with their scale. For example, each signature has the same
signal power or some other unit measure of area or volume, e.g., the temporal average of each
signature is one.
5.2 Pre-Processing Device
The Pre-Processing Device (PPD) takes the detailed data measurements collected by the RSD
and adjusts these measurements to improve their overall quality and reduce noise effects. These
improvements include:
• Adjusting constant-level offsets to all measurements of a single species to remove biases,
• Identifying outlier measurements and omitting them from processing,
• Filtering measurements to remove non-physical noise components such as tonal
disturbances and striping artifacts, and
• Adjusting the spatial location of the measured data points to a regular rectangular grid
using interpolation techniques.
Both the statistical and spatial structure of the data is used in this stage, as described below. The
resulting outputs are improved versions of the original data measurements, broken out by
measurement type, e.g., HC, CO2, and NO, among possible others.
Later, Section 7.1 will demonstrate each step in the pre-processing used in the Westminster
dataset by presenting examples for individual vehicle transits.
Adjusting Constant-Level Offsets: In this processing, each channel of data is treated as a
statistical measurement with an assumed constant value or offset when no species is present.
Thus, the measured value consists of a measured mixed signal, an additive noise signal, and a
possible constant offset value. To determine the value of the constant-offset value for this
channel of data, we first form a histogram of the values within the channel, denoted as p(bin),
where bin denotes the range of bin values corresponding to the overall range of values in the
original measured signal, such that each p(bin) value is the count of values within each bin.
5-6
-------
Assuming that the number of measurements in the channel is large, the spatial extent of the
measured mixed signal is small, and the additive noise is Gaussian-distributed, we can take the
logarithm of the histogram p(bin) and plot it as a function of the bin values in bin. The peak of
this plot will correspond to the bin value that is nearest the constant offset value. We solve for
the quadratic function of p(bin) versus bin to determine this offset value in the range of bin, and
then adjust all of the values in the measured signal to remove this offset value from them. This
also results in an estimate of the variance of the noise in the channel using the curvature of the
quadratic fit of the p(bin) versus bin values, which is used for outlier estimation, described next.
Outlier Removal: After Offset Adjustment, the signal may still contain large values that are
non-physical in nature due to erroneous operations within the RSD. Typically, these outlier
values are found near the vehicle due to erroneous light reflections (glints) caused by the
vehicle's shape as it is scanned by the RSD. To identify these values, we look for large values
that exceed a predetermined threshold value near the pixel positions identified by the RSD to be
vehicle pixels. The logic for this detection is as follows: A pixel must exceed a threshold value.
If it does, then the following conditions must also be true:
The pixel next to this pixel (either left or right) must be a vehicle pixel AND the pixel on the
other side of this pixel must not exceed a threshold.
OR
Both the pixel next to this pixel AND the pixel in front of this pixel occurring earlier in
measurement time must be vehicle pixels.
The threshold value for the detection is 1.96 times the noise standard deviation as determined in
the Offset Adjustment step. After this step, the values in the measured signal channels are largely
free of outliers due to vehicle pixel artifacts.
Filtering of Non-Physical Noise Components: As measured, the RSD signal may contain
periodic disturbances due to the mechanical nature of the measurement scanning process. Such
periodic disturbances are non-physical and are unrelated to the mass measurements being
collected. To reduce these noise artifacts, the following processing is performed:
1. The measured data in each channel is examined as a one-dimensional signal,
corresponding to the sequence of blue dots as shown in Figure 5-4.
2. The power spectrum of this signal is computed using standard frequency-domain
processing whereby a) the data is divided into blocks and windowed using a Hamming
window, b) the Fast Fourier Transform (FFT) of each windowed data block is computed,
and c) the magnitudes of the FFT values for each frequency bin are averaged across the
data blocks. A tonal noise signal will appear in the data as a peak in the power spectrum,
and the frequency of this peak is determined from the frequency bin value where the peak
occurs, denoted as estfreq. Finally, this value is then used in a two-pole, two-zero digital
infinite impulse response (IIR) filter with the form:
y[n] = x[n] + 2 cos(estfreq * pi) { 0.95 y[n-l] - 0.99 x[n-l] } + {0.95A2 x[n-2] - 0.99A2 y[n-2] }
5-7
-------
where x[n] is the measured input signal, and y[n] is the processed output signal for each
measured channel. After filtering, the signal is arranged into its zigzag pattern for further
processing.
Adjusting Measured Pixel Positions to a Rectangular Grid: The measured data locations as
shown in Figure 5-4 are not on a regular two-dimensional grid, and thus any estimation of
physical quantities based on these positions might distort the mass estimates based off of them.
To reduce these distortions, the measured data is interpolated to a rectangular grid, where the
blue dot positions represent the input to the interpolation process, and a corresponding set of
rectangular grid points corresponding to pixel positions that are evenly spaced in time across the
transit are used as output locations for the interpolation process. The interpolation is performed
in the x-direction only, thus corresponding to a one-dimensional interpolation of the data; the
direction dimension is neither adjusted nor interpolated.
After this interpolation is performed, a two-dimensional array of measurements for each channel
indexed by scan position m and scan number n is obtained, where m corresponds to the position
across the road and n corresponds to the passage of time along the road. We define these
measurements for each measurement channel using a vector representation for position and
indexed by scan number value n as
Xn = [xj>n 2-2,n ¦ ¦ * XM,n]
where M is the number of positions measured in a single scan and the range of n corresponds to
an appropriate time slot before the vehicle has arrived at the RSD instrument to an appropriate
time slot after the vehicle has passed the RSD instrument. Note that this representation still
corresponds to the linear model in Equation 5-1. In later portions of this description, we will
allow the signal x(t) for t=l to t=L to correspond to all of the interpolated measurements for a
particular channel through an appropriate assignment from t to the pair (m,n). For example, if
there are N scan numbers, this assignment is
{x(l),a;(2),..., x(L)} = x2,i,xM,i, xi,2, x2,2,. ¦ ¦ ,xm,n}
5.3 Separation/Estimation Device
This section describes the methodology's optional Separation/Estimation Device for determining
the separate emission rates of two or more different sources on a moving vehicle. If the emission
rates for a vehicle is to be determined without apportionment of emissions to separate source,
then this optional device is not needed as described by the flow diagram in Figure 5-1. However,
if the methodology is to be used to quantify emission rates for separate sources, such as for
evaporative HC emissions and exhaust HC emissions, then this optional device should be used.
The integration of this device with the methodology's other devices is described in Figure 5-2.
Later, Sections 7.2, 7.3, and 7.4 will demonstrate each step in the separation and estimation used
in the Westminster dataset by presenting examples for individual vehicle transits.
A flow diagram of the Separation/Estimation Device (SED) is shown in Figure 5-5. It consists of
two processing stages:
5-8
-------
Blind Source Separation (Stage 1): In this first stage of the Separation/Estimation Device, the
measured data is assumed to be of the form of the linear model in Equation 5-1. Blind source
separation algorithms are applied to the output of the Pre-Processing Device in Figure 5-2. The
resulting outputs are in the form of spatial patterns corresponding to plumes emitted from the
vehicle at different spatial locations around the vehicle. If the number of plumes identified is
fewer than the number of signal channels, the additional outputs produced from this stage are
labelled as noise components.
Emission Estimation (Stage 2): In this stage, those signals identified as plumes are combined
with the outputs of the Noise Reduction and Signal Correction block to estimate the emission
type contained in one or more identified plume patterns. For example, the plume associated with
tailpipe emissions can be combined with the improved NO channel data to determine the amount
of NO gas emitted from the exhaust location. As another example, the plume associated with
evaporative emissions at the fuel fill door location of the vehicle can be combined with the
improved HC channel data to determine the amount of HC gas emitted from the evaporative fuel
fill door location.
Figure 5-5. Flow Diagram of Separation/Estimation Device
r
Improved
Data
HC
C02
CO
NO
N02
Weights
Bumper Locns
Separation/Estimation Device
HC
C02
NO
Plume
Patterns
Blind
Source
Separation
Exh
Evap
UA
Emission
Estimation
Noise
Patterns
Emission
Components
-~ Exhaust HC
-~ Evap HC
-~Exhaust C02
¦> Exhaust NO
Two exemplary methods for extracting these components are described in a section below. The
result of this calculation is a set of images that correspond to selected gas types as emitted from
specific locations around the vehicle.
The overall goals and processing methodology for the Separation/Estimation Device is now
described. The approach leverages known results and algorithms in the signal processing
literature with unique design modifications and tuning for the methodology.
5-9
-------
Given the measurement model, the overall goal of the SED is to estimate the portion of a
particular gas being released from a particular location on the vehicle that is present in a
particular pixel measurement. This calculation is expected to be performed on a portion of an
extracted image generated from this linear model. To better understand this task, let the first
measurement at one pixel position for a given time instant be modeled as
x\(t) = x\\(t) + x\2(t) + ^13(2) Equation 5-2
where xn(7), xn(t), and xu(t) denote the hydrocarbon emissions emitting from three different
emission Locations #1, #2, and #3, respectively, where any noise m(t) contained in the first
measurement has been neglected. These locations could correspond to known point emission
locations, such as a tailpipe or fuel door, or they could correspond to patterns generated from
specific point locations, such as a leakage point under the vehicle hood. For the model in
Equation 5-2, these emissions are given by
aril(t) = ansi(t), x12{t) = ai2s2(t), and x13 = a13s3(t),
respectively. In this model, each Si(t) represents the spatial signature of gases emitted from the 7th
location on the vehicle as observed in the pixel xi(7) being analyzed. Thus, the goal is to process
the measurements in x(t) such that estimates of each xi i(7), xn(t), and xn(7) are generated. This
problem can be broken down into two tasks:
1. Process the measurements such that the spatial signatures Si(t), i={ 1,2,3} are reliably
extracted.
2. Use the 7th spatial signature and the original measurements to estimate a specific
component, such as xii(t) for a particular choice of i.
From the resulting images, an estimate of the total emissions coming from the particular spatial
location can be formed.
Stage 1: Blind Source Separation
Consider the Blind Source Separation stage. Assuming that the levels of additive noise and
artifacts are small, a linear model can be used to extract candidate spatial signatures from three
appropriately chosen channels of the five-channel measurement data as
s(t) = Bx(t)
where
s(t) = [si(t) s2(t) s3(t)]T
contains the estimated spatial signatures in a desired order, and
" bu
bi2
&13
614
bl5
B =
b2i
b22
&23
&24
&25
&31
&32
&33
&34
b'S5
5-10
-------
is a matrix of separation coefficients. If there is no noise or artifacts, then, the optimal solution
for the separation matrix satisfies
BA = I
where A is as defined in Equation 5-1 and I is an identity matrix with ones along its diagonal and
zeros everywhere else.
In practice, the number of spatial signatures may not be known. In addition, some residual noise
is present in the improved RSD signals. Thus, it is desirable to use an approach that can both
isolate remaining noise components and identify candidate spatial signatures that can be
analyzed to determine which of the signatures corresponds to significant parts of the improved
RSD data for specific species.
Given the measurement model, blind source separation is a well-known approach for processing
the measurements to produce candidate spatial signature signals. The FastICA algorithm14 can be
employed for this task. The FastICA algorithm describes an iterative approach for adjusting the
rows of a square matrix W according to the following criterion,
minimize/maximize
such that
where
£>{C(W(«))}
i=1
WRWr = s
R = i?{x(i)xT(i)}
C(yi(t)) is a contrast function, the matrix Z is the identity matrix for the FastICA algorithm,
E{M(t)} denotes the sample average of a matrix sequence M(t), and the candidate sources in the
vector y(t) are computed as
y(t) = Wx(()
This algorithm is appropriate for situations where the candidate plume spatial signatures do not
have significant spatial overlap, such that the constraint Z = I is appropriate. For some situations,
it is useful to model the overlap of the candidate plume signatures by allowing a non-diagonal
constraint matrix E that models this overlap. An example of a non-diagonal constraint matrix that
is appropriate for the separation task in the methodology is
1
p
0
0
0
p
1
0
0
0
0
0
1
0
0
0
0
0
1
0
0
0
0
0
1
14 A. Hyvarinen, E. Oja, "Independent Component analysis: algorithms and applications, Neural
Networks. Volume 13, Issues 4-5, June 2000, pages 411-430.
5-11
-------
where a positive value of p models the normalized cross-correlation between the exhaust and
evaporative spatial plume signatures. A typical value for p is p=0.15, although p values in the
range 0 < p < 0.4 are useful.
An iterative approach15 for adjusting W to solve the separation task in this methodology is as
follows:
1. Choose fi(yi) = dC{yt)/dyi or another appropriate function, e.g. fi{yt) — yf.
2. Do for each i — {1,2,..., 5}, where W — [wi W2 W3 W4 ws]T and
a\ is the ith diagonal element of S:
4. Set ~Wnew = P(y)Wnetu and s(t) = P(y)y(i) and repeat the above steps until convergence.
After separation, the candidate signatures in the vector y(t) need to be checked to see how well
they represent the improved RSD signals of one or more of the HC, CO2, CO, NO, and/or NO2
measurements. Knowledge of the typical plumes that can be emitted from vehicles on the road is
used here. For example, since CO2 concentrations above the ambient level are largely indicative
of exhaust emissions, the normalized correlations between the CO2 channel X2(t) and the five
candidate plume signatures in y(t) can be computed. Let pcozi denote these normalized
correlation values. The one with the largest absolute value of normalized correlation with index j
corresponds to the exhaust plume, and its sign can be used to adjust the amplitude of the
candidate exhaust plume as
«i(t) = sgn(pco2j)yj (t).
where the sign-function sgn(p) is 1 if p is positive-valued, 0 if p is zero-valued, and -1 if p is
negative-valued.
As an additional example, evaporative emissions, where they exist, are typically observed in the
HC measurements. The normalized correlations between the improved HC signal xi(t) and the
four remaining candidate plume signatures in y(t) can be computed as p nc.i. The one with the
15 S.C. Douglas, T.I I. DeFries, "Blind Source Separation under Signal Covariance Constraints: Criteria
and Algorithms," 2021 55th Asilomar Conference on signals, Systems, and Computers, 31 October 2021
- 3 November 2021.
R = E{x(t)x1 (<)}
^ = R 1 E{xfi(yi)} - E{fl(yi)}wi
3. Do for A: = 1 to 10 or until convergence, where W = [wi W2 W3 W4 w-,]T:
Wneiu = W+i(s-WRWr)w
new
5-12
-------
largest absolute value of normalized correlation with index k corresponds to the evaporative
plume, and its sign can be used to adjust the amplitude of the candidate evaporative plume as
s2(t) = sgn (pHc,k)yk(t)-
In this way, candidate plume signatures can be identified with the appropriate sign according to
the number of possible plumes that are observed in the improved RSD data, up to the number of
RSD data channels available. If the number of possible plumes is less than the number of RSD
data channels, then the identity assignment
Si(t) = Vi(t)
is made for the remaining data channels, as these remaining signals are noise spatial signatures.
These assignments of yi(t) to s_{hat}i(t) are also used to define the permutation matrix P(y) in
the algorithm. For example, if j=2 and k=l, then
P(y)
o
sSn{pHC,k)
0
0
0
sgn(pco2,j)
0
0
0
0
Other methods to perform this assignment and sign recovery could be used as well.
Stage 2: Emission Estimation
Consider the Emission Estimation stage, which performs estimations of the portions of these
spatial signatures in the original measurement data. Note that each improved RSD signal sample
x_i(t) has units of mass, whereas each plume signature s_{hat}j(t) are effectively unitless
because of normalization during the separation process. Thus, the primary goal of the estimation
task is to "recover" the mass units of each plume type for each species in an accurate way. These
estimates are computed using a linear model. For example, for the first measurement signal xi(Y),
the estimates are
xi(t) = hfs(£) (15)
= hnsiit) + huhit) + h13s3(t) (16)
= xn{t) + x12(t) + xi3(t) (17)
where the "hatted" quantities are estimates of their true values within the signal model. Thus, the
estimate of hydrocarbons emitted from Location #2 on the vehicle is denoted as
X12 (t) = hi2s2(t) (18)
These estimates are generally computed from the available signals using least-squares
techniques. Any one of a number of least-squares estimation techniques could be used, including
weighted least-squares and constrained weighted least-squares, amongst other methods. These
estimation tasks are formulated as a minimization of a cost function J(h) that depends on the
5-13
-------
measured and computed signals in a particular way, some with additional constraints on the
parameters in h. As illustrative examples, we describe two such approaches below.
Consider the weighted least-squares (LS) cost function
1 L
Jls(hi) =
t= 1
where L is the number of improved RSD signal samples corresponding to the HC RSD
measurements and wt is a weighting function across the L measurements for the improved RSD
signal images. In this context, t is the signal sample number, and each t can be mapped to a
particular (x,y) position in each plume image. Note that the exact mapping from l to (x,y)
determines the form of the weighting function for the estimation task. This weighting function
can be computed from the Weights W(v) output from the Vortex Shape Estimation Device by
extending the Weights across the improved RSD signal sample dimension t according to the
converse of the Weights computation. For example, if the Weight values are one-dimensional
and assigned by scan number, they can be extended to a two-dimensional weighting function by
replicating these Weight values across the width of the two-dimensional improved RSD signal
sample datasets, after which they can be assigned a one-dimensional index t associated with the
improved RSD signal samples. Other extensions of the Weights to a spatial arrangement
associated with the RSD measurement device are possible depending on the operation of the
RSD device.
The weighted least-squares cost function is minimized according to standard least-squares
methods. The resulting weighted least squares (LS) solution can be described as
h LS,i = R_1Pi
where
L
R = ^2 twf
t=l
L
p i = ^WtXi(t)st
t=1
are the weighted autocorrelation matrix and weighted cross-correlation matrix for the improved
RSD signal xi(t) being modeled. From these calculations, we can identify the estimated plume
emission for the chosen species to be analyzed from the weights and signals that have been
computed. For example, the evaporative HC plume samples can be identified according to the
description provided as
xn(t) = husi(t).
Other estimated plume emissions for different species can be computed similarly.
5-14
-------
Other least-squares methods can be employed for the Emission Estimation Stage as well.
Suppose instead the following constraints are desired to be imposed for each i
^ L 3
^ ^ ^ ] hinSn(t) = Sx = S^i
t= 1 n=l
where the quantities sx, S\.n. and Sd.i all contain sample average values:
Sx = [^1,1 &x,2
1 L ~
$x,i = ^ ^ si(t)
t=1
1 L
t-1
Then, we formulate the constrained least squares (CLS) problem
1 L
minimize JcLsO^i) = hj Rhi — 2hJ Pi + 7 ^ Wt^(t)
t=1
such that hf sx =
This is in the form of a quadratic linear programming problem with equality constraints. The
explicit solution for hi is the linear combination of two vectors hLs,i and hc,i as
hj = h^i + Ajhc.i,
where hLs,iis computed as before,
R Sx,i
Sd,i ~ hi^jSx
hC,iS*
It is straightforward to show that hir sx = Sd.i. This constrained least squares (CLS) solution works
best if the constraint is highly accurate; that is, the sample means are very close to the true
means.
5.4 Vortex Shape Estimation Device
The purpose of the Vortex Shape Estimation Device (VSED) is to characterize the expected
shape of the vortex for each vehicle as it drives past the RSD instrument. These characteristics
are needed to calculate vehicle emission rates from the RSD vortex mass measurements of
emission components as shown in the flow diagrams of Figures 5-1 and 5-2.
The categories of inputs are the improved CO2 data array, measured road speed and direction,
estimated vehicle size properties, RSD instrument measurement characteristics, wind speed and
direction, and estimated terrain surface roughness. The output categories are estimated for each
5-15
-------
vehicle transit. The general output categories are identification of the data location index (RSD
scan number) of the front and rear of each vehicle, weights that characterize the expected relative
magnitude of emissions at locations in the vortex, and the Vehicle Entrainment Time (VET)
expected for the vortex.
The flow diagram in Figure 5-6 shows the four steps that make up the Vortex Shape Estimation
Device. Each of the steps is described below.
Figure 5-6. Flow Diagram of Vortex Shape Estimation Device
Vortex Shape Calculation Device
Step 1. Air Speed Calculation Device
The shape of the vortex is influenced by the vehicle air speed, that is, the speed of the air moving
across the vehicle. The vehicle air speed is a function of the vehicle road speed and direction and
the wind speed and direction.
RSD systems may measure the wind velocity at a height above the pavement that is different
from the height that vehicles drive - typically 1 meter for light-duty vehicles. Therefore, the first
step is to calculate wind velocity at 1 meter from the wind velocity at the height that RSD
measures it. Then, the second step is to calculate the vehicle's air velocity at 1 meter from the
vehicle's road velocity and the calculated wind velocity at 1 meter.
The Danish Wind Industry Association relationship16 can be used to estimate the wind speed at
one height from wind speed measurements made at another height:
v = v_ref * ln(z/zO) / ln(z_ref/zo) Equation 5-3
16 http://xn--drmstrre-64ad.dk/wp-content/wind/miller/windpower%20web/en/tour/wres/shear.htm
5-16
-------
where:
v = wind speed at height z (meters) above the ground
v ref = measured wind speed at a height z ref (meters) above the ground
z = height (meters) above the ground for the desired wind speed v
zo = terrain surface roughness length (meters), estimated from Table 5-1, in the
current wind direction
Table 5-1. Terrain Surface Roughness Length Descriptions
Zo
Terrain Surface
Roughness Length
(m)
Surface Description
0.0024
Concrete runways, mowed grass.
0.03
Open agricultural area without fences and hedgerows with very scattered
buildings. Soft-rounded hills.
0.055
Agricultural land with some houses and 8 m tall hedgerows about 1250 m apart.
0.1
Agricultural land with some houses and 8 m tall hedgerows about 500 m apart.
0.2
Agricultural land with many houses, shrubs and plants, or 8 m hedgerows about
250 m apart.
0.4
Villages, small towns, ag land with many or tall sheltering hedgerows, forests,
and very rough and uneven terrain.
0.8
Larger cities with tall buildings.
1.6
Very large cities with tall buildings and skyscrapers.
The calculated wind direction is assumed to be the same as the measured wind direction.
The vehicle air speed vector at 1 m height is calculated from the wind speed vector at 1 m height
and the vehicle's road speed vector using standard vector algebra:
AS = RS - WS Equation 5-4
where
AS = Air Speed vector at 1 m elevation referenced to North heading
RS = Road Speed vector referenced to North heading
WS = Wind Speed vector at 1 m elevation referenced to North heading
Finally, the Air Speed vector at 1 m elevation is resolved into air speed components parallel and
perpendicular to the direction of vehicle motion, where • denotes the dot product for vectors and
|| . || denotes the length of a vector.
RP = Perpendicular vector to Road Speed vector
RP =
5-17
-------
AirSpeed Para
AirSpeed Perp
Equation 5-5a
Equation 5-5b
We used the following conventions for positive air-speed components in the vehicle reference
frame. Positive AirSpeed Para values represent air movement toward the windshield of the
vehicle. Positive AirSpeed Perp values represent air movement toward the left side of the
vehicle.
Step 3. Vehicle Characteristics Device
The purpose of the Vehicle Characteristics Device is to estimate the vehicle drag area and to use
the RSD improved CO2 measurements to identify the data location index (RSD scan number) of
the front and rear of each vehicle and the length of each vehicle. The drag area is used to
improve the estimate of the VET. The data location indexes of the front and rear of each vehicle
are used to time-align the improved RSD data arrays with the vortex weights that characterize
the expected relative magnitude of emissions at locations in the vortex. The front and rear
vehicle locations are in turn used to calculate vehicle length, which also influences the vortex
weights.
The inputs to the device are the measured road speed, the improved CO2 array, and the RSD scan
rate. The outputs of the device are the scan number of the last scan before the front of the
vehicle, the scan number of the first scan after the rear of the vehicle, vehicle length, and vehicle
drag area.
The data location indexes of the front and rear of each vehicle are determined by examining the
pixels where the outgoing RSD laser beam is not reflected back to the RSD since the laser beam
is occluded by the vehicle as exemplified by the missing blue dots in Figure 5-4. In that figure
the Last Scan Before Vehicle Front is Scan 2, and the First Scan After Vehicle Rear is Scan 11.
The length of the vehicle is then determined by:
Length Equation 5-6
= (FirstScanAfterVehicleRear - LastScanBeforeVehicleFront - 1) * Road Speed * 5280
RSD Scan Rate * 3600
where
Length
Fir st S can After V ehi cl eRear
Last S canB efore Vehi cl eFront
Road Speed
5280
RSD Scan Rate
3600
Vehicle length (ft)
First full scan after the vehicle rear
Last full scan before the vehicle front
Vehicle road speed (mile/hour)
Rate that RSD scans the vortex (scans/s)
3600 s/hr
5280 ft/mile
5-18
-------
The drag area is given by:
Drag Area (ft2) = Cd * FrontalArea Equation 5-7
where Cd = Coefficient of drag
FrontalArea = Frontal Area of the vehicle (ft2)
A default drag area of 10 ft2 can be used for light-duty vehicles. Optionally, to provide a more
precise value for drag area, the RSD license plate reader can be used with a state vehicle
registration database to look up vehicle year, make, and model, which can be used in turn to look
up specific drag areas for many vehicles.
Step 2. Vortex Entrainment Time Calculation Device
Vortex Entrainment Time (VET) is key to this methodology for measuring pollutant emission
rate using RSD. The Vortex Entrainment Time establishes a connection between the RSD-
measured pollutant Mass in Vortex and pollutant Release Rate (g/hr) and pollutant Emission
Rate (g/mile). The purpose of the Vortex Entrainment Time Calculation Device is to calculate
the Vortex Entrainment Time (VET) for each vehicle RSD transit.
Section 6.4 (below) describes the analyses of the September 2016 dataset and the October 2019
dataset collected to estimate the Vortex Entrainment Time of vehicles driving on the road. The
analysis of the September 2016 dataset found that the VET is a mild function of vehicle drag
area. The analysis of this October 2019 Westminster data found that the VET also depends
mildly on the AirSpeed Para and the emissions Release Location.
Regression analysis from the September 2016 dataset and from the October 2019 dataset shows
(see Section 6.4) that VET values can be determined by considering descriptors of vehicle,
vehicle operation, and environmental conditions as described by:
Equation 5-8
VET (s) = B * Release Location Factor * DragArea (ft2)A(l/3)
AirSpeed Para (mile/hr) A (1/2)
where
B
Release Location Factor
DragArea (ft2)
AirSpeed Para (mile/hr)
A constant determined by calibration
1.00, if release location is known to be the tailpipe
0.67, if the release location is unknown
Vehicle drag area from Step 3
Parallel component of the AirSpeed from Step 1
For releases from the vehicle rear, such as from the tailpipe or a fuel fill door located on the rear
of a quarter panel, the Release Location Factor will be near 1. If the emissions release location is
known to be the tailpipe, then the Release Location Factor of 1.00 should be used in Equation 5-
8. A Release Location Factor of 0.67 can be used if the actual release location is unknown. As
shown in Table 6-4, the Release Location Factor will be lower for release locations more forward
5-19
-------
on the vehicle. For example, analysis has shown that releases from under the hood have Release
Location Factors around 0.34.
The B coefficient value in Equation 5-8 is specifically for the RSD used in the October 2019
dataset. The B coefficient will be different for different RSDs due to their different optical
efficiencies, for example, retro-reflector efficiency, and their different optical strategies for
illuminating the vortex. The value of B for alternative RSDs can be determined using the
following procedure, based on currently available technology and methods:
1. Select a test vehicle of known Drag Area. Instrument the test vehicle to determine its
tailpipe exhaust CO2 Release Rate (g/hr) by either querying its CAN bus data stream or
its driver instrument display for fuel economy, or by directly measuring the tailpipe CO2
release rate with an external measurement device, such as a Portable Emissions
Measurement System (PEMS) or miniPEMS.
2. Drive the test vehicle past the RSD while collecting RSD detailed CO2 data of the vortex
and data to determine the AirSpeed Para for the transit.
3. Calculate the CO2 Mass in Vortex (g) using the collected RSD detailed CO2 data.
4. Calculate the VET (hr) using Equation 5-16 with the measured CO2 Mass in Vortex (g)
and the measured CO2 Release Rate (g/hr).
5. Calculate the value of B using Equation 5-8 with the calculated VET, known Drag Area,
determined AirSpeed Para, and a Release Location Factor of 1.00, which is the defined
value for tailpipe releases.
Step 4. Weights Calculation Device
The purpose of the Weights Calculation Device (WCD) is to provide weights that reflect the
probable distribution of relative emissions Mass in Vortex. The weights are used by the Emission
Calculation Device as a distribution to which RSD measurements of any pollutant are fit to
determine the mass of emissions in the vortex while enhancing the signal-to-noise ratio and
thereby improving the method detection limit of the pollutant emission rate. The weights are also
used by the Separation/Estimation Device to better assign RSD-measured pollutant mass to
separately located pollutant sources.
The inputs to the Weights Calculation Device are the vehicle length, which is determined from
the Vehicle Characteristics Device, and the AirSpeed Para at 1 meter, which is determined by the
Air Speed Calculation Device. The output of the Weights Calculation Device is a set of weights
as a function of time after the rear of the vehicle.
The weights can be expressed in terms of three factors: time after the vehicle rear, vehicle length,
and AirSpeed Para:
Equation 5-9
Weight = Time-Decay Factor * Length Factor * Air-Speed Factor
The dependencies for the three factors are shown in Figures 5-7, 5-8, and 5-9.
5-20
-------
Figure 5-7. Time-Decay Factor for Weights
0.2 0.3 0.4 0.5 0.6 0.7 0.8
Time After Vehicle Rear (s)
/proj1/EDARinDenver-OCT2019/Analysis_MLout^11122/Anal_MLout/OCT19_interpshapeCO2_5.sas 15JUN22 12:03
Figure 5-8. Vehicle Length Factor for Weights
U 4]
(0
2
-1 1
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
Time After Vehicle Rear (s)
Vehicle Length (ft) — 00-09
16-19
> 09-11
19-22
~ 11-13
22-26
13-16
26-44
0.9
1.0
/proj1/EDARinDenver-OCT2019yAnalysis_MLout/211122/Anal_MLoiit/OCT19_interpshapeC02_5.sas 16JUN22 12:38
5-21
-------
Figure 5-9. Air Speed Parallel Factor for Weights
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
Time After Vehicle Rear (s)
AirSpeed Para (mph) --- 00-18
33-38
'18-21 •-•-•21-24
38-45 45-52
24-28
52-60
28-33
1 60-99
/praj 1/EDARinDenver-OCT2019/Analysis MLout/211122/Anal_MLoi.it/OCT19_interpshapeC02_5 sas 06JUL22 09:55
5-22
-------
The Time-Decay Factor is shown in Figure 5-7 as a function of time after the vehicle rear. The
plot shows that the time-decay factor is an almost perfect exponential decay.
The Length Factor is shown in Figure 5-8 and is a function of time after the vehicle rear and the
vehicle length. The figure shows that after 0.5 seconds after the vehicle rear, the factor has a
value of 1, which indicates that vehicle length has no influence after 0.5 seconds after the vehicle
rear. The different curves for vehicle length indicate that the peak at 0.1 seconds is large for short
vehicles and decreases and approaches 1 for longer vehicles.
The Air-Speed Factor is shown in Figure 5-9 as a function of time after the vehicle rear and
AirSpeed Para. The figure shows that after 0.25 seconds after the vehicle rear, the factor has a
value of 1, which indicates that AirSpeed Para has no influence after 0.25 seconds after the
vehicle rear. For times shorter than 0.25 seconds, the figure shows that AirSpeed Para less than
about 40 mph are associated with air-speed factors less than 1, and AirSpeed Para greater than
about 40 mph are associated with air-speed factors greater than 1.
Examination of Figures 5-7, 5-8, and 5-9 indicates that after 0.5s after the vehicle rear, the
product of the three factors depends almost entirely on the exponential time-decay factor shown
in Figure 5-7. At shorter times after the vehicle rear, the influences of vehicle length and air
speed make substantial modifications to the decay - particularly for short vehicles and for low
values of AirSpeed Para.
The region at short times after the vehicle corresponds to the region close behind the vehicle rear
where the vortex has the largest mass of pollutants. Accordingly, the RSD gets a large part of its
signal from this region. Therefore, the weights for vehicle length and airspeed in this region are
important to achieving accurate emissions rate measurements with good detection limits.
While the values for the three factors that contribute to the weight could be read from Figures 5-
7, 5-8, and 5-9 one convenient set of parameterizations of the curves in those figures is given by
Equations 5-10a17, 5-10b18, and 5-10c19. The parameterization covers vehicle lengths from 10 to
27 feet and AirSpeed Para values from 16 to 67 mile/hr.
17 P:\EDARinDenver-OCT2019\Analysis_MLout\211122\Anal_MLout/ 0CT19_interpshapeC02_5.sas
18 P:\EDARinDenver-OCT2019\Analysis_MLout\211122\Anal_MLout/ rat21ength.xlsx/tab:Parms +
pred2 rat2decay and P:\EDARinDenver-OCT2019\Analysis_MLout\211122\Anal_MLout/
0CT19_interpshapeC02_7.sas (draft NLIN)
19 P:\EDARinDenver-OCT2019\Analysis_MLout\211122\Anal_MLout/ 0CT19_interpshapeC02_5.sas
5-23
-------
Time-Decay Factor =
where K
t
exp (K * t)
Equation 5-1 Oa
= -1.38 --1
-1.03 --1
s % for high altitudes (October 2019 study)
s"1, for altitudes near sea level (September 2016 study)
Time after the vehicle rear (s)
Length Factor = 1 + FastDecay + RampCorr Equation 5-1 Ob
where FastDecay = exp (Length lntercept + Length Slope * t)
Lengthlntercept
LengthSlope
Length(ft)
- 0.3088 * Slope - 2.2568
0.5211 * Length(ft) - 23.662
= 10.0, for LengthBin = 00-09
= 10.0, for LengthBin = 09-11
= 11.9, for LengthBin = 11-13
= 14.2, for LengthBin = 13-16
= 17.3, for LengthBin = 16-19
21.5, for LengthBin = 19-22
= 24.0, for LengthBin = 22-26
= 27.0, for LengthBin = 26-44
where
RampCorr
Ramp
(1+ exp(Length lntercept)) * Ramp
- 1.000, for t=0
-0.410, for t=0.05
0.030, for t=0.10
0, fort>0.0125
Air-Speed Factor = 1 + Air lntercept * exp(-18.8260 * t) Equation 5-10c
where Air_Intercept = 2.4720 - exp(exp( 0.7716 - 0.0240 * AirSpeedPara (mile/hr)))
AirSpeedPara (mile/hr)=
16.0, for AirParaBin = 00-18
19.5, for AirParaBin = 18-21
22.5, for AirParaBin = 21-24
25.0, for AirParaBin = 24-28
31.0, for AirParaBin = 28-33
33.5, for AirParaBin = 33-38
40.0, for AirParaBin = 38-45
49.0, for AirParaBin = 45-52
57.0, for AirParaBin = 52-60
67.0, for AirParaBin = 60-99
5-24
-------
5.5 Emission Calculation Device
The purpose of the Emission Calculation Device (ECD) is to calculate the pollutant Release Rate
and Emission Rate using the calculated quantities derived earlier from the RSD system
measurements and estimates of vortex properties.
The categories of inputs are the improved pollutant data arrays, RSD data location index for the
vehicle rear, vortex weights, RSD instrument geometry and operating characteristics, Vortex
Entrainment Time (VET), and vehicle road speed. The outputs are the time-based Release Rate
(g/hr) and the distance-based Emission Rate (g/mile) for each transit and each pollutant
measured by the RSD.
The Emission Calculation Device consists of Steps 5, 6, 7, and 8 as shown in Figure 5-10. These
steps are described below.
Figure 5-10. Flow Diagram of Emission Calculation Device
Emission Calculation Device (Single Pollutant)
Step 5. Pollutant RSD Signal Device
The purpose of the Pollutant RSD Signal Device is to calculate the mass equivalent of a pollutant
signal obtained by an RSD instrument from the vortex behind a moving vehicle.
The inputs to the device are the pollutant Improved Data array from the Pre-Processing Device
or the pollutant Emission Components array from the Separation/Estimation Device (if
separation is performed), Weights from the Weights Calculation Device, Vehicle Rear Scan
Number from the Vehicle Characteristics Device, and a Pollutant Conversion Factor that gives
the conversion between the RSD optical measurement quantity and the pollutant mass. The
output of the device is Pollutant RSD Signal (g).
The device's first step is to convert the optical mass measurement of each pixel in the pollutant
array into a mass value:
5-25
-------
Pollutant Mass in Pixel (g)
Equation 5-11
= Pixel Optical Mass (mole/m2) * Light Beam Area (m2) * Pollutant Conversion Factor
where
Pixel Optical Mass (mole/m2) = RSD-measured pollutant optical mass in each pixel
Light Beam Area (m2) = Cross-sectional area of the RSD light beam
Pollutant Conversion Factor = Factor that converts the RSD-measured optical mass
to pollutant mass
The RSD instrument used by this study reports gaseous pollutant Pixel Optical Mass in units of
mole/m2. Therefore, in this study for pure compound gaseous pollutants (CO, NO, NO2, CO2),
the Pollutant Conversion Factor is the Pollutant Molecular Weight (g/mole). For Exhaust HC,
Evaporative HC, and their sum, Total HC, which are mixtures of pure HC compounds, the
Pollutant Conversion Factor is the Pollutant Molecular Weight (g/mole) of the basis gas,
propane, in which the emission rates are to be calculated. Accordingly, for all RSD channels,
Equation 5-11 becomes:
Pollutant Mass in Pixel (g) Equation 5-11 a
= Pixel Optical Mass (mole/m2) * Light Beam Area (m2) * Pollutant MW (g/mole)
The device's second step is to convert the two-dimensional RSD data location indexes
referenced by scan position m and scan number n to one-dimensional data location indexes
referenced to the vehicle rear so that the pollutant data array can be spatially and temporally
aligned with the Weights. The data location indexes are the scan identifier values v. This process
is performed in two sub-steps. In the first sub-step, the Pollutant Mass in each Scan Number is
computed by summing values of the two-dimensional interpolated RSD measurements in each
instrument channel indexed by scan number n across the range of scan positions m as
M
Pollutant Mass in Each Sean(n) = sm>fl
m— 1
In the second sub-step, the Pollutant Mass in Each Scan is shifted by the Scan Number of the
Vehicle Rear using the conversion:
Equation 5-12
Scan Value v After Vehicle Rear = Scan Numbers - Scan Number of Vehicle Rear
In this way, we obtain the Pollutant Mass in Each Scan values indexed by the Scan Value v.
Typical ranges of the value v are from v=0 to v=20.
The device's third sub-step is to combine the array of pollutant Mass in Each Scan values with
the Weights to produce the Pollutant RSD Signal. This step can be considered as a fitting of the
Weights to the array of Pollutant Mass in Each Scan values, followed by taking the area under
the fit to the Pollutant Mass data array. Let W(v) be the Weights indexed by scan value v
5-26
-------
produced by the Vortex Shape Estimation Device, denoted as W(v), v={l,2,...,20}. The Pollutant
RSD Signal is computed as
20
^ W(d) • Pollutant Mass in Each Scan(v)
Pollutant ESD Signal - ^ _
f>w
„=o
Step 6. 100% Illumination Speed Device
The purpose of this device is to calculate the RSD 100% Illumination Speed (100%IS), which is
the road speed at which the vehicle/vortex would have to move to produce an RSD signal that
would equal the RSD signal produced if the RSD light beam illuminated the scan path once and
only once. The 100%IS is independent of other variables including vehicle, road speed,
pollutant, release rate, release location, and wind.
The inputs to the 100% Illumination Speed Device are the geometrical and operating properties
of the RSD instrument. For this set-up, these inputs are the RSD laser beam radius, laser beam
scan rate, scan path length, and number of pixels per scan. The device output is the 100%
Illumination Speed:
100% Illumination Speed (m/s) Equation 5-13
= Number of Pixels/Scan * Effective Pixel Area (m2/pixeO * Scan Rate (scan/s)
Scan Length (m)
Step 7. Mass-in-Vortex Calculation Device
The Mass-in-Vortex Calculation Device calculates the mass of pollutant in the vortex. The
calculation uses the pollutant RSD Signal (g), the vehicle Road Speed, and the RSD's 100%
Illumination Speed.
In general, RSD instruments do not illuminate the entire vortex. They illuminate a sample of the
vortex. Thus, RSD signals are proportional only to the fraction of the vortex that they illuminate.
The fraction of the vortex that RSDs illuminate is related to the geometry of the RSD's light
beam and its illumination of the vortex.
As the vehicle and vortex move faster, the distance between consecutive scans in the vortex gets
larger, and therefore the fraction of the vortex that is illuminated decreases. Consequently, as
road speed increases, the RSD signal tends to decrease. The RSD Signal (g) is just the sum of the
pollutant masses measured by the RSD in all RSD scans of the vortex or in all RSD pixels of the
vortex. Therefore, to determine the mass in the entire vortex, the RSD signal must be corrected
for the road speed.
5-27
-------
The purpose of the Mass-in-Vortex Calculation Device is to calculate the pollutant Mass in
Vortex from the pollutant mass that was illuminated and measured by the RSD. The Mass in
Vortex is given by:
Equation 5-14
Mass in Vortex (g) = RSD Signal (g) * Road Speed (mile/hr)
100% Illumination Speed (mile/hr)
where
Mass in Vortex (g) = Mass of the pollutant in the entire vortex
RSD Signal (g) = Pollutant RSD signal output by the Pollutant RSD Signal
Device
Road Speed (mile/hr) = Vehicle road speed measured by the RSD system
100% Illumination Speed (mile/hr) = 100% illumination speed from the 100%
Illumination Speed Device
Step 8. Emission Rate Calculation Device
The purpose of the Emission Rate Calculation Device is to calculate the Release Rate (g/hr) and
Emission Rate (g/mile) from the calculated pollutant Mass in Vortex using the calculated value
of the Vortex Entrainment Time (VET) and the measured vehicle Road Speed.
A portion of the emissions released from a vehicle is temporarily stored or entrained in the
swirling vortex that follows a moving vehicle. This entrainment process is a dynamic equilibrium
consisting of the emissions released from the vehicle, flow of a portion of the emissions into the
vortex behind the vehicle, and the stripping of emissions from the vortex by the air passing over
the vortex as the vortex moves down the road. This process is shown in the left side of Figure 5-
11. For a vehicle operating under steady-state conditions, which can be defined as constant road
speed, constant wind speed and direction, and constant emissions release rate, the mass of a
given pollutant in the vortex will oscillate around an average value.
This steady-state entrainment process is similar to the continuously stirred tank model in
chemical engineering as shown in the right side of Figure 5-11. For the stirred tank model, a
stream of liquid, which contains a solute, flows into the tank, and an equivalent flow exits the
bottom of the tank. Under steady-state conditions, the mass of solute in the tank is proportional
to the solute release rate with a proportionality constant called the turnover time:
Mass in Tank (g) = Turnover Time (hr) * Release Rate (g/hr)
Equation 5-15
Under steady-state conditions, if the solute release rate is low, the mass in the tank will be low. If
the release rate is high, then the mass in the tank will be high. The turnover time is characteristic
of the inlet flow rate, the volume of the tank, and the mixing in the tank.
5-28
-------
Figure 5-11. Continuously Stirred Tank Analogy
Vehicle
Solute
Release
(g/hr)
Solute
Release
(g/hr)
Following the stirred tank model, Equation 5-15 is expressed in terms of entrainment of released
emissions into the vehicle's vortex:
Mass in Vortex (g) = VET (hr) * Release Rate (g/hr)
Equation 5-16
where the proportionality constant is now called the Vortex Entrainment Time (VET). From
Equation 5-16, a large VET value means that the ratio of Mass in Vortex to Release Rate is large.
There are two contributions to the size of VET: 1) the efficiency of entrainment of emissions
released from the vehicle, and 2) the volume of the vortex.
When most of the released emissions bypass the vortex and only a small fraction becomes
entrained, the VET is small. This is more likely to happen for emissions released farther up front
on the vehicle and especially when there is strong sideways air movement caused by wind.
Vehicles that have large drag areas tend to have large vortexes. Large vortexes can store more
pollutant mass for the same release rate than a small vortex can. Thus, large vortexes tend to
produce larger RSD signals and tend to have larger VETs.
A rearrangement of Equation 5-16 provides Equation 5-17, which expresses the pollutant
Release Rate based on an RSD measurement of Mass in Vortex and the VET.
Release Rate (g/hr) = Mass in Vortex (g)
VET (hr)
Equation 5-17
5-29
-------
Then, the Emission Rate from Equation 5-17 divided by the Road Speed gives the Emission
Rate:
Emission Rate (g/mile) = Release Rate (g/hr) Equation 5-18
Road Speed (mile/hr)
where
Mass in Vortex (g) = Pollutant mass in vortex calculated from the Mass-in-
Vortex Calculation Device
VET (hr) = Vortex Entrainment Time calculated from the Vortex
Entrainment Time Calculation Device
Road Speed (mile/hr) = Vehicle road speed as measured by the RSD system
Overall, a direct expression for the Emission Rate can be obtained by substituting the expression
for the Mass in Vortex (Equation 5-14), into the expression for Release Rate (Equation 5-17),
and then into the expression for Emission Rate (Equation 5-18) to produce:
Equation 5-19
Emission Rate (g/mile) = RSD Signal (g)
100% Illumination Speed (mile/hr) * VET (hr)
This fundamental expression shows that a vehicle's instantaneous emission rate (g/mile) can be
determined from the RSD signal from pollutant mass in the vortex, the RSD instrument's
geometry, and an estimate of the Vortex Entrainment Time of the vehicle driving past the RSD.
5-30
-------
6.0 Quantities Needed by the RSD Emission Rate Method
6.1 Vehicle Air Speed and Direction
While the wind speeds were measured by EDAR at 6 meters above the pavement, light-duty
vehicle plume dispersion is affected by wind speeds closer to the ground where the vehicles are
located. We estimated the wind speeds at 1 meter above the pavement using the 6-meter wind
speed measurements and a relationship20 used by the Danish Wind Industry Association that
describes wind speeds at different heights as a function of surface roughness lengths, which are
described in Step 1 in Section 5.4 using Equation 5-3 and Table 5-1.
The land to the northwest of the Federal Parkway site was a dry creek bed with low grassy
vegetation and a few widely spaced trees and bushes and no buildings for at least 200 m. The
southeast side of the road had a tall hedgerow along the sidewalk with two-story buildings
beyond. Based on the descriptions in Table 5-1 and consideration of the topography of the
Westminster measurement site, we chose zo = 0.1 m. Using this value of zo and z ref = 6 m and
Equation 5-3, the estimated wind speed at 1 meter above the surface would be 56.2%
[=ln(l/0.1)/ln(6/0.1)] of the measured wind speed at 6 meters. The distribution of estimated wind
speeds at 1 meter is shown in Figure 6-1. We assumed that the wind direction at 1 meter above
the pavement was the same as the measured wind direction at 6 meters above the pavement.
Figure 6-1. Estimated Wind Speeds at 1 Meter above Pavement
a
E
"O
0)
cu
o.
(/)
"D
c
"D
CU
¦M
ro
E
'-P
i/i
LU
30 2000 3000 4000 5000
FREQUENCY
/proj1/EDARinDenver-OCT2019/Analysis/analyze_Fleet_Wind_1.sas 06MAY20 17:28
CUM.
CUM.
PCT
4 16
4022
8519
2.66
26.82
4227
12746
3.31
40.13
3338
16084
0.51
50.64
2499
18583
7.87
58.51
2432
21015
7.66
66.16
2192
23207
6.90
73.06
2142
25349
6.74
79.81
1675
27024
5.27
85.08
1455
28479
4.58
89.66
968
29447
3.05
92.71
710
30157
2.24
94.94
496
30653
1.56
96.51
392
3
045
1,23
97.74
269
3
314
0.85
98.59
175
3*
489
0.55
99.14
118
3
607
0.37
99.51
57
3
654
0.18
99.69
41
3
705
0.13
99.82
22
3
727
0.07
99.89
16
3
743
0.05
99,94
7
3
750
0.02
99.96
3
3'
753
0.01
99.97
7
3*
760
0,02
99.99
1
3
761
0,00
99,99
1
3
762
0.00 100.00
1
31763
0.00 100.00
0
3
763
0.00
00.00
0
3'
763
0.00
00.00
0
3
763
0.00
00.00
0
3
753
0.00
00.00
0
3
763
0.00
00.00
0
3
763
0.00
00.00
0
3
763
0.00
00.00
0
3*
763
0.00
00.00
0
3
763
0.00
00.00
0
3
763
0.00
00.00
0
31763
0.00 100.00
0
3
763
0,00
00.00
0
3
763
0,00
00.00
0
3
763
0.00
00.00
0
3*
763
0.00
00.00
0
3
763
0.00
00.00
0
3
763
0.00
00.00
0
3*
763
0.00
00.00
0
3
763
0.00
00.00
0
3
763
0.00
00.00
0
3
763
0.00
00.00
0
3
763
0.00
00.00
0
31763
0,00 100,00
1 http://xn--drmstrre-64ad.dk/wp-content/wind/miller/windpower%20web/en/tour/wres/shear.htm
6-1
-------
The dispersion of emissions from a moving vehicle is affected by the apparent movement of air
around the vehicle, which is made up of two components in the vehicle's reference frame: 1) the
wind at the height of the vehicle, which we have estimated as described above, and 2) the
apparent air movement as a consequence of the forward motion of the vehicle. For example, if
there is no wind, a vehicle driving on the road at 30 mph experiences an apparent air movement,
in the vehicle's reference frame, of 30 mph blowing toward the front of the vehicle. For another
example, if a vehicle is driving north at 20 mph, and if there is a 20 mph tail wind from the
south, in the vehicle's reference frame there is zero apparent air movement. That condition is as
if the vehicle were motionless on a calm, no-wind day.
Therefore, to analyze the EDAR emissions data, and because plume dispersion may be affected
by the apparent air movement in the vehicle's reference frame, we need to calculate the air speed
and direction from the lm wind speed and direction and the vehicle's road speed and direction.
This is a vector calculation. Figure 2-1 shows that all vehicles moved on a bearing of 72 degN
(approximately ENE) and the speed measured by the EDAR instrument. Figure 6-2 shows the
road speed distribution of the fleet vehicles in the study.
Figure 6-2. Distribution of Fleet Vehicle Road Speeds
¦o
0)
>
CUM.
CUM.
FRECb
freo.
pc&To
P0
-------
Speed = lOmphV and Air Direction = -90 degV. To serve as a check, the following are the
measured quantities and the calculation results21 for the vehicle transit on 10/20/2019 at 10:11:13
AM:
Thus, in this vehicle's reference frame, air is moving toward the front left of the vehicle from 30
degrees left of the vehicle's centerline at 23.8 mph. This is the air movement resultant in the
vehicle's reference frame as a consequence of driving ENE (72 degN) at 44.5 mph in a 47.6 mph
wind coming from the west (279 degN) as measured at 6m above the pavement.
The air movement calculations were performed for all fleet vehicle transits of the EDAR
instrument. Figures 6-3 and 6-4 show distributions of the air speed and direction in the vehicle
reference frame. Figure 6-5 shows a plot of air speed vs. air direction in the vehicle's reference
frame. Note that almost all air movement directions are ±20 degV with respect to directly in
front of the vehicle, but the air speeds have a wide range from 10 to 80 mphV.
21 P/EDARinDenver-OCT19/Analysis/Westminster_OCT19Results_200124Reprocess-200219.xlsx
Vehicle Speed (measured)
Vehicle Direction (measured)
6m Wind Speed (measured)
Wind Direction (measured)
lm Wind Speed (calculated)
Air Speed (calculated)
Air Direction (calculated)
44.5 mph
72 degN
47.6 mph
279 degN
26.8 mph (=0.562 * 47.6 mph)
23.8 mphV
-30 degV
6-3
-------
Figure 6-3. Vehicle Reference Frame: Fleet Air Speed Distribution
Q.
£,
"O
U)
-------
Figure 6-5. Vehicle Reference Frame: Fleet Distribution of Air Movement
120 j
110;
100^
90-
_ 80
-60 -45 -30 -15 0 15 30 45 60
Air Direction (degV)
/projl/E DARinDenver-OCT2019/Analysis/analyze_Fleet_Wind_ 1.sas 06MAY20 17:28
6.2 Vehicle Footprint Length and Width
As described above and demonstrated in Figure 5-4, when a vehicle is on top of the pavement
tape, the laser beam is not returned to the overhead EDAR instalment. This causes missing
detailed data values, which when imaged as in Figure 6-6, produce a "vehicle footprint." In this
figure, the vehicle footprint is the evenly colored blue area to the left of the red/yellow/green
vortex area which contains the emissions mass. The size of the footprint is related to the size and
shape of the vehicle. Since the vehicle footprint is made up of pixels with missing values, we can
calculate the dimensions of the footprint by examining the pattern of the missing pixels.
Figure 6-6. Image of an EDAR Detailed Data Array
Vehicle
Footprint
6-5
-------
The length of the footprint, which is in the direction of vehicle motion, is a function of the
vehicle speed, which is measured by the EDAR system, and the vehicle length. In this study, the
laser beam always scanned at a constant 20 scans/s rate. For a given vehicle length, vehicles
moving at low speeds produce more scans with missing-value pixels than vehicles moving at
high speeds. For example, a 17.6-foot-long vehicle moving at 30 mph would have about 8
missing-value footprint scans, while the same vehicle moving at 60 mph would have about 4
missing-value footprint scans.
The width of the footprint is affected by the vehicle width, the vehicle height, and the length of
the laser scan on the pavement tape. Because of the triangular scan pattern of the laser beam, as
shown in Figure 2-3, tall vehicles, as well as wide vehicles, produce a wide footprint. For
example, the top of a 53-foot, 102-inch wide, 13.5-foot-tall, 18-wheeler box trailer would be so
close to the overhead EDAR instrument that the footprint would be the entire 256 pixels wide.
On the other hand, if the EDAR instrument were 15 feet above the pavement with a 12-foot
pavement tape, a 60-inch-high, 66-inch-wide car would have a footprint only about 82 inches
wide or about 145 pixels.
The number of scans and number of pixels with missing values in each transit, pavement tape
length, laser scan rate, and vehicle speed were used to calculate the length and width of each
transit's vehicle footprint.
Locating vehicle front and rear - Before we could calculate the length and width of each
transit's vehicle footprint, we needed to determine which EDAR scans were associated with the
front and rear of the vehicle. Because a CO2 signal is always present in the vortex of a vehicle
with a fossil-combusting engine, we used the EDAR CO2 detailed data to locate the vehicle
footprint. While we could locate the footprints of a small number of transits by examining the
CO2 detailed data by eye, because the Westminster dataset contains more than 30,000 transits, we
developed an automated method for determining the scans associated with the front and rear of
each footprint.
We used the raw CO2 detailed data of the 1180 test vehicle transits as a dataset to develop an
automated algorithm for determining the scan number of the first scan after the vehicle's rear
bumper, which we call BumperCounter=l. For each of those transits, we made a plot of the sum
of the CO2 measurements in each scan vs. scan number overlaid with a plot of the number of
missing (blank) CO2 values in each scan vs. scan number. Then, we examined each transit's plot
by eye and recorded the scan number that we judged was the first scan after the rear of the
vehicle.
We wrote a SAS program22 that used hand-observed after-vehicle scan numbers of the 1180 test
vehicle transits to develop a draft automatic method for determining those values. The first scan
after the end of the footprint was defined as the scan after the last scan where the number blank
CO2 pixels was greater than three pixels. Three pixels were used as the detection threshold since
occasionally three pixels could have blank CO2 values as a consequence of HEAT's raw data
post-processing that was not indicative of the footprint. The program determined the last scan
22 P:\EDARinDenver-OCT2019\Analysis_MLout\211108/OCT19_bumper.sas
6-6
-------
before the beginning of the footprint by reversing the order of sorting of the scan numbers and
then applying the same detection method.
The program was then applied to the Westminster dataset to validate the front and rear footprint
assignments by visual checks against a random set of the fleet vehicle transits.
Footprint length - Figure 6-7 shows an example of scans assigned to a vehicle transit. The
black trace and right axis show the counts of the number of blank pixels in each scan. On the
basis of the automated assignment, a vertical black, dashed reference line is placed at the last
scan that intersects the vehicle at After-Vehicle Scan =0. The next scan is the first-after-vehicle
scan, which is assigned After-Vehicle Scan=l. In a similar manner, with reverse scan sorting, the
last-before-vehicle scan is located at After-Vehicle Scan = -7 and is marked by a small green
triangle symbol. The result is that all scans between After-Vehicle Scan = -7 and 1 have at least
three blank pixels, which indicates the presence of the vehicle. Because the vehicle footprint has
seven scans and the measured vehicle speed was 45.5 mph, we calculate a vehicle length of 23
feet (=45.5 * 5280 * (7/20) / 3600). Since we are using just whole numbers of scans, this
calculated length is approximate. The red trace and left axis of Figure 6-7 show the CO2
scansums for the transit. Notice that the CO2 scansum is low before After-Vehicle Scan =1. After
the vehicle clears the laser beam, the CO2 scansum rises rapidly and then decays back toward
zero as the exhaust CO2 disperses.
Figure 6-7. CO2 ScanSums and Blank Pixel Counts for a Car
(45.5 mph, Series_Transit=505_000299)
fN
E
o
E.
fN
o
u
<
o
LU
"O
re
©
o
E
3
l/>
c
re
u
90 ;
80:
70
60
50
40
30
20
10
0
-10
-20
-30
h
-70 -60 -50 -40
-30
-20 -10 0 10 20
After-Vehicle Scan
30
40
50
60
256
240
224
208
192
176
160
144
128
112
96
80
64
48
32
16
0
70
CM
O
U
T3
ED
k—
~
in
re
ED
c
~
o
u
¦3
><
h.
c
_ra
CO
/prqj 1/EDARinD en ver-OCT2019/Analysis_M Lout/211108/Anal_MLout/OCT19_bumper.sas 01DEC21 11:32
6-7
-------
Figure 6-8. CO2 ScanSums and Blank Pixel Counts for a Vehicle with Trailer
(19.8 mph, Series_Transit=505_000262)
-20 -10 0 10 20
After-Vehicle Scan
/proj 1 /E D ARin Den ver-OCT2019/Analysis_MLout/211108/Anal_MLout/OCT19_bumper_trailers.sas 19NOV21 12:37
CM
O
u
¦o
cu
D
in
ra
0)
c
D
O
u
ai
><
CL
C
CD
Figure 6-8 shows a result for a vehicle pulling a trailer with a footprint between After-Vehicle
Scan = -52 and 1 (i.e., 52 scans) traveling at 19.8 mph for a calculated vehicle/trailer length of 75
feet. The red CO2 scansum trace clearly shows CO2 at the hitch, zero CO2 on the sides of the
trailer because the trailer blanks out virtually all of the 256 pixels, and a large CO2 signal behind
the trailer.
Footprint width - To estimate the width of the vehicle footprint, the length in feet of each scan
with blank pixels is calculated as the length of the retro-reflective pavement tape (12 feet) by the
number of blank pixels in each scan to the number of pixels the number of pixels in each scan
(256). Then, we assign the second longest scan length in feet to the vehicle footprint width.
Using the second longest scan length reduces the influence of protrusions from the vehicle body
such as side-view mirrors.
Vehicles with trailers - Once we developed an automated method to locate the front and rear of
a vehicle in a transit, we wrote code23 to attempt to identify vehicles that were pulling trailers.
The code looked for abrupt, large changes in the number of blank pixels in the scans between the
front bumper and the rear bumper as an estimate of the presence of a hitch. About 159 transits of
the Westminster dataset were identified by this method as having trailers. Because we decided to
put understanding the emissions of vehicles with trailers as part of future work, we did not
further investigate these transits.
23 P:\EDARinDenver-OCT2019\Analysis_MLout\211108/OCT19_bumper.sas
6-8
-------
Footprint size trends - For many of the transits produced by vehicles registered in Colorado,
the registration database contained various types of vehicle descriptions, for example, fuel type
(gasoline, diesel, electric, ...), vehicle type (car, motorcycle, incomplete, ...), body type (bus,
ambulance, convertible, pickup truck, ...), and GVWR. We examined vehicle size trends as a
function of these descriptors. For example, Figure 6-9 shows the footprint length and width as a
function of GVWR. The plots show that the footprint length and width tend to increase with
GVWR.
6-9
-------
Figure 6-9. Footprint Size as a Function of GVWR for the Westminster Set
-o
i
o
o
0 - 3,750 pounds
10 15 20 25 30
35
8,501 - 10,000 pounds
«
.
•
fix* '
. -V 1
i «•
5 10 15 20 25 30 35
-------
6.3 100% Illumination Speed for this Study
The 100% illumination speed (100%IS) is a characteristic constant for a given RSD instrument
geometrical set-up that is used to correct the RSD signal for the speed of the vehicle, as used in
Equation 5-14. This subsection describes how the 100%IS is calculated for the EDAR set-up
used in this study. The 100%IS for other RSD set-ups might be calculated similarly.
The purpose of Step 6 is to calculate the RSD 100% Illumination Speed (100%IS), which is the
road speed at which the vehicle/vortex would have to move to produce an RSD signal that would
equal the RSD signal produced if the RSD light beam illuminated the scan path once and only
once. The 100%IS is independent of other variables including vehicle, road speed, pollutant,
release rate, release location, and wind.
As shown in Figure 5-4, in the reference frame of the vehicle/vortex, the laser beam optically
samples the vortex with a zigzag pattern from above the roadway. Of course, the instrument can
make measurements only where pixels are illuminated, that is, where blue dots are. The pollutant
mass in the white areas between the scans and around the blue dots is not illuminated and
therefore does not contribute to the RSD signal. The pitch of the zigzag in the vehicle/vortex
reference frame is a function of the road speed. At higher speeds, the pitch is larger. After pre-
processing, the arrangement of the processed measurements is on a rectangular grid, but the
shape and size of the measurement region remains the same. Because the diameter of the laser
beam is independent of road speed, but the distance in the vortex between successive scans
depends on road speed, the fraction of the vortex that is illuminated by the light beam depends on
road speed.
In addition, the overlap of pixels is a function of vortex speed. At road speeds below the 100%IS,
all pixels have some degree of overlap with other pixels.
To calculate the 100%IS, the speed at which the total area illuminated by the RSD would be
equal to the total area of the scan path is determined. The scan path is the path that the RSD
instrument is scanning. The fraction of vortex illuminated is a function of the road speed and
characteristics of the RSD instrument. When the RSD instrument scans the roadway from above:
Fraction of Vortex Illuminated = Area of Illumination for 1 second Equation 6-1
Area of Scan Path for 1 second
The Area of Illumination for 1 second is the sum of the areas illuminated by each pixel for all
pixels illuminated in 1 second. The areas of all pixels are to be summed even if pixels overlap.
This is appropriate since while overlapping pixels cause over-sampling of the pollutants in the
vortex, overlapping pixels do contribute to the RSD signal.
An example serves to illustrate Equation 6-1. Suppose the vortex, which moves at the same
speed as the vehicle, is moving at 25 m/s. The scan length is 3.66 m (=12 feet), which is the
length of the retro-reflective tape on the pavement. Thus, the Area of the Scan Path for 1 second
is 91.5 m2 (=25m * 3.66m). Since the RSD laser beam has a radius of 1.0 cm, scans the retro-
reflective tape 20 times per second, and each scan has 256 pixels, the Area of Illumination for 1
second is 1.61 m2 (=71*(0.01m)2 * 20 * 256). Thus, the Fraction of Vortex Illuminated is 1.75%
(=1.61 / 91.5). This means that the mass of pollutant in the vortex moving at 25 m/s past the
6-11
-------
instalment is actually 57 times larger (=1/0.0175) than the mass illuminated by the RSD laser
beam.
This example shows that the fraction of the vortex illuminated is given by:
Fraction of Vortex Illuminated Equation 6-2
= Number of Pixels/Scan * Effective Pixel Area (m2/pixeO * Scan Rate (scan/s)
Road Speed (m/s) * Scan Length (m)
Note that Equation 6-2 is specific to this RSD instrument, which scans the full width of the
vortex from above the pavement and which is accordingly believed to obtain a representative
optical sample of the vortex. For other types of RSD instruments that are believed to obtain a
representative optical sample, Equation 6-2 would be replaced with a different appropriate
relationship. RSD instruments that make measurements using a horizontal beam at a fixed height
above the pavement may or may not be able to get a representative optical sample of the vortex.
To derive an expression for the 100%IS, Equation 6-2 is simply solved for Road Speed for a
value of Fraction of Vortex Illuminated =1, which produces Equation 5-13.
For the RSD set-up used in this study, the 100%IS would be calculated as:
256 pixels/scan * n *(0.01 m)2/pixel * 20 scan/s
3.66 m
which equals 0.44 m/s (=0.97 mph).
6.4 Vortex Entrainment Time (VET) Functionality
Vortex Entrainment Time (VET) is key to this method for measuring pollutant emission rate
using RSD because it establishes a connection between the RSD-measured pollutant Mass in
Vortex and pollutant Release Rate (g/hr) as described by Equation 5-17.
The Vortex Entrainment Time (VET) can be defined by a re-arrangement of Equation 5-17:
VET (hr) = Mass in Vortex (g) Equation 6-3
Release Rate (g/hr)
The VET is influenced by 1) the volume of the vortex, and 2) the efficiency of entrainment of
emissions released from the vehicle into the vortex. Consider Equation 6-3 for the case of a large
vehicle and a small vehicle that have the same pollutant release rates. The larger vehicle will
have a larger vortex, which, at equilibrium, will contain a larger pollutant mass and produce a
larger RSD signal than the smaller vehicle will. Accordingly, the larger vehicle will have a larger
VET. Now, consider another comparison of two vehicles of the same size and shape having the
same pollutant release rate, but one vehicle has the release at the vehicle rear and the other
vehicle has the release under the hood. The vehicle with the under-the-hood release will likely
have a smaller VET since a smaller portion of its release is likely to become entrained in the
6-12
-------
vortex compared to the vehicle with the rear release - especially if the vehicles are in a strong
cross wind.
Characteristic VET values depend on the size and shape of the vehicle, the location of the release
from the vehicle, and the speed and direction of air moving across the vehicle (i.e., the vehicle's
air velocity). VET values do not depend on the pollutant release rate. Thus, methods to compute
the VET value can use the physical outline of the vehicle in the measurement data, the spatial
locations of large-amplitude plume components in the processed measurements, and the wind
velocity as measured by the RSD system during the transit event.
The dependencies of VET on vehicle properties, vehicle operation, pollutant release location,
and ambient conditions were studied using the test vehicle data from the September 2016 and
October 2019 studies. Test vehicles with metered pollutant releases were driven past the RSD.
For each transit, the release rate was metered and the mass in the vortex was calculated from the
RSD data. Then, Equation 6-3 was used to calculate the VET for each transit.
The analysis that is described below effectively separates the dependence of VET into three
multiplicative factors for: 1) vehicle air speed, 2) the location on the vehicle of emissions
released, and 3) the drag area of the vehicle.
Air Speed Effects
The first step in examining VET functionalities looks at the effects of air speed. Because all
fossil-fueled vehicles have exhaust plumes containing high concentrations of CO2, because most
light-duty vehicles release exhaust at a location at their rear, and because RSD instruments get
strong signals from the CO2 in exhaust plumes, we will analyze the trends of VETs calculated
from CO2 signals to determine how vehicle air speed influences VET. Indeed, as will become
apparent later, we will use the CO2 in light-duty vehicle exhaust plumes as a reference for
evaluating the effect of emissions location on VET.
In this October 2019 study, exhaust gas was metered from EV-1 and EV-2, which were the
electric vehicles that had been fitted with fake tailpipes and bottle gas. Therefore, the CO2 from
EV-1 and EV-2 can be used to calculate VETs. The only difference in the bodies of EV-1 and
EV-2 was that EV-1 had the fake tailpipe on the left rear corner and EV-2 had the tailpipe on the
right rear corner. While exhaust gas was released from all five test vehicles, the exhaust gas from
the F150, GMC, and Subaru was natural engine exhaust, and its composition and flow rate was
neither metered nor measured. Therefore, the exhaust data from those three vehicles cannot be
used to calculate VETs.
The calculated VET values from EV-1 and EV-2 exhaust CO2 releases are plotted with red and
blue symbols against the AirSpeed Para in Figures 6-10 and 6-11. Clearly, VETs are higher for
lower air speeds. Regression analysis of this CO2 VET data, as well as the analysis of EvapHC
VET data on all five test vehicles, indicated that VETs followed a trend that was approximately
proportional to the inverse of the square root of the AirSpeed Para. The solid lines in Figures 6-
10 and 6-11 are fits using that functionality. The figures show substantial scatter in the VET
values. Because the flow of air around the vehicle is complex, turbulent in the vortex, and varies
across replicates, the VETs vary to produce the observed scatter. However, on average over a
6-13
-------
period of time, VET values for a given operating condition converge to a finite, repeatable value.
The figures show that a typical value for emissions from the rear of a light-duty vehicle with a
30-mph airspeed is around 4 seconds.
We examined the ratio of the measured VET to the fit VET to determine if the residual scatter in
the measured VETs was dependent on AirSpeed Perp, that is, the speed of the air blowing
perpendicular to the direction of vehicle motion. Figures 6-12 and 6-13 show the ratios plotted
against AirSpeed Perp. The plots show that for EV-1 the ratios increase with increasing
AirSpeed Perp, and for EV-2 they decrease with increasing AirSpeed Perp.
Those trends can be understood by considering the fake tailpipe location and the direction of
cross air movement. Positive values of AirSpeed Perp represent air blowing toward the left side
of the vehicle; negative values of AirSpeed Perp represent air blowing toward the right side of
the vehicle. EV-1 had its fake tailpipe on the left rear corner of the vehicle. Therefore, positive
values of AirSpeed Perp tend to move exhaust gas toward the centerline of the vortex and
thereby tend to increase the mass of CO2 in the vortex, which in turn produces a higher VET
value. Negative AirSpeed Para values tend to move EV-l's exhaust gas farther to the left and
therefore away from the vortex, which tends to decrease the VET value. Because EV-2's fake
tailpipe is on the right rear, the trend of the ratio measured VET to predicted VET is opposite of
that for EV-1. While the scatter in Figures 6-12 and 6-13 is substantial, it appears that cross air
movement of 3 mph typically causes changes of around ± 30% with respect to the VET value
with no cross-air movement.
The results in Figures 6-12 and 6-13 show that as AirSpeedPerf gets large the VET value is
affected. Even more extreme values of AirSpeed Perp could reduce the VET value to zero, which
essentially means that the release is blown so strongly to the side that it has no chance of
becoming entrained in the vortex. We expect that the effect is worse for emissions release
locations farther forward on the vehicle, such as for under-hood evaporative emissions releases.
For strong side winds, the EDAR instrument would have little chance of obtaining a usable
emissions signal. An EDAR flag could be developed to avoid outputting results when cross
winds are likely to blow plumes away.
Low AirSpeed also can be a risk to good RSD measurements. All RSDs are based on
measurement of the pollutants entrained in a vortex that follows a vehicle. At low vehicle air
speeds, it may be possible that no vortex forms at all or that the vortex is so poorly defined that
RSDs cannot get reliable measurements. Low AirSpeeds can occur when a vehicle is moving
slowly in calm air, but they can also occur when a vehicle is driving at normal speeds but in a
strong tailwind.
6-14
-------
Figure 6-10. Exhaust CO2 VET vs. AirSpeed Para for EV-1
(/)
a)
ia
re
OJ
a>
CL
CM
O
u
LU
>
11
10
9
8
7
6
5
4
3
2
1
0
--
A ^ A *A*
4 * K a A
4 AA 4^A . a
aA ^777?
A A AA. ii1 4a
a V*£a Ai 4 A A
^ A 4 i\
10 20 30 40 50
AirSpeed Para (mph)
/proj 1 /E DARi nDenver-OCT2019/Analysis _MLout/211220/Anal JVILout/OCT19_TOT_2_TH D.sas 28JUL22 13:17
60
Figure 6-11. Exhaust CO2 VET vs. AirSpeed Para for EV-2
w
10
1/)
re
v
CL>
a:
11
10
9
8
7
6
5
4
3
2
1
A A V -
A 1 ^A A
* VaA 4
10 20 30 40 50 60
AirSpeed Para (mph)
/proj 1/EDARinDenver-OCT2019/Analysis_M Lout/211220/Anal_MLout/OCT19_TOT_2_THD.sas 28JU L22 13:17
6-15
-------
Figure 6-12. Residual VET Trend vs. AirSpeed Perp for EV-1
LU
>
-o
0)
+-»
u
a>
Q_
LU
>
"O
CL>
3
10
re
a>
2.0
.8
.6
.4
.2
.0
0.8
0.6
0.4
0.24
-10
• i
" 1
•
. •>
•
• •
• • •
% • • • •
~ . . •
r * • . ••
#? \ « *
* ••
• • <
• #• • •
* *%•
•
• v ¦ • •
x* •
• •• • •• ,
- • • •
•
•
-5 0 5 10
AirSpeed Perp (mph)
/proj 1/EDARinDenver-OCT2019/Analysis_M Lout/211220/Anal_MLout/OCT19_TOT_2_THD.sas 28JUL22 13:17
15
LU
>
¦o
-o
¦o
3
(/)
re
-------
Release Location Effect
To determine the effect of release location, we need to have RSD transits with metered releases
from different release locations at different air speeds. In this October 2019 study, artificial
exhaust and artificial EvapHC emissions were metered in releases from the five test vehicles.
The dependence of VET on emissions release location and vehicle AirSpeed Para was
determined from these releases.
Artificial evaporative emissions (propane) were metered from all five test vehicles at three
different locations and at 6400, 3200, 1600, 800, 400, 200, 100, and 50 mg/mile emission rates.
Some of the resulting tests can be used to calculate VETs for releases from the locations as long
as natural ExhHC emissions and natural EvapHC emissions are small relative to the artificial
evaporative releases.
For EV-1, the artificial exhaust contained no HC, and since EV-1 was an all-electric vehicle, it
had no gasoline on board. Its evaporative HC should be quite small since it would be derived
only from outgassing of HC from vehicle construction materials. However, we also used only the
test data from the 6400, 3200, and 1600 mg/mile artificial evaporative emission rates from EV-1
to maximize the signal-to-noise ratio of the RSD HC signals. Finally, we could use no EV-2
evaporative test data to calculate VETs because EV-2's artificial exhaust HC was quite high (402
ppm propane). Thus, none of EV-2's artificial evaporative HC emission rates were substantially
higher than its exhaust HC emission rate.
Similarly, we used only the test data from the 6400, 3200, and 1600 mg/mile artificial
evaporative emission rates for the F150, GMC, and Subaru test vehicles to ensure that the
artificial evaporative release rates were substantially larger than the natural ExhHC and EvapHC
emissions from those vehicles.
The calculated VET values from the 6400, 3200, and 1600 mg/mile artificial evaporative
releases from EV-1, F150, GMC, and Subaru test vehicles are plotted in Figures 6-14, 6-15, 6-
16, and 6-17. For EV-1 and Subaru, the three release locations were at the fuel fill door
(Door=purple) at the rear of the quarter panel, on top of the under-vehicle fuel tank
(Tank=orange), and on top of the engine under the hood (Hood=green). For the F150 and GMC,
the fuel fill door (Side = blue) was just aft of the driver's door on the left side of the vehicle.
Test regressions of the trends of VETs for artificial EvapHC (EV-1, F150, GMC, and Subaru) as
a function of release location and AirSpeed Para indicated, just as for CCh-based VETs, that
VET was inversely proportional to approximately the square root of the AirSpeed Para. In
addition, the regressions indicated that the proportionality constants were connected to the
release location. Based on those findings, we used regression to determine the proportionality
constants for all of the selected test vehicle transits. For the EvapHC releases, we used regression
weights proportional to the 6400, 3200, and 1600 mg/mile emission rate values to account for the
less variable VETs associated with the higher emission rates. The resulting proportionality
constants are shown in the fourth column of Table 6-1. The first three columns of Table 6-1
show the test vehicle identifier, drag area, and emission release location for the data under
consideration. The solid curves in Figures 6-14, 6-15, 6-16, and 6-17 show the regression fits of
the measured VETs.
6-17
-------
Figure 6-14. VET vs. EvapHC Release Location and Air Speed for EV-1
11:
l/l
a
V)
re
_cu
cu
0;
U
X
Q.
re
>
LU
I—
LU
>
10
9
8
7
6
5
4
3
2
1
0
10 20 30 40 50
AirSpeed Para (mph)
60
LEGEND ' » ' DOOR Data
DOOR Fit
1 HOOD Data
' HOOD Fit
TANK Data
TANK Fit
/proj1yEDARinDenver-OCT2019yAnalysis_MLout/211220/Anal_MLoutfOCT19_TOT_2_THD.sas 03SEP22 08.29
Figure 6-15. VET vs. EvapHC Release Location and Air Speed for F150
w
10
o
i/)
re
_Q)
0)
CL
U
X
Q.
re
>
LU
LU
>
11
10
9
8
7
6
5
4
3
2
1
0
10 20 30 40
AirSpeed Para (mph)
50
60
LEGEND * 1 » SIDE Data
SIDE Fit
1 HOOD Data
' HOOD Fit
TANK Data
TANK Fit
/proj1/EDARinDenver-OCT2019/Analysis_MLout/211220/Anal_MLout/OCT19_TOT_2_THD.sas 03SEP22 08:29
6-18
-------
Figure 6-16. VET vs. EvapHC Release Location and Air Speed for GMC
11
LU
>
10 20 30 40 50
AirSpeed Para (mph)
60
LEGEND » * » SIDE Data
SIDE Fit
' HOOD Data
¦ HOOD Fit
TANK Data
TANK Fit
/proj1/EDARinDenver-OCT2019/Analysis_MLout/211220/Anal_MLout/OCT19_TOT_2_THD.sas 03SEP22 08:29
Figure 6-17. VET vs. EvapHC Release Location and Air Speed for Subaru
w
l/l
CD
V)
ro
_cu
(0
Ol
U
I
Q.
to
>
LU
LU
>
11
10
9
8
7
6
5:
4
3:
2
1:
o:
10 20 30 40
AirSpeed Para (mph)
50
60
LEGEND ' » * DOOR Data
DOOR Fit
1 HOOD Data
' HOOD Fit
TANK Data
TANK Fit
/proj1/EDARinDenver-OCT2019/Analy5is_MLouV211220/Anal_MLoutfOCT19_TOT_2_THD sas 03SEP22 08'29
6-19
-------
Table 6-1. Release Location Factors for Test Vehicles
Test
Drag Area
Release
Proportionality
Constant
Release
Location
Factor
Vehicle
(ft2)
Location
VET vs. sqrt
(AirSpeed Para)
EV-1
8.0
Tailpipe C02
20.2
=1.00
EV-1
8.0
Door
20.4
1.02
EV-1
8.0
Tank
12.1
0.60
EV-1
8.0
Hood
7.2
0.36
EV-2
8.0
Tailpipe C02
16.9
EV-2
8.0
Door
n/a
n/a
EV-2
8.0
Tank
n/a
n/a
EV-2
8.0
Hood
n/a
n/a
F150
Tailpipe C02
n/a
n/a
F150
Side
8.2
F150
Tank
15.0
F150
Hood
9.9
GMC
Tailpipe C02
n/a
n/a
GMC
Side
10.6
GMC
Tank
12.2
GMC
Hood
4.2
Subaru
10.2
Tailpipe C02
n/a
n/a
Subaru
10.2
Door
22.0
Subaru
10.2
Tank
15.4
Subaru
10.2
Hood
8.7
To get a better view of the trends in the proportionality constants, they are shown in Table 6-2 in
a vehicle-versus-release-location grid.
Table 6-2. VET Proportionality Constant vs. Vehicle and Release Location
EV-1
EV-2
F150
GMC
Subaru
Tail
20.2
16.9
Door
20.4
22.0
Tank
12.1
15.0
12.2
15.4
Side
8.2
10.6
Hood
7.2
9.9
4.2
8.7
Table 6-2 shows a trend of decreasing VET proportionality constant as location moves from
vehicle rear to front (Tail Door Tank Side Hood). This makes physical sense since plumes of
releases tend to spread horizontally and vertically as they move away from their release point.
Thus, a smaller portion of pollutant releases from the front of a vehicle is likely to become
entrained in the vortex that follows the vehicle rear in comparison with releases from the vehicle
rear.
We would like to estimate the relative VET proportionalities among the five release locations
even though only 14 of the 25 cells in Table 6-2 have measured values of the VET
proportionality constant. We smoothed the relative VET proportionalities by assuming that the
6-20
-------
VET proportionality constants differ by location factors across the five locations and also differ
by vehicle factors across the five vehicles. We built a categorical model24 that used these
assumptions to arrive at the fitted VET proportionality constants given in Table 6-3.
Table 6-3. Fit of VET Proportionality Constant vs. Vehicle and Release Location
EV-1
EV-2
F150
GMC
Subaru
Tail
20.2
16.9
22.6
17.3
23.9
Door
19.5
16.3
21.8
16.6
23.0
Tank
13.2
11.0
14.7
11.3
15.6
Side
9.5
8.0
10.7
8.1
11.3.
Hood
6.9
5.8
7.7
5.9
8.2
The model had an r-square of 0.88, an F-value of 144, and the effect of release location was
significant at greater than the 99% confidence level. Even though the model did not find that
vehicle ID was a statistically significant factor, we left it in the model since we know that
vehicles with different drag areas will have different VET proportionality constants. Comparison
of the measured VET constants in Table 6-2 with the corresponding fit values in Table 6-3 shows
good agreement. The agreement for the F150 and GMC test vehicles, which had fewer transits
on which to base their VET constants, is not quite as good as for the other test vehicles.
We then used any column from Table 6-3 to calculate the VET proportionality constant relative
to the tailpipe (Tail) location as shown in Table 6-4. The Relative VET factors for release
location are appealing because they directly relate to the relative front-to-rear location of the
release point: Hood releases are effectively at the firewall (one-third to the rear: 0.34 factor);
Sides releases (one-half to the rear, 0.47 factor); Tank releases (two-thirds to the rear, 0.67
factor); quarter-panel fuel-fill-Door releases (almost all the way to the rear, 0.96 factor), and
Tailpipe releases (at the rear, 1.00 factor).
Table 6-4. Relative VET Proportionality vs. Release Location
Release Location
Relative VET
Tail
=1.00
Door
0.96
Tank
0.65
Side
0.47
Hood
0.34
It is important to note that none of the factors are zero and the factors range from 0.3 to 1. This
indicates that there is always a significant chance that a portion of emissions released from a
vehicle can get into the vortex that follows the vehicle and that the emissions will always
potentially be present for detection by an RSD instrument.
24 P:\EDARinDenver-OCT2019\Analysis_MLout\220817\Anal_MLout\RefVehs/ RelVET.sas
6-21
-------
Drag Area Effect
In the September 2016 study, artificial EvapHC was metered at a constant Emission Rate (10,913
mg/mile) from four late-model, light-duty test vehicles while driving the vehicles at four
different road speeds. That emission rate was chosen to be large so that any natural EvapHC or
ExhHC emissions from the test vehicles would be trivially small in comparison.
Table 6-5 shows the September 2016 study results of the 50 transits made on each of the four test
vehicles with a metered, constant EvapHC Emission Rate. The results show that vehicles with
larger drag areas had higher VETs than those with lower drag areas. The results also show that
for these four light-duty vehicles, drag area had a mild effect on VET. Specifically, the ratio of
largest to smallest drag area was 2:1, but the ratios of VETs was only 1.3:1. Thus, for this dataset
VET appears to be proportional to approximately the one-third root of the drag area.
Table 6-5. Relative VET by Vehicle Drag Area
Test
Drag Area
Relative
Vehicle ID
(ft2)
VET
1
7.2
0.85
3
10.7
1.10
4
6.7
0.95
5
13.4
1.06
Overall, the analysis indicates that light-duty vehicle VETs follow the relationship given by
Equation 5-8 with VET proportional to the relative location of the release point from front to
rear, proportional to the one-third root of the drag area, and inversely proportional to the square
root of AirSpeed Para. We have not yet quantified the dependence of VET on AirSpeed Perp.
6.5 Vortex Shape (Weights) Functionality
Because in most instances evaporative emissions from well-maintained, latest-technology
vehicles are quite low, attempting to quantify or even just detect evaporative emissions using
remote sensing devices is challenging. While RSD instruments can make thousands of detailed
measurements around each vehicle driving past the instrument, each individual measurement has
low signal and high noise. We need to find ways to enhance RSD signals.
When we think about it, we realize that we do not need to know exactly where the emissions are,
that is, we do not need to analyze the emissions location in great detail - even though RSDs
provide such information. We anticipate that a portion of emissions from vehicles get entrained
in the vortex and otherwise disperse behind the vehicle in patterns that may be predictable. If we
could predict the pattern, then we could use the pattern as a template for quantifying the
emissions in the pattern or cloud of emissions and thus enhance the signal-to-noise ratio.
In this subsection, we analyze the patterns of emissions of vehicles moving through the air to
understand the pattern or shape of the emissions clouds around moving vehicles. We minimize
the effects of noise for this analysis by using either large artificial gas releases or large natural
6-22
-------
vehicle emissions (exhaust CO2). And we do this with the knowledge that, whether the release
rates are tiny or large, the patterns are the same - even though the magnitude of emissions in the
pattern is different. Thus, once we understand the shape well enough to predict it, we can use the
pattern to quantify the emission magnitude - to convert the thousands of detailed measurements
collected for a single vehicle transit into a single emissions value.
Early Trends from September 2016 Data
Although the scientific literature contains results of many computational fluid dynamics (CFD)
studies that detail how gases flow around moving bodies, for this methodology a more general
description of the dependence of RSD signals from pollutants in the vortex on vehicle operation
and pollutant release characteristics is preferred.
The weight functionality characterizes the shape of the vortex as viewed by the RSD when
pollutants are released from a vehicle under a range of conditions that are expected from in-use
vehicles.
The scansum, which is the sum of all pixels in each RSD scan for each pollutant, is calculated.
Figure 6-18 shows scansum time traces for fifteen transits in the September 2016 study when
artificial evaporative HC (butane) was released at 10,913 mg/mile from a test vehicle when it
was driving under the RSD at 37.5 mile/hour. Each dot in the figure is one scansum. The fifteen
transits are made up of four replicate transits with releases from the fuel fill door (Door=purple),
under the hood (Hood=green), on top of the fuel tank (Tank=orange), and inside the left-rear
wheel well (Well=red).
The traces in Figure 6-18 have been aligned in time at the vehicle rear (Scan=0). The traces in
the figure are generally made up of a peak at about Scan 2. After the peak, the traces decay in a
variety of paths toward a scansum of zero. This diversity of scansum trace shapes is attributed to
turbulence and instrumental noise. Because of this diversity and to get a clearer picture of the
overall tendency of the scansum trace shape, the 207 traces for 10,913 mg/mile HC releases were
averaged across all four test vehicles, all four test speeds, all four artificial evaporative HC
release locations, and all replicates to produce a grand mean trace, which is shown by the black
dots on a linear scale in Figure 6-19 and a log scale in Figure 6-20. These figures show that after
the initial peak the RSD signal tends to decay exponentially.
6-23
-------
Figure 6-18. ScanSum Traces for Vehicle with 10,913 mg/mile HC at 37.5 mph
(N
E
o
E,
E
3
l/>
C
re
u
l/>
U
0.02
After-Vehicle Scan
/proj 1/EDARatTTI-SEP2016/RunLoss/Analysis/TTI_analysis_HCgi-p3_3.sas 13JUL22 10:56
To examine the influences on the grand scansum shape, the 207 scansum time traces were
averaged in categories of vehicle identity, road speed, and HC release location. Figures 6-21, 6-
22, and 6-23 show the log of the average scansum time traces for the four different levels of each
of those categories. Each curve is the average of approximately 50 traces. Just as seen in the
grand average log plot in Figure 6-20, the curves in Figures 6-21, 6-22, and 6-23 are
characterized by a peak at Scan=2 followed by an exponential decay. The plots indicate that all
levels of those categories have close to, but perhaps not exactly, the same scansum time trace
shape.
Note that while the shapes of the traces are quite similar, the magnitudes of the traces, as seen by
vertical shifts of curves in the log plots of Figures 6-21, 6-22, and 6-23 do vary with the different
levels of vehicle, road speed, and pollutant release location even though the HC emission rate
was a constant 10,913 mg/mile.
Overall, the analysis of the first measurement dataset with 207 RSD transits on four test vehicles
indicated that the shape of scansum time traces depends on time after the vehicle rear and seems
to be independent of road speed and vehicle shape. Each scansum time trace shape is
characterized by a peak at Scan 2 followed by an exponential decay.
6-24
-------
Figure 6-19. Grand Average HC ScanSum Trace for HC Releases
10 20 30
After-Vehicle Scan
/proj1/EDARatTTI-SEP2015/RunLoss/An£]lysis/TTI_an£]lysis_HCgrp3_5.S£]s 13JUL22 09:41
Figure 6-20. Log of Grand Average HC ScanSum Trace for HC Releases
CM
E
Id
0
E,
E
3
(/)
C
to
u
(/)
U
1
0.200
0.020
0.002
10 20 30 40
After-Vehicle Scan
/proj1/EDARatTTI-SEP2016/RunLoss/Analysis/TTI_analysis_HCgrp3_5.sas 13JUL22 09:41
6-25
-------
Figure 6-21. Log of HC ScanSum Traces Averaged by Vehicle ID
0.200
c
re
u
(/>
u
<0
U)
ro
1-
d>
5
0.020
0.002
10 20 30
After-Vehicle Scan
Vehicle ID — i — 3 —4 — 5
/proj 1/EDARatTTI-SEP2016/RunLoss/AnalysisfiTI_analysis_HCgrp3_5.sas 13JUL22 09:41
Figure 6-22. Log of HC ScanSum Traces Averaged by Road Speed
0.200
CM
E
o
E,
E
3
(/>
c
rc
u
(/>
u
<0
U)
ro
1-
d>
5
0.020
0.002 ^
-10
10 20 30
After-Vehicle Scan
Speed (mph) — 12.5 — 25 — 37.5 — 50
1 proj 1/EDARatTTI-SEP2016/RunLoss/AnalysisyTTI_analysis_HCgrp3_5.sas 13JUL22 09:41
6-26
-------
Figure 6-23. Log of HC ScanSum Traces Averaged by Release Location
Evap Location Door Hood Tank Well
/proj 1 /EDARatTTI-SEP2016/RunLoss/Analysis/TTI_analysis_HCgrp3_5.sas 13JUL22 09:41
Refined Trends from Westminster October 2019 Data
Analysis of the September 2016 staged data taken on test vehicles indicated that the shape of
scansum traces for both exhaust and evaporative emissions in the vortex had the same shape and
could be described as a peak followed by an exponential decay. We wanted to confirm that
finding using the Westminster data. Further, we wanted to determine the dependence of the
shape of scansum traces on vehicle and vehicle operation variables.
The fleet vehicles that produced the 30,000 transits in the Westminster dataset have a variety of
shapes and sizes and most use gasoline or diesel fuel and have exhaust plumes containing CO2.
Therefore, we used the CO2 data collected by the EDAR instrument to examine influences on the
shape of the scansum traces.
The CO2 scansum traces were used to quantify the scansum trace shape. However, to provide
accurate scansum shapes, the CO2 pixel zero baseline must be accurate. The MatLab processing
of the raw EDAR data used a histogram of the log of the pixel counts to initially zero-adjust the
measured CO2 detailed data. These initial MatLab baseline offset adjustments will be described
in Section 7.1. We then used a SAS program25 to verify and further adjust the zero of the CO2
scansums.
25 P:\EDARinDenver-OCT2019\Analysis_MLout\211122\Anal_MLout/OCT19_shapeC02.sas
6-27
-------
We provided the second and final zero-baseline adjustment by considering the CO2 scansums
between the last-before-vehicle scan and the first-after-vehicle scan. Since exhaust CChis not
usually present in front of the tailpipe exit, all of these CO2 scansums should be zero. The SAS
program found the median of the scansums in this zone and adjusted all CO2 pixels for the transit
to make the median be exactly zero.
Figure 6-24 shows an example of the procedure. The red scansum trace shows that the sum
provided by the initial MatLab zeroing is slightly negative during the vehicle footprint. The
green scansum trace is after the median scansum value (at the small orange triangle) is set to
zero.
Figure 6-24. Example Showing CO2 ScanSum Zero Adjustment
224
J/J
Q)
192 X
'£
CM
160 O
u
128 J
CO
M-
96 0
b.
V
-Q
64 E
z
32
/proj 1/EDARinDenver-OCT2019/Analysis_M Lout/211122/Anal_MLout/OCT19_shapeC02.sas 21DEC21 10:51
-20 -15 -10 -5 0 5 10 15 20 25
After-Vehicle Scan
Signal Adj C02 (CORR zero?) zeroed_co2_mole_scan
zeroed_co2 mole scan ftpt 1» ' medianscannum
We expected that vehicle air speed has a more intimate influence on vortex shape than vehicle
road speed since the movement of air around the vehicle body produces the vortex. Further,
based on our examination of average plume contours for different air speed directions, we
expected that the component of the air speed parallel to the major axis of the vehicle is the most
important component to the formation of the vortex. Figure 6-25 shows a histogram of the air
speed parallel component for fleet vehicle transits in the Westminster dataset.
We also expected that the scansum trace shape could also be influenced by the vehicle length.
Figure 6-26 shows a histogram of vehicle length for the fleet vehicles in the Westminster dataset
as estimated as described in Section 6.2. Most vehicles in this dataset have a length between 9
and 26 feet.
6-28
-------
Figure 6-25. Distribution of the Parallel Component of the Vehicle Air Speed
CUM. CUM.
0 100 200 300 400 500 600 700 800 900 1000 1100 1200 1300 1400 1500 1600
FREQUENCY
fcroj1/EDARinDenver-OCT2019/Analysis_MLout/211122/Anal_MLoutfOCT19_interpshapeC02_2.sas 09MAY22 15:09
Figure 6-26. Distribution of Vehicle Length for Westminster Fleet Vehicles
CUM. CUM.
0 1000 2000 3000 4000 5000
FREQUENCY
/proj 1 /E DARi nD enver- OCT2019/An alysi s_MLout/211122/Ana I MLo ut/OCT19_i nterp sha peC02_2. sas 09 MAY22 15:09
6-29
-------
Figure 6-27 shows a scatter plot of all transits as a function of Vehicle Length and AirSpeed
Para. Black symbols represent fleet vehicles, and red symbols represent test vehicles. Figure 6-28
zooms in to the region of abundant data.
Because turbulence is always present behind a moving vehicle, scansum shapes are variable -
even for replicate transits. Accordingly, we examined averages of shapes of CO2 scansum traces.
We averaged CO2 scansums of transits in categories of Vehicle Length and AirSpeed Para.
Figure 6-29 shows the averages of CO2 scansum traces for the Westminster fleet transits in
categories of Vehicle Length. Figure 6-31 shows the averages taken for essentially the same
transits but in categories of AirSpeed Para. The curves in the two plots have the same general
shape - a peak followed by a decay. Figure 6-30 and Figure 6-32 show log versions of the same
data. These plots show that beyond After-Vehicle Scan 8 the decays are close to exponential
since the lines are straight. Also, the slopes of the straight portions for different levels of Vehicle
Length and AirSpeed Para are all about the same - except for low values of AirSpeed Para where
the data is less abundant.
At this point in the analysis, we are concerned mainly about the shape of these average CChtime
traces - not their magnitudes since many things affect the relative magnitudes of the traces
including engine displacement, engine RPM during the transit, and frontal area of the vehicle.
We present the analysis of the shape of the time traces.
6-30
-------
Figure 6-27. Full Distribution of Vehicle Length and Parallel Air Speed
110
100
90
80
70
U\
C 60
50
"u
CD
>
0 10 20 30 40 50 60 70 80 90 100 110 120
Air Speed Parallel Component at 1 m (mph)
RefVeh • • • 0 • • • 1
/proj1/E DARin Den ver-OCT2019/Analysis_M Lou t/211122/Anal_M Lout/O CT19_shapeCO 2.sas 06JAN22 15:39
Figure 6-28. Zoomed Distribution of Vehicle Length and Parallel Air Speed
30
±3 20
U)
C
0)
-I
0)
.y 15
-C
QJ
>
10
• :;\vaU-
:'A">'
V Vvv _ ¦¦
10 15 20 25 30 35 40 45 50 55 60 65 70 75
AirSpeed Parallel Component at 1m (mph)
/proj 1 /E DARin Denver-OCT2019/An alysis_M Lou 1/211122/An al_MLou t/OCT19_sh apcCO 2. sas 06JAN22 15:39
6-31
-------
Figure 6-29. CO2 Mass Trace Averaged by Vehicle Length
400
£
5 350
(/)
C 300
l/l
1/1 250
re
^ 200
rsi
8 150
a)
> 100
*-P
re
0) 50
Ol
0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
After-Vehicle Scan
Vehicle Length (ft) —00-09 •-•-•09-11 •-•-•11-13 13-16
—- 16-19 19-22 —" 22-26 26-44
/proj 1IEDARinDenver-OCT2019/Analysis_MLouU211122/Anal_MLout/OCT19_interpshapeC02_5.sas 03SEP22 10:23
Figure 6-30. Log CO2 Mass Trace Averaged by Vehicle Length
1000.0
c
re
u
1n
c 100.0
(/)
(/)
re
^ 10.0
CM
O
u
1.0
_re
<0
a:
0.1 n , , , , , , , , , , , , , , ,
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
After-Vehicle Scan
Vehicle Length (ft) —00-09 «-• 09-11 *-*-Mi-i3 13-16
—- 16-19 19-22 —— 22-26 -"26-44
/proj 11EDARinDenver-OCT2019/Analysis_MLout/211122/Anal_MLoutfOCT19_interpshapeC02_5.sas 03SEP22 10:23
6-32
-------
Figure 6-31. CO2 Mass Trace Averaged by AirSpeed Para
After-Vehicle Scan
AirSpeed Para (mph) — 00-18 18-21 21-24 24-28 28-33
33-38 —- 38-45 m 45-52 52-60 60-99
/proj 1 /EDARinDenver-OCT2019/Analysis_MLout/211122/Anal_MLout/OCT19JnterpshapeC02_5.sas 03SEP22 10:23
Figure 6-32. Log CO2 Mass Trace Averaged by AirSpeed Para
/proj 1 /EDARinDeriver-OCT2019/Analysis_MLout/211122/Anal_MLout/OCT19_interpshapeC02_5.sas 03SEP22 10:23
After-Vehicle Scan
AirSpeed Para (mph) —00-18 •-•-•18-21 •-•-•21-24 24-28 «-*-¦ 28-33
33-38 —- 38-45 45-52 52-60 60-99
6-33
-------
Since the plots in Figure 6-30 and Figure 6-32 indicated that the average slopes beyond After-
Vehicle Scan 8 were close to the same value for Vehicle Length categories and AirSpeed Para
categories, we quantified the average exponential decay rate for the entire Westminster dataset
for After-Vehicle Scan 10 to 20 (0.5 to 1.0 s) using a regression model26. Figure 6-33 shows the
fit of the Westminster CO2 scansum traces to an exponential decay with a decay constant of -1.38
+- 0.03 (2 standard errors) s"1.
Figure 6-33. Fit of 30,559 CO2 ScanSum Traces to an Exponential Decay
0.2 0.3 0.4 0.5 0.6 0.7 0.8
Time After Vehicle Rear (s)
/proj1/EDARinDcnvcr-OCT2019/Analysis_MLout/211122/Anal_MLout/OCT19_intcrpshapcCO2_5.sas 03SEP22 10:23
We also determined the decay constants for AirSpeed Para and Vehicle Length strata of the
dataset. The analysis indicated that the decay constant was the same across all levels of those two
variables, which confirms the exponential trends seen in Figure 6-30 and Figure 6-32. Table 6-6
shows mean decay constants and the 95% confidence intervals for the strata and the entire
dataset. The table shows that the decay constant has no significant trend across the two
stratification variables, since the dataset mean value of-1.38 s"1 is within the 95% confidence
interval of almost all strata.
These results indicate that the stripping of pollutants from the low-pressure zone, which is quite
close to the vehicle rear, into the wake behind the vortex is a first-order process, which can be
expressed by the first-order differential rate law:
26 P:\EDARinDenver-OCT2019\Analysis_MLout\211122\Anal_MLout/ 0CT19_interpshapeC02_5.sas
6-34
-------
dM = k * M Equation 6-1
dt
where M = Mass of pollutant
t = time
The rate law says that the rate that the pollutant mass moves out of the low-pressure zone
(dM/dt) and into the wake behind the vortex is directly proportional to the mass of the pollutant
in the low-pressure zone. If the ambient air has a zero pollutant mass, the integrated form of
Equation 6-1 gives the time dependence of the mass in the tail of the vortex:
M(t) = Mo * exp (k * t) Equation 6-2
where Mo is the mass in the low-pressure zone just behind the vehicle.
The RSD instrument takes measurements at constant time intervals since it uses 20 scans per
second. Because the pollutants leave the vortex following the time dependence of Equation 6-2,
the RSD-measured decay of pollutants in the vortex appears to be independent of the
stratification variables. Thus, in terms of time, the vortex length is constant, but in terms of
distance, the vortex length is proportional to AirSpeed Para.
Table 6-6. Exponential Vortex Time-Decay Constants for Various Dataset Strata
Stratification
Bin
Ntransits
k Decay Constant (s1)
Variable
LCLM95
Mean
UCLM95
00-18
503
-1.38
-1.68
-1.98
18-21
422
-0.52
-1.04
-1.57
21-24
793
-1.30
-1.55
-1.80
AirSpeed
Para
(mph)
24-28
28-33
1610
3235
-1.37
-1.27
-1.51
-1.37
-1.64
-1.47
33-38
6119
-1.27
-1.34
-1.41
38-45
8989
-1.38
-1.43
-1.49
45-52
6598
-1.24
-1.30
-1.37
52-60
1980
-1.23
-1.34
-1.45
60-99
310
-1.11
-1.49
-1.86
00-09
166
-0.60
-1.09
-1.58
09-11
985
-0.99
-1.15
-1.30
11-13
4456
-1.33
-1.41
-1.49
Vehicle
Length
(ft)
13-16
16-19
11525
9053
-1.36
-1.34
-1.41
-1.40
-1.46
-1.45
19-22
3166
-1.22
-1.33
-1.45
22-26
917
-1.14
-1.28
-1.42
26-44
206
-0.96
-1.29
-1.62
44-73
77
-0.87
-1.46
-2.06
73-99
8
-0.66
-2.54
-4.42
None
All
30559
1.35
-1.38
1.41
6-35
-------
In the next modeling step, we divided all CO2 scansum traces by the exponential decay shown in
Figure 6-33 to determine the residual of the traces for further analysis. We assigned the residuals
to Vehicle Length bins and calculated the average residual trace and its 95% confidence limits
for each vehicle length bin.
As an example, Figure 6-34 shows the result for the 19-22 ft vehicle length bin. The blue and red
symbols show the lower and upper 95% confidence limits for the mean. The analysis revealed
that for all After-Vehicle Scans > 8, a constant value of the Length Factor for each Length Bin
plot would stay between the red and blue limits. For example, in Figure 6-34, a value of about
1.5 was within the upper and lower confidence limits for After-Vehicle Scans >8. Therefore, the
mean value at After-Vehicle Scan 8 or 9 was held constant for all later After-Vehicle Scans as
shown in the figure. For all Vehicle Length bins, these "padded" values remained inside the 95%
confidence limits for all Vehicle Length bins.
Figure 6-34. Average Residual CO2 Scansum Trace for 19-22 ft Length Bin
After-Vehicle Scan
Legend -—Mean • LCLM95 • UCLM95
/proj1/EDARinDenver-OCT2019/Analysis_MLout/211122/Anal_MLout/OCT19_interpshapeC02_5.sas 03SEP22 10:23
The results of this procedure are shown in Figure 6-35 for the eight Vehicle Length bins up to 44
feet. The top two bins (44-73 ft and 73-99 ft) had too few transits to provide reliable results.
Since the analysis focuses on the shape of the traces, each average unnormalized trace in Figure
6-35 was then normalized by dividing the trace by the asymptotic value that was used for After-
Vehicle Scans > 8. This produced the normalized traces by Vehicle Length bin shown in Figure
6-36. Figure 6-36 shows a constant asymptotic value of 1 for all After-Vehicle Scans > 8, and all
of the Vehicle Length effects are concentrated in the first several scans (Scans 0 to 6, 0.0 to 0.3
s) just behind the vehicle. Since the average AirSpeed Para value for each Vehicle Length bin
and for the entire dataset was near 40mph, the curves in Figures 6-35 and 6-36 are the effects of
Vehicle Length for airspeed Para values near 40 mph.
6-36
-------
Figure 6-35. Average Residual CO2 Scansum Traces for Vehicle Length Bins
"O
cu
N
"15
E
o
c
c
3
u
(0
Li.
O)
C
-------
Next, we determine the effects of AirSpeed Para on CO2 trace shape by removing the effect of
Vehicle Length, as well as the time decay. First, all transits were divided by the exponential time
decay of Figure 6-33. Second, all transits were divided additionally by the Vehicle Length bin
traces of Figure 6-36. Next, the residual traces were grouped by AirSpeed Para bins, and the
average residual trace and the 95% confidence interval for each AirSpeed Para bin was
calculated. For example, Figure 6-37 shows the result for the 18-21 mph AirSpeed Para bin. The
results indicated that for After-Vehicle Scans > 4 the unnormalized AirSpeed Factor could be
padded at a constant value for the remainder of the scans without straying outside of the 95%
confidence intervals.
The unnormalized AirSpeed Para residual traces for the various bins are superimposed in Figure
6-38. It is apparent the trace shape effects of AirSpeed Para are confined to the first four scans
behind the vehicle, which is the first 0.2 seconds of the vortex. Just as for the Vehicle Length
factors, we divide each AirSpeed Para residual trace by its asymptotic value to produce Figure 6-
39.
Figure 6-37. Average Residual CO2 Scansum Trace for 18-21 mph AirSpeed Para
Bin
Legend -"Mean •—»LCLM95 UCLM95
/proj 1 /EDARinDcnvcr-OCT2019/Analysis_MLout/211122/Anal_MLoutfOCT19_intcrpshapcC02_5.sas 03SEP22 10:23
Figure 6-39 shows that if the AirSpeed Para is 40mph, the Air-Speed Factor is 1 for all times
after the vehicle rear. If the AirSpeed Para is different from 40mph, Figure 6-39 gives the factor
to make the correction to the expected scansum time trace. AirSpeed Paras below 40 mph have
values less than 1 near the vehicle rear, and therefore we expect less pollutant mass near the
vehicle rear.
By calculating the Air-Speed Factors for Vehicle Length subsets of the dataset, we found that the
Air-Speed Factors as shown in Figure 6-39 were relatively independent of Vehicle Length.
6-38
-------
Figure 6-38.
^ 1.5:—
"g 1-4'
^ 1.3
| 1.2
O 1-1
C 1.0
c
3 0.9
i' °-8
u 0.7
lE °-6,
0.5:
0.4:
O.3L
Average Residual CO2 Scansum Traces for AirSpeed Para Bins
6 7 8 9 1011 1213141516171819 20
After-Vehicle Scan
AirSpeed Para (mph)
00-18
33-38
'18-21 •-•-•21-24
38-45 45-52
24-28
52-60
28-33
1 60-99
/proj 1JEDARinDenver-OCT2019/Analysis_MLout/211122/Anal_MLout/OCT19_interpshapeC02_5.sas 03SEP22 10:23
Figure 6-39. Normalized Average Residual CO2 Scansum Traces by AirSpeed
1.4:
1.3i
Time After Vehicle Rear (s)
AirSpeed Para (mph) 00-18 •-•-•18-21 •¦•-•21-24 24-28 •-•-" 28-33
33-38 •—• 38-45 • 45-52 — 52-60 ™ 60-99
/projl/EDARinDenver-OCT2019/Analysis_MLout/211122/Anal_MLout/OCT19_interpshapeC02_5.sas 03SEP22 10:23
6-39
-------
The 207 test vehicle transits of the September 2016 dataset were sufficient to show that the shape
of the vortex time traces is characterized by a peak at Scan 2, which is 0.1 seconds after the
vehicle rear, followed by an exponential decay. Those 207 transits also indicated that time traces
near the vehicle rear had additional dependencies, but the number of transits was insufficient to
determine the functionality of the dependencies.
The 30,559 transits of the October 2019 Westminster study confirmed that the exponential decay
rate in the tail of the vortex (beyond After-Vehicle Scan 8, which is beyond 0.4 s after the
vehicle rear) is independent of vehicle, road speed, vehicle air speed parallel component, release
location, vehicle length, and pollutant.
The shape of the vortex peak, which extends from the vehicle rear to After-Vehicle Scan 8,
which is 0.4 seconds after the vehicle rear, is additionally influenced by vehicle length and air
speed parallel component. The analysis of the exhaust CChin the 30,559 transits quantified the
vortex scansum time trace.
Overall, the shape of the scansum time trace is expressed as the product of the Time-Decay
Factor of Figure 6-33, the Vehicle-Length Factor of Figure 6-36, and the AirSpeed-Para Factor
of Figure 6-39. Parameterizations of these three factors are given by Equations 5-10a, 5-10b, and
5-10c in Section 5.4.
The parameterizations express the expected scansum time trace shape of any vehicle emission in
the vortex as a function of Vehicle Length and the component of the air speed in the direction of
vehicle motion (AirSpeed Para), which is a function of vehicle velocity and wind velocity.
The average scansum trace shape is important to know because in the current method it is used to
smooth and reduce the noise in the EDAR signals obtained from each transit.
6.6 Future Improvement: EvapHC Release Location Detection
The data analysis in Section 6.4 showed that when the EvapHC source location is known, the
Vortex Entrainment Time estimate improves, directly enhancing release rate and emission rate
calculations. Therefore, we did a scoping analysis to investigate the possibility of detecting
emissions zones around the vehicle to determine the approximate location of the EvapHC
sources. Further refinement of zone boundaries, ranking plumes based on physical likelihood,
and outlier removal may help in EvapHC predictions of greater reliability.
Vehicles can emit EvapHC from various locations. The location of the source affects the
emission dispersion and changes the measured EDAR signature. Initially, we looked at the
average scansum traces of test vehicle transits grouped by evaporative release location. Figures
6-40, 6-41, and 6-42 show EDAR scansums v. after-vehicle scan number for aggregated Door,
Hood, and Tank EvapHC releases from test vehicle transits at 22.5 mph. The black lines track
blank CO2 pixels to show where the vehicle is. Blank CO2 pixels occur when the vehicle passes
over the pavement retro-reflective tape. The blue lines track RSD HC signal strength. Since each
scan takes 50 ms, these scansum traces track emissions over time. We see that the signal trace
varies depending on the release location. While the Door, Hood, and Tank release rates were
equal, the Door signal shows the largest vortex peak. In contrast, portions of the Hood release are
blocked from RSD detection during the vehicle transit. These emissions, being farther forward
6-40
-------
on the vehicle, are partially masked because of the coincident vehicle footprint, thereby reducing
a portion of the signal that the RSD instrument would otherwise obtain. Tank releases are a
middle ground between Door and Hood releases; tank emissions have a medium-strength peak
that is partially masked by the vehicle footprint.
Knowledge of post-vehicle peak signal intensity alone cannot distinguish among Door, Hood,
and Tank releases. For instance, a medium-emission rate Door release and a high-emission rate
Tank release may have similar maxima. Additionally, integrating the area tends to
underrepresent the emission rate of Tank releases due to the emissions masking by the vehicle
footprint. As HC levels increase within the vehicle footprint, the vortex peak decreases. From
observing the gradual buildup of emissions during the vehicle footprint in Figure 6-41 (After-
Vehicle Scans -11 to 0), we theorized that with additional spatial analysis of the EDAR array, we
could estimate the release location of a given transit.
We created aggregate heatmaps to detect release location trends. Averaging enhances location
information by mitigating outliers and reducing apparent noise to more clearly image trends than
can be seen in the heatmap of a single transit. Figure 6-43 shows RSD field test results from
releases of artificial EvapHC and artificial ExhHC from the EV-2 all-electric test vehicle. Each
panel, which was made by averaging approximately fifty transits, includes a vehicle footprint,
depicted by the white pixels, moving towards the left. For each panel, substantial perpendicular
air speed was blowing from the bottom of the figure towards the top. The ExhHC emission rate
was metered at 1660 mg/mile and was released from the tailpipe, whose location is denoted with
a black circle. EvapHC was released at 1600 mg/mile from the Hood, Tank, or Door with
positions denoted by the black triangle in the second, third, and fourth panels. No EvapHC was
released in the first panel.
The ExhHC releases are seen as regions of high intensity in the panels, directly behind (to the
right of) the black circle, which weaken with increasing distance from the point of highest
intensity. The clouds of ExhHC appear with similar spread and false color in all four panels.
By examining the differences among the panels, we begin to see the regions that evaporative
emissions tend to populate. The second panel with its under-Hood EvapHC release shows a high-
mass HC plume on the right side of the vehicle (near the top of the panel) as well as light
EvapHC Mass in Vortex area behind the vehicle (right of the footprint). In the third panel, the
Tank EvapHC release, originating from below the vehicle, shows an even dispersion of HC
behind the vehicle. In the fourth panel, the fuel Door EvapHC release shows a high concentration
(red pixels) of HC mass at the release location with a bright yellow plume being pulled into the
vortex.
The different patterns of EvapHC releases from different release locations suggests that a pattern
recognition algorithm could be used to estimate the approximate location of EvapHC sources.
Estimated EvapHC emissions location coupled with the measured AirSpeed Perp would be used
to improve the selection of an appropriate relative VET factor for the effect of release location
(see Table 6-4) for the transit.
6-41
-------
Figure 6-40. EDAR Scansums v. Scan Number for DOOR Evaporative Releases
-10 -5 0 5 10 15 20 25
After-Vehicle Scan
Signal Adj HC (float zero)
/proj 1 /EDARin Denver-OCT2019/Analysis_M Lout/yymmdd/Anal_MLout/OCT 19_shape_edit.sas 17NOV21 16:03
0.052
0.046
0.040
Figure 6-41. EDAR Scansums v. Scan Number for HOOD Evaporative Releases
/proj 1/EDARin Den ver-OCT2019/Arialysis_MLoutJyymmdd/Anal_MLout/OCT19_shape_edit.sas 17NOV21 16:03
cTi
a.
<
Q
LU
(U
U\
(0
1-
u
0.052
0.046
0.040
15 -10 -5 0 5 10 15
After-Vehicle Scan
Signal Adj HC (float zero)
6-42
-------
Figure 6-42. EDAR Scansums v. Scan Number for TANK Evaporative Releases
0.052
CM
jE 0.046
jy
o
E 0.040
E
3
tn
z
ro
u
if)
"5
c
a
"t7>
a.
<
~
LU
a)
ai
ro
L.
a>
<
-10 -5 0 5 10 15 20 25
After-Vehicle Scan
Signal Adj HC (float zero)
/projl/EDARin Derive r-OCT2019/Analysis_M Lou t/yymmdd/Anal_MLout/OCT19_shape_edit.sas 17NOV21 16:03
Figure 6-43. EV-2 HC Releases from Different Locations in a Transverse Air Flow
Evap Release Location
Under
HOOD
TANK
Posn
Fuel
DOOR
I
S
q5
x
CL
c
CO
co
ro
2
O
X
,i
H—'
ro
0)
a:
^ EvapHC: 1600 mg/mile
O ExhHC: 1660 mg/mile
6-43
-------
The dispersion patterns of EvapHC for different release locations were enhanced in the panels of
Figure 6-43 by averaging the arrays of approximately 50 high-EvapHC individual transits per
panel. For standard RSD field operation, EvapHC release location would need to be estimated
from just a single transit when the EvapHC release rate may or may not be high. Under these
circumstances, noise in the RSD signal becomes an issue because noise in individual pixels
obscures the EvapHC signal.
To reduce the effects of noise, we began the development of zones around and behind the vehicle
to locate the origin of various emissions. Zones help classify emissions surrounding a vehicle. By
comparing the average signals of different zones, we protect against outliers and can lower the
detection limit. Figure 6-44 shows a conceptualization of zones that could be used to help
determine emission release location. If, for instance, the zone-average EvapHC signals were
greater than the average noise level in Zones 4 and 6, and if the airspeed data indicated
substantial air flow from vehicle left to vehicle right, one possible explanation would be an
under-hood EvapHC release. The morphology of these emissions would match that of the second
panel of Figure 6-43, where an under-hood evaporative release dispersed to one side of the
vehicle and also to the vortex. However, if similarly sized EvapHC signals were found in Zones
5 and 6, but not in Zone 4, then the finding could indicate an EvapHC release location closer to
the rear of the vehicle as for the Tank location in the third panel of Figure 6-43. Thus, it is
possible that classifying EvapHC release location by assessing the relative strength of RSD
signals found in distinct zones can help explain the emissions and its associated factors.
During our preliminary analysis27, we identified rectangular zones around and behind the
vehicle. From the test vehicle data, we correctly identified the evaporative release location at a
rate greater than 50%. A single transit with a high-release rate at the Hood of a test vehicle is
shown in Figure 6-45. In this figure, the vehicle footprint is moving down towards the bottom of
the page, with a back bumper position at about 30 on the y-axis. Here, we see a significant plume
emanating from the side of the vehicle. Test vehicle data, however, represent ideal conditions
with known releases of large amounts of EvapHC. Due to the typically low emission rates of
fleet vehicles, we did not apply our model to the fleet. Novel test transits, in addition to a more
sophisticated or machine-learning model, could enable fleet-level EvapHC emissions detection.
Test transits tailored to enhance release location detection could be part of a future evaporative
emission study. We would like to improve release location detection because of the enhancement
it brings to Vortex Entrainment Time, which directly affects release rate and emission rate
calculations. During the 2019 Westminster study, the test transits focused around two all-electric
sedans. We now know that Vortex Entrainment Time is mildly correlated with both vehicle
shape and emission release location. Due to the high variability of a single transit, multiple Door,
Tank, and Hood test releases on larger SUVs, light duty trucks, and medium/heavy duty trucks
would strengthen the release location prediction model. Furthermore, the latest noise reduction
and separation techniques have enhanced the overall signal-to-noise ratio. Such improvements
promote focusing on medium-to-low HC release rates, which have greater applicability to fleet-
level evaporative emissions.
27 P:/EDARinDenver-OCT2019/Analysis_MLout/211118/Anal_MLout/
OCT 19_hoodEvapProfile_standard. sas
6-44
-------
Figure 6-44. Example Zones for Detecting Releases from Vehicle Locations
2
/ 4
12
10
14
O)
u
6
8\
CLI
>
5
7
K 9
13
X
11
Figure 6-45. EvapHC Hood Release for a Single Transit in a Transverse Air Flow
7 20191020 000505 car 001813
HC C02 CO NO N02
10
Bfli
i l l ill! Ill i
I 11!
tj J I I!'!
10
10
10
10
20
I.lllll u ii'Ll'J1"
Milflllii '' ' lIUlMt1
¦Ipl?'
20
20
20
20
i Mill iiii iii
ImlcCTSvffl'ffln
30
30
1
30
&
30
li
30
0 100 200 0 100 200 0 100 200 0 100 200 0 100 200
EvapPiume ExhPlume Noisel Noise2 Noise3
10 Hililil
10
10
i lilhl
i' i j ;i]!|
10
1 1
II il'l ¦'fi v'1 I'll1 fr'
10
20 1111
20
20
'nil
20
20
o
CO
30
30
I
30
i '¦ iji'ji 'ifi' 111 ''i
''n 1;
30
0 100 200 0 100 200 0 100 200 0 100 200 0 100 200
6-45
-------
7.0 Demonstration of Signal Analysis of RSD Detailed Data
7.1 Adjustment and Improvement of RSD Detailed Data
Figure 7-1 displays raw EDAR detailed data for an example vehicle transit from the Westminster
dataset. The direction of vehicle motion for each panel is vertically toward the bottom. The
colors shown represent relative amounts of each pollutant for the five data channels: HC, CO2,
CO, NO, and NO2. No corrections have been applied to this example dataset.
Figure 7-1. Detailed Data Patterns for Example Westminster Transit
7 20191023 000512 car 001188
HC C02 CO NO N02
The Pre-Processing Device performs the following four signal improvements to the incoming
EDAR measurements presented to it, in the order shown:
1. Adjust constant-level offsets
2. Remove outliers
3. Filter non-physical components
a. De-stripe in the direction of vehicle travel using multi-tonal cancellation (HC only)
b. Adaptive notch filter in the direction of the across-road scans
4. Interpolate measured transit data to a rectangular grid
Each channel of incoming EDAR measurements is treated independently and identically in terms
of its processing, with two exceptions:
• When filtering non-physical components, the HC channel is processed differently from
the other measurement channels. The HC channel data is processed using multi-tonal
cancellation followed by adaptive notch filtering, whereas the other channels (CO2, CO,
NO, NO2) are processed using adaptive notch filtering only.
• When applying adaptive notch filtering to all data channels, the NO2 channel is used to
adaptively estimate the frequency of the notch disturbance to be removed.
7-1
-------
Table 7-1 shows which processing step is performed on each of the data channels.
Table 7-1. Application of Improvement Steps to Transit Data
Improvement Step \ Data Channel
HC
co2
CO
NO
no2
1. Offset Adjustment
X
X
X
X
X
2. Outlier Removal
X
X
X
X
X
3. De-striping via Multi-Tonal Cancellation
X
4. Adaptive Notch Filtering
X
X
X
X
X0
5. ZigZag Interpolation to Rectangular Grid
X
X
X
X
X
X - Applied to each data channel independently
O - Used in procedure across all applied channels
Each of the five signal adjustments is now considered.
Adjusting Constant-Level Offsets
Processing Rationale - As provided, the EDAR detailed data measurements have been
calibrated by HEAT such that the values nominally represent mass measurements (mole/m2).
Since mass is always non-negative, these calibrations would appear to be straightforward to do.
On examination of the data, however, it became clear that measurement noise causes some of the
values in the EDAR data to be negative, which is non-physical. This means that the zero
reference of each channel is not obvious. Moreover, the zero reference may not be set exactly by
the EDAR instrument. An inaccurate setting of the zero reference would create a bias of the mass
measurements28 in any data channel, thus leading to a bias in any calculations using these mass
measurements. Fortunately, the statistical behavior of this noise is a feature that can be used to
set a zero reference for each measurement channel in each transit independently.
The procedure for finding the zero reference is based on the following concept. If the noise in
each channel is additive and Gaussian with a bell-shaped histogram, then the peak of this noise
will define the zero reference. The peak can be found precisely by the following steps:
1. Compute the histogram of each measurement channel in each transit.
2. Take the logarithm of the number of counts in each histogram bin. The logarithm turns
the histogram count values around the zero reference into a parabola facing downward.29
3. Fit a parabola to the log-histogram points for values below a data-dependent threshold.
Currently, this threshold is set as 20% of the maximum histogram pixel count for each
bin count. This threshold removes bin values that contain both outliers and likely plume
pixels.
28 Note that zero-offset inaccuracies have little effect on EDAR's routinely reported exhaust concentration
values.
29 The expected parabolic shape is a consequence of the functional form for the bell-shaped curve.
7-2
-------
4. Find the peak of this parabola. This is the zero reference for the measurement channel for
the given transit.
This method does not make use of the locations of the signal-free measurements in the transit.
The values used for this offset estimation process include all pixels in each channel except the
following:
1. Pixels that have been determined to be clipped values. As delivered, the Westminster data
has been adjusted by HEAT with certain pixels clipped to an arbitrary negative value. It
is presumed that these pixels have been deemed to be "too negative to be valid." The
negative value that replaces these pixels is non-physical and alters the histogram counts
in an erroneous manner. These pixels are easily identified, as they are set to the same
minimum value. These pixels are detected and removed from the histogram formation.
2. Pixels that have been artificially adjusted to near-zero. As delivered, the Westminster
data has a number of pixel values that are unnaturally close to zero, yet these pixels are
not vehicle pixels. The non-zero value of these pixels registers as a peak in the histogram,
and this peak is ignored in the processing.
3. Pixels counts that fall below a percentage threshold. The noise histogram need only have
pixel counts that extend across the two sides of the parabola after the log has been
applied. Thus, we limit the range of these pixel counts to those bins whose counts are at
least 20% of the maximum count of any one bin. The 20% rule used in this detection
appears to provide good performance for the image sizes and pixel counts contained in
the Westminster dataset.
This method assumes that there are many samples (pixels) in each measurement channel that
have little to no signal in them. When a plume is present, it typically has a limited spatial extent,
such that the number of low-level values in the plume is small relative to the number of
measurements in the entire transit. Therefore, this assumption is often reasonable. The method
might have difficulty in situations where a) the measured transit is short, with not many scan
lines after the vehicle, b) the plume has a large spatial extent in terms of the number of nonzero
values relative to the number of sample measurements in the transit, and/or c) the background of
the transit channel is not flat, e.g. there is an unnatural shape or curvature to the overall channel
image. In the first two of these cases, the method is data-starved for the particular channel. In the
last case, the image data does not fit the histogram model.
Figure 7-2 shows an annotated plot for all pixels in the CO2 raw detailed data in Figure 7-1. The
CO2 pixel optical masses were binned with a bin width of 0.00145 mole/m2. Figure 7-2 shows
the log of the pixel count in each bin vs. the midpoint value of each bin. Bins from -0.03 to about
0.01 contain counts of CO2 optical mass values that are dominated by noise and create the
concave-downward parabolic shape. A quadratic fit to these points provides an estimate of the
zero reference and the standard deviation of the noise. These values correspond to most of the
pink area of the second panel of Figure 7-1. Values larger than about 0.01 mole/m2 are smaller in
number than for noise, but they begin to deviate from the parabola because they are influenced
by signal. The figure shows that there are only three pixels that have CO2 optical masses greater
than 0.40 mole/m2. These three important pixels are for the CO2 mass just behind the tailpipe exit
and are the red pixels in the second panel of Figure 7-1.
7-3
-------
Figure 7-2. Histogram of CO2 Pixel Counts for Series=512 Transit=1188
100
c
3
o
U 10
"a>
x
Q_
Noise!
- ± 3 Standard Deviations
7 Small
I Signals
Medium
Signals
Large
Signals
1
-0.05 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45 0.50 0.55
C02 Optical Mass (mole/m2)
/proj1/EDARinDenver-OCT2019/Analysis_MLout/220817/Anal_MLout/OCT19_C02_bin_plot.sas 27FEB23 12:46
Processing Examples - We now provide examples of the offset estimation method as applied to
example transit data from the test vehicle datasets. In each figure, the five raw images from the
measured transit are shown in the first row, and the log-histograms and quadratic data fit plots
are shown in the second row. The log-histograms shown are only for the portion of the data that
is dominated by noise, for example, in Figure 7-2 for CO2 optical masses less than 0.01 mole/m2.
Figure C-l (in Appendix C) provides an example of a test vehicle with a low amount of HC
evaporative emissions, a simulated CO2 exhaust emission, no other tailpipe emissions, and
moving at a low speed. In this case, the log-histograms form inverted parabolas in each data
channel, and the data is easily fitted to a parabolic shape.
Figure C-2 provides an example of a test vehicle with a high amount of HC evaporative
emissions, a simulated CO2 exhaust emission, no other tailpipe emissions, and moving at a low
speed. Again, the log-histograms form inverted parabolas in each data channel, and the data is
easily fitted to a parabolic shape.
Figure C-3 provides an example of a test vehicle with a low amount of HC evaporative
emissions, simulated HC/CO2/CO/NO exhaust emissions, and moving at a low speed. Again, the
log-histograms form inverted parabolas in each data channel, and the data is easily fitted to a
parabolic shape.
Figure C-4 provides a second example of a test vehicle with a low amount of HC evaporative
emissions, simulated HC/CO2/CO/NO exhaust emissions, and moving at a low speed. In this
7-4
-------
case, the HC and CO log-histograms show non-parabolic shapes, and the amplitudes of the CO
channel in particular deviate significantly from the model. This behavior was observed rarely,
but it tended to occur with transits that exhibited some amount of CO emissions.
Figure C-5 provides an example of a test vehicle with a high amount of HC evaporative
emissions, simulated HC/CO2/CO/NO exhaust emissions, and moving at a low speed. In this
case, the CO log-histogram exhibits a highly non-parabolic shape. It was more likely to find an
erroneous CO offset model in high HC emission environments for this test vehicle, suggesting
that there is an interaction between HC and CO measurements when HC emission values are
high and CO emissions are non-zero.
Other examples of the failure of the offset adjustment procedure to adequately characterize the
CO offset value can be found. Figure C-6 shows an example from the fleet where the shape of
the log-histogram CO values is concave upward. This situation implies that the noise variance is
negative - an impossibility - and means the additive noise model is not appropriate for this
channel of this dataset.
Processing Summary - As conceived and developed, offset adjustment provides a simple and
accurate method for determining the constant offsets in each channel of an EDAR measured
transit when the data in each channel obeys an additive noise model. The method assumes a
constant background level for each measured channel, an assumption that is physically justified.
In most scenarios, it works well. It also provides an estimate of the background noise level in the
form of a noise signal power for each channel on a per-transit basis, determined from the
curvature of the inverse-parabolic model of the log-histogram data points.
The primary drawback of the offset adjustment procedure is the parsimony of the additive noise
model for certain measurement scenarios. For example, when the CO channel measurements are
significantly positive, indicating the presence of a CO signal, the offset adjustment procedure can
fail. Such situations can likely be detected and flagged, although such detection methods are
currently not implemented. A qualitative review of these particular scenarios indicates that they
tend to occur in situations where the underlying shape of the CO channel background is not very
flat and has some form of curvature to it in terms of Scan Position. It is unclear why the CO
channel would exhibit this artifact. It may be useful to study this phenomenon further and either
correct for it or provide feedback to the RSD manufacturer to help address it in future data
collection campaigns. For the Westminster data, if fleet measurements are considered plentiful
enough such that some fraction of them could be rejected, a simple detector based on the
following reasoning could be built:
a. Find all transits with significant CO components.
b. Determine the similarity of the CO channel to the CO2 channel in a normalized way (i.e.
based on relative shape similarity and not dependent on mass or plume size). If the two
channels are dissimilar, reject the transit.
For future work, some possible enhancements to the procedure can be developed. A partial list
follows.
7-5
-------
1. Use a goodness-of-fit measure between the log-histogram values and the inverse-
parabolic model to identify transit channels and/or transits that do not fit the additive
noise model. The goodness-of-fit could take into account various quantities, such as the
error in the inverse-parabolic fit or the number of pixels being used in the overall model.
2. Test the value of the variance estimated in each channel - for example, a negative
estimated variance value - to determine data channels for each transit that do not fit the
additive noise model. Flag these channels and/or transits for further analysis and/or
alternative processing.
3. Explore the dependence between the CO channel and other data channels in terms of
degree-of-fit to the additive noise model. An example question thread that could be
explored: Does the goodness-of-fit of the CO channel to the additive noise model
correlate with the goodness-of-fit of other data channels? If so, how often does this
occur? Can this correlation be used as a flag to identify problematic transits for omission
or further processing?
Removing Outliers
Processing Rationale - The EDAR instrument relies on a reflective strip on the roadway to
reflect laser light back to the detector. Thus, the measurements are only valid when this reflective
strip is not covered by an object, such as a vehicle's body. As provided by HEAT, the EDAR
measurements that have been determined by the EDAR instrument to be occluded are set to zero
values. These determinations are not error-free, however. Problems can occur near the corners of
the vehicle where large outlier values or "spikes" can sometimes be observed. If these outlier
values were left in the measurement, they would greatly distort the results of later processing
stages, particularly the separation processing.
For this reason, the following procedure is used to find these outlier values and set them to zero:
1. From the histogram generated from the constant-level adjustment procedure, determine
the width of the noise histogram, also known as the standard deviation of the noise.
2. Test all non-zero measurement values that are next to the vehicle footprint. If these
measurement values are larger than a threshold - currently set as twice the noise standard
deviation - these are possibly "spike" values.
3. Make sure that these large values are not next to other large values and are also behind
the vehicle footprint, because those are likely legitimate measurements. Those
measurements are kept. The remaining detected values are outliers and are removed.
Processing Examples - We now provide examples of the outlier removal method as applied to
example transit data from the test vehicle datasets. In each figure, the five raw images from the
measured transit are shown in the first row, and corresponding images of the transit data with the
outlier pixels removed are shown in the second row. Due to auto-scaling of the colormaps, the
images in the second row have a different color palette due to the different amplitude ranges of
each of the data channels. Thus, a "tell-tale sign" of an outlier pixel being removed is an overall
7-6
-------
color shift of a second-row image to a lighter shade of blue or yellow, as one or more large-scale
pixels have been removed.
Figure C-7 (in Appendix C) provides an example of the EV-2 test vehicle with a low amount of
HC evaporative emissions, simulated HC/CO2/CO/NO exhaust emissions, and moving at a low
speed. As can be seen, the main difference between the first- and second-row images is the HC
data channel image. All other images have a similar background shade. Moreover, the detected
outliers are found to be typically at the sides, corners, and the front of the vehicle. The HC
emissions are clearly more evident in the processed images.
Figure C-8 provides an example of the EV-2 test vehicle with a high amount of HC evaporative
emissions, simulated HC/CO2/CO/NO exhaust emissions, and moving at a low speed. Again, the
main difference between the first- and second-row images is the HC data channel image. All
other images have a similar background shade. As in the previous case, the detected outliers are
found to be typically at the sides, corners, and the front of the vehicle. The HC emissions are
again clearly more evident in the processed images.
Processing Summary - The purpose of outlier detection for the EDAR measurement instrument
is to identify and remove erroneous pixels that occur near the boundary of the vehicle footprint.
To provide some context, the EDAR instrument data was originally delivered with this vehicle
footprint obscured by a larger rectangle that also removed pixels that were near the back of the
vehicle bumper. This was viewed as a loss of data, so the EDAR instrument data was re-
delivered with the rear bumper pixels left intact. Outliers were then discovered in the original
data measurements that were clearly "wrong," as they were large, isolated pixels next to the
vehicle. The outlier detection method identifies these pixels in a per-transit technique that is
data-dependent within each channel. The use of a per-channel variance calculation for setting the
detection threshold allows this test to be statistically robust.
The only potential drawback of this technique is its connection to the offset estimation step
described previously. The offset estimation method can fail, and if it does, then the variance
calculation it produces and used in the outlier detection method to set the detection threshold can
be incorrect. This is expected to be a rare occurrence. Thus, this technique is likely to be useful
as-is for future data processing campaigns.
De-Stripinq via Multi-Tonal Cancellation
A preliminary analysis of the Westminster dataset indicated that every transit measurement
contains non-physical additive noise components. These components show up as periodic sine
wave artifacts in each channel of each transit and are structured. It is unclear why these artifacts
are present, but they can be identified, characterized, and largely mitigated through numerical
processing.
Two types of non-physical artifacts were identified. These two artifacts are dealt with
independently in the data processing. The first of the processing methods to be applied to the
Westminster data is termed multi-tonal cancellation. The second of the processing methods is
7-7
-------
termed adaptive notch filtering. This subsection discusses multi-tonal cancellation; the next
subsection discusses adaptive notch filtering.
Processing Rationale - The first type of artifact is something that mainly affects the HC channel
measurement in each EDAR dataset in the Westminster dataset. The artifact shows up as a series
of lines and periodic waves that extend in time from the front to the rear of the transit data in the
direction of vehicle motion. These lines and periodic waves are random in amplitude and phase
for each position across the vehicle width. The resulting image looks like it has both "stripes"
and sine waves in it. The stripes are oriented vertically and look like streaks of light and dark
pixels. The sine waves are undulating vertically and appear to produce light and dark regions on
the order of every four to five scan lines. To preview such an image, the upper-left corner of
Figure C-9 (in Appendix C) shows an example HC image from a particular EDAR dataset in the
Westminster dataset. Other examples of these types of images are shown in the upper-left corner
of Figure C-10 through Figure C-20.
Because this type of data artifact appears in other EDAR measurement sets from other
instruments in other measurement campaigns, several different methods for its removal have
been developed.
a. Adaptive Linear Prediction: Apply a common single-channel adaptive least-squares
linear predictor (4 samples of prediction, 11 taps for filtering) from back-to-front of each
transit. This operation is termed adaptive line enhancement in the adaptive filtering
literature.
b. Multi-Transit Prediction: Use noise-only data fields from several transits before and
several transits after to predict the current transit's noise field. This operation is similar to
how a video encoder performs when compressing a video signal.
c. Multi-Tonal Cancellation: Employ multi-tonal cancellation of each data column with
fixed noise frequencies. This operation is a form of adaptive noise cancellation common
in digital signal processing when a) the noise disturbances are known to be sinusoidal in
nature and b) the frequencies of these tonal disturbances are known.
Figure 7-3 shows examples of both linear prediction (middle) and multi-transit prediction (right)
as applied to a particular HC dataset example shown on the left. This figure does not show an
example of multi-tonal cancellation. This figure is meant to illustrate the types of improvements
that can be obtained using the first two methods listed above. In the final analysis for this report,
it was determined that the third method not shown in Figure 7-3 - multi-tonal cancellation - was
most suitable for the Westminster dataset. Thus, it is the only one discussed in depth in this
report.
7-8
-------
Figure 7-3. Example of Linear and Multi-Transit Prediction for HC Channel
Original HC Data Linear Prediction Multi-Transit Prediction
An analysis of the entire Westminster dataset showed that the frequencies of these disturbances
in the HC channel were essentially the same for every transit, although the amplitudes and
phases of these disturbances change on a per-position basis uniquely across each transit. Figure
7-4 shows the power spectra of the five data channels for all 33,636 transits in the direction of
vehicle motion, where we only considered the data away from the vehicle along the edges of
each transit - usually the first few and last few pixels of each side of each transit. The purpose of
this analysis is to attempt to ignore any emission signal that may be present in the transit data
without having to build a detector for this condition. Thus, the resulting scan line power
spectrum is simply the frequency content of the noise field of the vertical lines of all signal-free
Westminster data in the aggregate. The HC data has a scan line power spectrum with peaks at the
following digital frequencies, where f = 1 corresponds to half the sampling rate: f = 0, f =
0.21875 (= 7/32), f = 0.5, and f = 1. We assume that these periodic disturbances are additive. So,
to remove them, we process the HC channel according to the method described in Section 5.2.
We call this processing multi-tonal cancellation.
7-9
-------
Figure 7-4. Noise Power Spectra for RSD Channels in the Vehicle Direction
10"'
3 10"2
£
4—1
o
CD
Q.
W
w
CC 10-3
10"4
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2
Normalized Spatial Frequency (half cycles/ScanSum line)
Figure 7-5 shows the power spectra before and after processing using the multi-tonal
cancellation method and the best competing method to this chosen one at the time of writing -
multi-transit prediction - for all of the transits in the Westminster dataset, where we again only
considered the data away from the vehicle along the edges of each transit. In this example, we
applied a fourth tonal cancellation frequency at f = 0.779 (= 399/512) to address the small tonal
component at this frequency but found that there was no improvement at that frequency; hence,
this fourth tonal cancellation was removed for final processing. As can be seen, multi-tonal
cancellation does a better job of reducing the tonal peaks of the HC noise field in the direction of
vehicle motion than does multi-transit prediction, except at f = 0, which corresponds to the offset
baseline. The improvement at f = 0.5 is about 12.5 dB, and the improvement at f = 1.0 is about
10.5 dB.
7-10
-------
Figure 7-5. De-Striping Evaluation using HC Noise Power Spectra
¦ Measured HC
¦after Multi-Transit Prediction (3+3 tr.)
¦ after Tonal Cancellation (4 freqs+DC)
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1
Normalized Spatial Frequency (half cycles/ScanSum line)
Processing Examples - To illustrate the effects of multi-tonal cancellation, we consider only the
HC data channel from selected transits of the Westminster dataset corresponding to the following
known measurement conditions:
Test Vehicles
EV-1
EV-2
EvapHC Release Locations
DOOR
TANK
HOOD
EvapHC Emission Rates
200 mg/'mile (low)
6400 mg/mile (high)
Nominal Road Speed
22.5 mph
Considering one example from each possible combination of the above conditions results in
twelve transits from the Westminster dataset.
For each selected transit, we present one figure per page. An example of the first figure page is
shown in Figure C-9 in Appendix C. The upper-left corner of this figure shows the original
measured HC data for the selected transit after outliers have been removed. The right side of this
figure shows plots of the scanlines corresponding to pixel positions 100 (top right of the figure)
and 200 (bottom right of the figure), respectively, which are easily located vertical lines
correspondi ng to pixel values shown in the images on the left of the figure. Two lines on each of
the two right-side plots are provided: the original scan line in red, and the processed scan line in
7-11
-------
blue in which multi-tonal cancellation has been applied. The two lines on each plot show an
example of the type of interference that multi-tonal cancellation reduces in its operation.
On each page, the lower-left image shows the processed HC data after multi-tonal cancellation.
Comparison of the upper-left and lower-left images illustrates the visual improvement that multi-
tonal cancellation provides in reducing interference for the HC data channel.
Figure C-9 through Figure C-20 provide the twelve pairs of the figure pages for the different EV-
1 and EV-2 vehicle transits discussed previously. Although these twelve figure sets will not be
discussed individually in this report, several comments are now provided that summarize the
types of improvements that the processing method achieves.
Improvement #1: When plumes are present, multi-tonal cancellation enables them to be
more easily seen visually in the data. This improvement is evident for strong plume
signals for the portions of the plumes that are ten or more scanlines after the vehicle (e.g.
Figures C-9, C-l 1, C-15, C-17). This improvement is also evident for weak plume signals
near the vehicle (e.g. Figures C-13, C-19, C-20).
Improvement #2: Multi-tonal cancellation removes oscillations and offsets that are clearly
erroneous, while preserving waveforms that do not exhibit such effects. Good examples
of this performance can be seen in the plots in Figure C-19 and Figure C-9, in which the
plots at Line 100 each show a fairly clean original signal and the plots at Line 200 show
an extremely noisy signal that is improved significantly through multi-tonal cancellation.
Improvement #3: Multi-tonal cancellation addresses constant-level offsets that vary with
pixel position, effectively performing image de-striping. These improvements are found
in all Figures C-9 through C-20. They are exhibited in the signal plots as well. See
Figures C-10, C-12, C-14, and C-19 for plots that show negative scan line values that are
clearly incorrect.
Processing Summary - Multi-tonal cancellation is a processing method that addresses two
specific problems in the EDAR measurements. One of these problems - striping - is common to
all multispectral "pushbroom" sensors. The other problem - tonal components in the direction of
motion - is clearly present in the Westminster HC data, but the source of these artifacts is
unclear. Both artifacts appear to be additive, such that estimation and subtraction is a viable
methodology to address them.
While the multi-tonal cancellation method works well, there are some open issues regarding its
performance that could be addressed in future work:
1. Improve the performance of the method for low frequency striping effects. From Figure
7-5, it appears that the performance of multi-transit prediction is better at lower
frequencies than multi-tonal cancellation. It is possible to combine the two techniques to
obtain better performance than using either one alone. Such a combination would be
useful to consider.
2. Develop quantitative emission-based strategies for evaluating the performance of these
methods. The quantitative evaluation provided in Figure 7-4 and Figure 7-5 is well-
founded but only includes the noise field in the Westminster data. Determining the
7-12
-------
quality of the noise suppression should consider the level of signal present in the data as
well.
Adaptive Notch Filtering
Processing Rationale - The second type of artifact is a tonal disturbance in the form of a low-
level sine wave that appears in the one-dimensional signal that makes up the zigzag measurement
of an entire channel of a transit. Figure 7-6 shows the spatial structure of this scan. To make the
one-dimensional signal, we put the scan lines for each channel in an end-to-end fashion,
removing the common value at the ends of each scan line
Figure 7-6. Spatial Structure of EDAR Pixels for ZigZag Collection Pattern
Figure 7-7 shows an example NO2 dataset as an image (lower left) and its corresponding one-
dimensional signal as a time plot (upper plot). The flat regions in the upper plot are the portions
of the EDAR scan that are the vehicle footprint. The red boxes in the NO2 image and one-
dimensional time plot represent two scan lines - a "zig" and a "zag" - of the EDAR
measurement process across the roadway. These 512 samples are plotted in the smaller plot on
the lower right, also outlined in red. The zoomed NO2 signal shows an obvious tonal artifact of
varying amplitude across this scanning process. This tonal artifact shows up as a moire-type
noise pattern in the NO2 image in the lower left and is similar to the type of interference an old
cathode ray tube (CRT) television set would show when it had oscillatory interferences in its
display electronics.
7-13
-------
Figure 7-7. Example of EDAR NO2 Signal Collected for One Transit
10 r—^
6000 6200
Sample Number
6400
This sine wave can be seen in the NO2 channel of each transit, as this channel is usually signal-
free except for this sine wave. The frequency of this sine wave is different for each transit, and it
appears to slowly change over the entire set of 33,636 measurements. This issue was determined
through an analysis of a slope error in the test vehicle data that occurred when the notch filter
frequency falls to a low value. More recently, it has been determined that the frequency of this
interfering sine wave is a function of temperature. Figure 7-8 shows an analysis that illustrates
this fact. The upper half of this figure shows the field value "EDARAmbientTemperature"
contained in the metadata of the Westminster measurement dataset for the first transit of each
hour of the Westminster data collection campaign - from 12:00am, Sunday, October 20, 2019Z
to 2:00pm, Thursday, October 25, 2019 - whenever such data is available. The lower half of this
figure shows the notch filter frequency determined via frequency analysis on each of these
transits in blue, along with a best linear fit of the "EDA RAmbi entTem perature" value to this
notch filter frequency determined by optimal least-squares methods to be:
f = 1.0342 - 0.01592 * EDAR Ambient Temperature.
Thus, the frequency of the noise disturbance in the one-dimensional scan signals of the EDAR
instalment is approximately negatively linear with temperature. This correspondence likely
means that some physical aspect of the EDAR instrument is sensitive to temperature. Identifying
the cause of this disturbance could lead to mitigation strategies in future data collection
campaigns.
7-14
-------
Figure 7-8. Change in Tonal Disturbance Frequency for Test Vehicle Transits
70
60
a>
I 50
L—
(1)
CL
E 40
0
T3
0
N
"ro
E
30
0.6
• EDAR Ambient Temperature
•*s.
.*
•• 'A
•
s •
* • *
. - v \
1
*•
v;
•
• •
* * . V*
. •*
• N •
•
%
1
*v\*
1 1 1
20
40
60
80
100
120
140
E 0.4
o
c
o
c
0
3
CT
0
0.2
Adaptive Frequency Value
EDAR Ambient Temperature - Linear Fit
A
<
j
A*
/ T
i 1
K;/
iv;
'v
\ r
** ;
v.
\ i
v
£*'
%m
i r
20 40 60 80
Experimental Hour
100
120
140
To reduce this sine wave component, we apply the processing described in Section 5.2. Because
the processing uses a notch filter with an adaptive frequency, we call this processing adaptive
notch filtering. This form of adaptive notch filtering has one effective adaptive parameter: the
notch frequency. The notch frequency for each transit is estimated from the NO2 channel using
standard power spectrum methods. For each transit, the same filter is applied independently to all
data channels.
For the HC channel, both the multi-tonal cancellation and adaptive notch filtering help to
improve data integrity. The striping effects in the HC channel are generally much larger than the
effects due to the presence of the tone in the zigzag form of the HC measurement data. Because
of this, the multi-tonal cancellation is applied first to the HC channel, followed by adaptive notch
filtering. For the other four EDAR channels - CO2, CO, NO, and NO2 - only adaptive notch
filtering is applied.
7-15
-------
Processing Examples - We illustrate the effects of adaptive notch filtering through selected
transits of the Westminster dataset corresponding to the following known measurement
conditions:
Test Vehicles
EV-1
EV-2
EvapHC Release Locations
DOOR
TANK
HOOD
EvapHC Emission Rates
200 mg/mile (low)
6400 mg/mile (high)
Nominal Road Speed
22.5 mph
Considering one example from each possible combination of the above conditions results in
twelve transits from the Westminster dataset.
Figure C-21 (in Appendix C) through Figure C-32 show images from these twelve transits in the
same data ordering as Figure C-9 through Figure C-20. The top row of each figure slide shows
the five channels of transit data, where the HC channel has been processed using multi-tonal
cancellation. The bottom row of each figure slide shows the five channels of transit data after
adaptive notch filtering. These data can be explored and compared individually from top row to
bottom row to see the qualitative effects of the processing. In some cases, the improvements are
subtle, whereas in others, the improvements are more obvious.
Highlighted improvements include the following:
Figure C-22 uses the dataset used to generate the images in plots in Figure 7-7. In this figure,
images from all five EDAR channels are shown, both before (top) and after (bottom) adaptive
notch filtering. The improvement in the NO2 channel is obvious, although this result is not very
interesting given the design of the processing method. Examining the HC, CO, and NO channels,
however, there is clearly an improved visual structure in the associated plume images after
processing, as the noise field in these images is of lower amplitude. This qualitative
improvement likely results in a quantitative improvement in mass assessment as this data is
processed using subsequent methods.
Figure C-26 shows an improvement in plume structure from before to after adaptive notch
filtering processing as well. The periodic components that appear in the original data within the
CO2 and NO channels is reduced after processing.
Figure C-29 illustrates an improvement in the plume structure of the HC channel through
adaptive notch filtering. The bottom HC image looks clearer and has a better defined plume
structure as compared to the top HC image. Note that these effects are most easily seen in
situations where the data channel contains a strong signal of interest.
Note that if the tonal disturbance is not strong, then the improvement provided by adaptive notch
filtering is not as apparent. Examples where this is likely the case include Figure C-25 and
Figure C-30. Even when the processing provides minimal qualitative effects, there are
7-16
-------
improvements that can be gleaned. For example, Figure C-31 shows an HC channel that has a
better-defined plume structure after processing (bottom) as compared to before (top). It is clear,
however, that there are other artifacts in the data, particularly in the NO and NO2 channels for
this transit, that are not being addressed by adaptive notch filtering processing. Fortunately, these
other noise artifacts are spatially-correlated across multiple data channels, which implies that
they can be addressed using spatially-oriented methods such as blind source separation.
Processing Summary - Adaptive notch filtering is a modification of a fixed notch filtering
method originally developed for the September 2016 EDAR dataset collected at TTI. The
primary modification to this method was to make the frequency of the tonal reduction adaptive,
as the Westminster dataset exhibited a changing tonal frequency value.
Adaptive notch filtering is based on the fundamental way the EDAR instrument functions. The
EDAR instrument collects a one-dimensional signal over time. This one-dimensional signal is
then mapped to a two-dimensional array through the position of the scanning laser on the
roadway and the travel time of the vehicle as it passes under the instrument. Any noise or errors
that are generated as part of the scanning process are ideally addressed in the form that these
disturbances were introduced. This is why adaptive notch filtering is inherently a one-
dimensional filtering technique.
It is clear that the tonal disturbance in the EDAR data that adaptive notch filtering addresses is
related to something physical and/or numerical about the way the EDAR instrument collects its
data. Periodic signals in sampled physical waveforms can be due to many things: physical
resonances, detuned oscillators within electronics, aliasing due to under-sampling, and even
interference patterns due to the combinations of high-frequency signals. One cannot figure out
what causes this disturbance without additional knowledge of the EDAR instrument itself. In
addition, the only way to completely remove this interference would be to modify the data
collection and/or processing of the EDAR data before it is delivered by the data collection
contractor. Adaptive notch filtering represents a reasonable way to address the tonal artifacts in
the EDAR data as delivered absent these modifications.
Interpolating Measured Pixel Positions to a Rectangular Grid
Processing Rationale - Each channel of each transit is measured according to the zigzag pattern
shown in Figure 7-6. So far, we have treated each zigzag measured pattern as if it is a two-
dimensional image, showing pixel values on a rectangular grid according to an approximate
spatial mapping of the EDAR data scan onto the roadway over time. It is clear, however, that this
type of representation is not entirely accurate. In reality, each dataset is a zigzag sampling of the
two-dimensional image of the mass in each data channel that passes under the EDAR instrument.
This zigzag pattern does not characterize mass in all regions similarly.
To illustrate how the zigzag sampling of the EDAR instrument affects the representation of the
two-dimensional data, Figure 7-9 shows a simple cartoon of what a completely smooth
exponentially decaying plume after a vehicle might look like using the standard color palate
employed throughout this report. The left-most panel shows what the EDAR instrument would
collect in such an idealized situation, where we show the measured values as a two-dimensional
7-17
-------
image. It is clear that there is "something wrong" in this image. There are discontinuities on the
left and right sides of this image despite the fact that this idealized plume has no discontinuities
in it. The center panel shows what is occurring with the measurement device. As the instrument
scans left-right-left-right-..., the plume signal is decreasing in the vertical direction only. If one
knows the position of the scanning element, then the representation in the center panel can be
interpolated in the vertical direction to reconstruct the underlying plume accurately. This
interpolation restores the equal spacing that is expected in a two-dimensional Cartesian
coordinate sampling of an image. The right-most panel shows this interpolated image. It was
generated using the measurements in the center panel only. The smoothness of the original data
is recovered, and there are no discontinuities in it. More importantly, the right panel represents a
spatially-accurate di stribution of the measured quantity - in this case, mass - that the left-most
panel can only approximate.
Figure 7-9. Cartoon Demonstrating Interpolation to Rectangular Grid
Measured Data ZigZag Location Model ZigZag Interpolated Image
Fortunately, the representation provided with the EDAR data allows us to recover the zigzag
nature of the sampling and perform this interpolation easily. This interpolation is done after the
adaptive notch filtering described in the previous subsection, and thus takes advantage of all
noise reduction methods applied previously, including the de-striping of the HC transit data and
the filtering of disturbances via adaptive notch filtering.
As an initial check of the methodology, Figure 7-10 shows an example interpolation performed
on the CO2 channel of an example transit. In this figure, the left-most "image" is the raw data.
This data exhibits the same discontinuities seen in the left panel of Figure 7-9. The center image
is an interpolated-and-upsampled version of the data in the left panel, where we have used the
proper zigzag placement of the data values to do the linear interpolation. It is important to
recognize that this enhanced image is not simply a two-dimensional extension of the data on the
left. The physical location of the zigzag pattern as illustrated in the center image of Figure 7-9
has been used to interpolate the values in the zigzag pattern in the left image of Figure 7-10 to
7-18
-------
make them two-dimensional with equal spacing between each row value in the center image of
Figure 7-10. A highly detailed plume structure with physically plausible plumes is reconstructed.
However, this upsampled version of the data has too many scan lines and would require
modification of all subsequent processing steps within the Westminster data analysis. Thus, we
instead use a subsampled version of this upsampled dataset, as shown on the right-most image of
Figure 7-10. Comparing the left-most image with the right-most image, one can surmise that the
interpolation method used here reconstructs a more physically plausible plume structure and thus
a more accurate representation of the mass measurements being collected by the EDAR
instrument.
Figure 7-10. Example of Interpolation of COa Data to a Rectangular Grid
Raw CO2 Data Interpolated + Upsampled Then Downsampled
seriesnum = 509, runnum = 2863 seriesnum = 509, runnum = 2863 seriesnum = 509, runnum = 2863
7-19
-------
Processing Examples - We illustrate the effects of zigzag interpolation through selected transits
of the Westminster dataset corresponding to the following known measurement conditions:
Test Vehicles
EV-1
EV-2
EvapHC Release Locations
DOOR
TANK
HOOD
EvapHC Emission Rates
200 mg/mile (low)
6400 mg/mile (high)
Nominal Road Speed
22.5 mph
Considering one example from each possible combination of the above conditions results in
twelve transits from the Westminster dataset.
Figure C-33 (in Appendix C) through Figure C-44 show images from these twelve transits in the
same data ordering as Figure C-21 through Figure C-32. In these datasets, the HC and CO2 data
channels show specific plume structures, so these two channels are used to illustrate the behavior
of the zigzag interpolation process in each case. This choice allows us to use more of the printed
or viewed page for zoomed-in image content. The four figures show the HC and CO2 data
channels for the selected transit. The two images on the left show the HC channel before and
after zigzag interpolation after both have been processed with multi-tonal cancellation and
adaptive notch filtering. The two figures on the right show the corresponding CO2 data channel
before and after zigzag interpolation after both have been processed with adaptive notch filtering.
The elongated view of each image allows one to more easily assess the effects of zigzag
interpolation as well.
These data can be explored and compared individually to see the qualitative effects of the
processing. In some cases, the improvements are subtle, whereas in others, the improvements are
more obvious.
As an example, Figure C-33 shows plume structures for both the HC and CO2 data channels that
are less "blocky" after interpolation as compared to before interpolation. The tell-tale sign for the
zigzag artifact is plume discontinuities that occur every two scan lines. There are fewer of these
discontinuities in the interpolated images.
As another example, Figure C-39 has an HC channel with a plume that is more physically
plausible after interpolation as compared to before interpolation. Generally, plumes should have
smoothly changing amplitudes in the direction of vehicle motion, and the HC interpolated image
exhibits this character in this example.
Figure C-40 shows obvious "blocky" artifacts in both the HC and CO data channels before
interpolation that are largely eliminated through the zigzag interpolation process. Finally, Figure
C-40 illustrates the measurement issue with the EDAR instrument in the CO2 channel. The
discontinuities in the CO2 plume on the right side of the image after the vehicle are largely gone
in the interpolated image.
7-20
-------
Processing Summary - As a point of background, almost all traditional optical cameras in use
today - phone cameras, laptop cameras, and the like - use a two-dimensional lens-based
collection process with an image sensor that measures pixels on a rectangular grid. A traditional
document copier or scanner uses a one-dimensional detector of parallel sensors and moves the
image across this parallel detector array uniformly, again creating pixels on a rectangular grid
but with potentially non-uniform spacing. The EDAR instrument has effectively only one sensor
that must move in a two-dimensional pattern to sense an image. It differs both from a traditional
camera and a copier or scanner. Moreover, because this instrument must move its sensing point
around, there are physical limits to where and when the device can sense information.
Using a zigzag scanning motion creates a sampling process that is non-uniform in the direction
of vehicle motion depending on the lateral position of the information across the roadway. Pixels
sensed in the center of the EDAR field of view are uniformly sampled. Pixels sensed near either
edge of the EDAR field of view are non-uniformly sampled and inherently have less spatial
detail in them.
The interpolation process described in this report takes care of the baseline issue with the zigzag
measurement process of the EDAR instrument: The pixels of the original EDAR measurement
are not true assessments of the mass in any Cartesian coordinate position except along the center
line of the camera field of view. After interpolation, each pixel of the interpolated image
represents a scaled version of the actual spatial information contained in the pixel quadrant.
An important issue is now raised. Because the EDAR instrument is performing spatial sampling
of a two-dimensional image field, the theory behind its sampling operation can be understood
using traditional 2D signal processing concepts. For example, for a plume to be properly
sampled, its spatial shape must satisfy a sampling criterion as understood viaNyquist sampling
theory. The zigzag form of the scanning process complicates this analysis somewhat, but it is
possible to do. But most importantly, it is not clear that the EDAR instrument is performing an
adequate sampling of the spatial plume field. Said more simply, we do not really know how
"spatially rough" the plumes are behind travelling vehicles. Maybe there are peaks and valleys in
these plumes that are being missed by the EDAR instrument, or maybe the plumes themselves
are smooth. As far as we are aware, no one has measured the smoothness of a plume behind a
travelling vehicle. If these physical details about the underlying signals were known, then the
EDAR data collection process could be designed to specify a scanning rate to ensure that the
entire mass of the plume could be captured and assessed. In any event, for the purposes of better
quantifying pollutant release rate (g/hr) and emission rate (g/mile), it would be worthwhile to
explore new collection strategies - and even new sensor designs - that would capture more scan
lines in the direction of vehicle motion with a much smaller spacing or timing between them.
For the original EDAR design goal of measuring pollutant concentrations, the zigzag scanning
method is, by no means, a limitation or problem. The reason is that EDAR's determination of
concentrations is based on measurement of the ratio of a pollutant signal to the CO2 signal at
each pixel. The locations or spacings of the pixels are of little consequence to accurate
determinations of concentration as long as many pixels are in the vortex.
7-21
-------
7.2 Blind Source Separation by Independent Component Analysis
Processing Rationale - A vehicle powered by an internal combustion engine (ICE) emits
compounds as it travels along the road. These emitted compounds create different spatial patterns
depending on the location of their emission from the vehicle. Some of these spatial patterns
correspond to multiple material types. For example, for an internal combustion engine vehicle,
multiple emitted materials are typically expended from the exhaust, including CO2, CO, HC, and
NOx. The spatial patterns of these emitted compounds are largely the same if they are present. In
other cases, the spatial pattern corresponds to one material type. For example, a vehicle leaking
hydrocarbon vapor from the fuel fill door will not have any other compounds, such as CO2 or
NO, emitted from the same fuel fill door location. There are typically a small number of
locations around the vehicle from which compounds are emitted, where the chemical makeup of
these emissions is different at each emitted location. This situation is what allows blind source
separation processing to yield useful results.
Blind source separation is used in this project to identify unique spatial patterns that can
correspond to different plumes of emitted materials around the vehicle. Blind source separation
tries to make output images that are spatially different from each other using the five channels of
RSD data (HC, CO2, CO, NO, and NO2) obtained for each transit. This processing is performed
on the improved data from each transit separately after the pre-processing steps are performed.
The mathematical procedure is described in Section 5.3.
Note that the spatial patterns produced after separation are images that look like plume images of
a particular compound, but their amplitudes are no longer in units of mass. In other words, the
separation procedure produces normalized plume patterns as outputs. They do not have the
correct amplitude to correspond to any particular mass of any particular chemical compound.
That is why a second step of processing - estimation - is performed, to be described later.
Since there are five RSD channels going into the BSS procedure, five possible plume patterns are
produced from the procedure. If a smaller number of channels from the RSD are selected for
processing, then the number of possible plume patterns produced from the BSS procedure is also
reduced. Because of how ICE-powered vehicles work, only two types of situations are expected.
Either 1) only an exhaust plume is present, or 2) both an exhaust plume and an evaporative
plume are present. The exhaust plume will contain some amount of CO2. The evaporative plume,
if it is present, will be most similar to the HC data channel. Thus, we use the procedure described
in Section 5.3 to select two of the BSS spatial patterns to identify an exhaust plume and an
evaporative plume for each transit. What remains after these patterns are considered noise
plumes and are discarded. No processing is performed on these noise patterns other than to
identify them as noise.
Processing Examples - In order to show various processing scenarios of interest, we have
selected the dataset types shown in Table 7-2 from the test vehicle set to illustrate BSS results in
this section.
7-22
-------
Table 7-2. Criteria for Selection of Processing Examples
Test Vehicles
EV-1:
Metered CO2 artificial exhaust emissions
EV-2:
Metered HC, CO2, CO, and NO artificial exhaust emissions
Subaru:
Sedan shape with natural exhaust emissions
F-150:
"Capped" pickup truck shape with natural exhaust emissions
EvapHC Release Locations
DOOR
TANK
HOOD
EvapHC Emission Rates
6400 mg/mile (high)
800 mg/mile (medium)
200 mg/mile (low)
Nominal Road Speed
22.5 mph (slow)
45.0 mph (fast)
Using these choices along with some controlled zero-emission EV transits, and selecting fewer
fast-speed transits, results in 52 different separation examples. These are shown in Appendix D.
All of these examples have the following form. Ten images are shown. The top five images are
the transit data channels input to the standard BSS procedure, labeled by compound type and
given by HC, CO2, CO, NO, and NO2. The bottom five images are the separated plume outputs,
labeled as evaporative plume (EvapPlume), exhaust plume (ExhPlume), and the three remaining
output channels, termed Noisel, Noise2, and Noise3, respectively.
Regarding all of the results, some general comments can be made:
1. The exhaust plume, when present, typically looks most like the CO2 input channel. Note
that BSS uses all data channels to estimate each plume. So, in fact, this plume is typically
cleaner than the CO2 channel alone when other data channels also have clear exhaust
plume structures, such as in the EV-2 transits.
2. The evaporative plume, when present, typically looks mostly like the HC channel with
any HC exhaust emissions removed. For the EV-1 transits, there is no HC exhaust
emissions, so successful processing should result in the HC channel and the EvapPlume
looking highly similar. For the EV-2 transits, there is HC exhaust emissions present, so
successful processing should result in EvapPlume appearing to be the portion of the HC
channel that is different from ExhPlume.
3. Any apparent plume signal in Noisel, Noise2, or Noise3 is indicative of some aspect of
the process - either inherent within the measurements or due to the processing - that does
not fit the standard evaporative emissions / exhaust emissions model. However, if these
"noise plumes" correspond to an apparent signal within the CO, NO, or NO2 channels
that looks different from the CO2 channel, then BSS has successfully isolated these
"erroneous plumes" into channels that effectively isolate them from the EvapPlume and
7-23
-------
ExhPlume outputs. Such a processing result is considered successful given the nature of
the input signals to the BSS process.
4. One tell-tale error that can occur with standard BSS processing is a type of "over-
subtraction" of the evaporative plume portion of the HC transit signal, causing a negative
value that is typically hole-shaped due to the concentrated nature of the exhaust plume at
the vehicle tailpipe. This "blue-hole" phenomenon is understandable from the constraints
imposed by standard BSS processing, which assumes no spatial overlap of plumes. These
observed artifacts were the inspiration behind the development of the BSScov separation
procedure developed under this project and discussed in the next subsection.
With these general comments in place, we now discuss the individual transit results.
Figure D-l shows a separation example for EV-1 where no emissions of any kind are present. In
this case, BSS produces a near-zero EvapPlume and a near-zero ExhPlume, as it should. No
spurious signals are created.
Figure D-2 shows a separation example for EV-1 where only CO2 tailpipe emissions are present.
In this case, BSS produces a zero EvapPlume and an ExhPlume that is nearly-identical to the
CO2 input signal - a correct result.
Figure D-3 shows a separation example for EV-1 where a high-level evaporative release occurs
from the simulated fuel-fill door away from the simulated tailpipe location where CO2 is
released. In this case, BSS correctly estimates an EvapPlume similar to the HC input signal and
estimates an ExhPlume similar to the CO2 input signal - a correct result. There is little spatial
overlap of these two plumes.
Figure D-4 shows a separation example for EV-1 where a medium-level evaporative release
occurs from the simulated fuel-fill door away from the simulated tailpipe location where CO2 is
released. As in Figure D-3, BSS correctly estimates an EvapPlume similar to the HC input signal
and estimates an ExhPlume similar to the CO2 input signal - a correct result. The evaporative
plume is weaker due to the lower-level release rate. There is little spatial overlap of these two
plumes.
Figure D-5 shows a separation example for EV-1 where a low-level evaporative release occurs
from the simulated fuel-fill door away from the simulated tailpipe location where CO2 is
released. As in Figure D-3 and Figure D-4, BSS correctly estimates an EvapPlume similar to the
HC input signal and estimates an ExhPlume similar to the CO2 input signal - a correct result.
The evaporative plume is weaker still due to the low-level release rate. There is little spatial
overlap of these two plumes.
Figure D-6 shows a separation example for EV-1 for a high-speed transit, in which a medium-
level evaporative release occurs from the simulated fuel-fill door away from the simulated
tailpipe location where CO2 is released. As in Figure D-4 and Figure D-5, BSS correctly
estimates an EvapPlume similar to the HC input signal and estimates an ExhPlume similar to the
CO2 input signal - a correct result.
7-24
-------
Figure D-7 shows a separation example for EV-1 where a high-level evaporative release occurs
from the simulated gas tank underneath the center of the vehicle rear and close to the simulated
tailpipe location where CO2 is released. There is a large spatial overlap between the evaporative
plume and the exhaust plume. In this case, BSS estimates EvapPlume as the portion of the HC
input signal that overlaps with, yet is spatially different from, the CO2 input signal that largely
makes up ExhPlume. While this result looks reasonable, it is in fact incorrect for this particular
transit, as no HC was released from the tailpipe for EV-1. This result is largely due to the
constraint imposed by standard BSS procedures that use statistical independence as the
separation measure. EvapPlume and ExhPlume are two plumes that are spatially distinct, but
they do not fit the measurement scenario in this case due to plume overlap. The BSScov
procedure discussed in the next subsection was designed for this type of scenario.
Figure D-8 shows a separation example for EV-1 where a medium-level evaporative release
occurs from the simulated gas tank underneath the vehicle rear and close to the simulated tailpipe
location where CO2 is released. As in Figure D-7, there is a large spatial overlap between the
evaporative plume and the exhaust plume. Like this previous example, BSS estimates
EvapPlume as the portion of the HC input signal that overlaps with, yet is spatially different
from, the CO2 input signal that largely makes up ExhPlume. Again, the result is reasonable for
an independence-based separation system but does not fit this measurement scenario. EvapPlume
is also noisier due to the weaker evaporative emissions release.
Figure D-9 shows a separation example for EV-1 where a low-level evaporative release occurs
from the simulated gas tank underneath the vehicle rear and close to the simulated tailpipe
location where CO2 is released. Due to the low-level release, there is less apparent spatial
overlap of the two plumes. BSS estimates EvapPlume as the portion of the HC input signal that
overlaps with, yet is spatially different from, the CO2 input signal that largely makes up
ExhPlume. The result appears to be more reasonable here because of the low-level nature of the
evaporative release.
Figure D-10 shows a separation example for EV-1 for a high-speed transit, in which a medium-
level evaporative release occurs from the simulated tank near the simulated tailpipe location
where CO2 is released. Because of the weak nature of the evaporative signal, and its dispersed
spatial signature, there is little spatial relationship between it and the tailpipe plume seen in the
CO2 channel. Thus, standard BSS does a good job of maintaining the evaporative signal in
EvapPlume.
Figure D-l 1 through Figure D-15 all show separation examples for EV-1 where the evaporative
release occurs from underneath the hood. For this vehicle shape, the precise release location
under the hood, and wind velocity, the resulting evaporative plume largely appears on the side of
the vehicle and away from the simulated tailpipe location where CO2 is released. Since there is
little spatial overlap between the evaporative and exhaust plumes, EvapPlume and ExhPlume in
all four separation examples are distinct and largely are correct in their isolation of the HC
channel in EvapPlume and the isolation of the CO2 channel in ExhPlume. The main differences
in all four of these examples is the relative levels of signal in the HC channel input to the
standard BSS procedure.
7-25
-------
Figure D-15 shows a separation example for EV-2 where no emissions of any kind are present.
Like Figure D-l, BSS produces a near-zero EvapPlume and a near-zero ExhPlume - as it should.
No spurious signals are created.
Figure D-l6 shows a separation example for EV-2 where tailpipe emissions with four different
pollutants - HC, CO2, CO, and NO - are present, and there are no evaporative emissions. In this
case, BSS produces a nearly-zero EvapPlume and an ExhPlume that is similar to all four of the
different plumes for each of the four different pollutants. The BSS procedure correctly merges
the four different compound signals into one plume pattern.
Figure D-l7 shows a separation example for EV-2 where tailpipe emissions with four different
pollutants - HC, CO2, CO, and NO - are present, and where a high-level HC evaporative release
occurs from the simulated fuel-fill door which is located near the rear bumper but on the other
side of the vehicle's tailpipe location. This is our first example of a composite release of both
evaporative HC and exhaust HC in a transit signal. Here, EvapPlume has a "blue hole" in the
tailpipe location, indicating that standard BSS has over-subtracted the emissions component in
the evaporative plume and created a negative-valued signal in the tailpipe location. This result is
to be expected due to the independence criterion used by standard BSS techniques and is one of
the motivating scenarios for which BSScov was designed. Standard BSS does a good job of
estimating the evaporative plume in EvapPlume from the multiple input signals.
Figure D-l8 shows a separation example for EV-2 where tailpipe emissions with four different
pollutants - HC, CO2, CO, and NO - are present, and where a medium-level HC evaporative
release occurs from the simulated fuel-fill door which is located near the rear bumper but on the
other side of the vehicle's tailpipe location. In this case, standard BSS processing produces a
reasonable result, isolating the evaporative plume in EvapPlume from the exhaust plume in
ExhPlume, and correctly combining the multiple tailpipe signals into one ExhPlume. No
spurious artifacts are created in any Noise outputs.
Figure D-l9 shows a separation example for EV-2 where tailpipe emissions with four different
pollutants - HC, CO2, CO, and NO - are present, and where a low-level HC evaporative release
occurs from the simulated fuel-fill door which is located near the rear bumper on the other side
of the vehicle's tailpipe location. In this case, standard BSS processing produces a fairly
reasonable result for EvapPlume and ExhPlume, but it also generates spurious artifacts in at least
one of the Noise outputs near the tailpipe location. The performance of standard BSS in this case
is somewhere in-between the results of the previous two examples.
Having provided these three examples, we can consider the performances of the remaining EV-2
transit examples in Figure D-20 through Figure D-27 as being similar in characteristics to one of
these three above. Examining the figures, we find that:
BSS performance similar to Figure D-18 (reasonable): Figures D-20, 21, 23, 26, 28
BSS performance similar to Figure D-l9 (in the middle): Figures D-22, 24, 27
BSS performance similar to Figure D-l7 (blue hole): Figure D-25
7-26
-------
These behaviors are likely due to the specific plume structures and emission levels of these
various transit examples matching those of the three initial EV-2 examples provided above.
Figure D-29 through Figure D-40 show separation examples for the Subaru test vehicle, which
has natural CO2 tailpipe emissions and perhaps a small amount of HC tailpipe emissions as well,
although the latter is likely much below the release amounts from the simulated evaporative
emissions in every measured case. In addition, it is possible for the Subaru to emit other
compounds, like CO, depending on its engine state. Examining these twelve figures, we can
classify them into three different performance categories:
Reasonable BSS performance, CO present: Figures D-32, D-34, D-36. In these cases, the
vehicle appears to emit similar exhaust plumes containing both CO2 and CO. ExhPlume
largely follows the spatial pattern of these CO2 and CO emissions. The HC channel
contains emissions that are largely different from the tailpipe emissions, and these are
correctly isolated in EvapPlume.
Reasonable BSS performance, no CO emissions present: Figures D-30, D-36, D-39. In
these cases, the vehicle appears to emit only a CO2 exhaust plume. ExhPlume largely
follows the spatial pattern of the CO2 emissions. The HC channel contains emissions that
are largely different from the tailpipe emissions, and these are correctly isolated in
EvapPlume.
Artifacts in the CO channel: Figures D-29, D-31, D-33, D-35, D-37, D-40. In these cases,
the vehicle appears to emit at least two different kinds of plumes in the CO2 and CO
channels. This type of behavior is non-physical, as internal combustion engines would
generate similar CO2 and CO plumes if they are being produced from the same engine -
even if the engine generating the CO is from another nearby vehicle or combustion
source. Thus, it is initially unclear how to evaluate the performance of standard BSS in
these cases. However, it can be seen that, if the CO channel is thought to be in error, then
the so-labelled erroneous components within the CO channel are isolated in the Noisel
channel - a correct result. In addition, the separation performance for EvapPlume and
ExhPlume under this interpretation appears to be reasonable.
Figure D-41 through Figure D-52 show separation examples for the F-150 pickup truck test
vehicle, which has natural CO2 tailpipe emissions and perhaps a small amount of HC tailpipe
emissions as well, although the latter is likely much below the release amounts from the
simulated evaporative emissions in every measured case. Examining these twelve figures, we can
classify them into two different performance categories:
Reasonable BSS performance, no CO emissions present: Figures D-48, D-49, D-52. In
these cases, the vehicle appears to emit only a CO2 exhaust plume. ExhPlume largely
follows the spatial pattern of the CO2 emissions. The HC channel contains emissions that
are largely different from the tailpipe emissions, and these are correctly isolated in
EvapPlume.
Artifacts in the CO channel: Figures D-41, D-42, D-43, D-44, D-45, D-46, D-47, D-50,
D-51. In this case, a large number of the measured transits for the F-150 test vehicle
7-27
-------
contain artifacts in the CO data channel. These artifacts are non-physical and make
evaluating the separation performance of standard BSS challenging for this vehicle type.
A particular standout in this group is Figure D-46, which appears to contain three
different plume structures in the CO2, CO, and NO channels despite this scenario being
an impossibility from a physical perspective.
Processing Summary - The goal of BSS processing in this project is to isolate specific
candidate plume patterns from spatially-oriented RSD measurements of moving vehicles without
regard to, and without knowledge of, the exact emission characteristics of any one vehicle being
analyzed. If one knows the exact nature of the emissions from a vehicle, then one can design a
specific processing method to extract the requisite plume patterns from the vehicle. The problem
of course is the lack of knowledge of this exact nature for any one vehicle. Applying a precise
processing method to a vehicle with an unknown state would likely result in erroneous results.
This is the reason for using BSS methods in this project. They generally work well without
having precise knowledge of the emissions characteristics of any one vehicle being analyzed.
The examples provided show that standard BSS often does a reasonable job of estimating the
exhaust plume of vehicles and, when present, isolating the evaporative plume of emitted HC
components from the exhaust plume. This is the first step in performing an estimation of the
evaporative and exhaust components of a specific pollutant, such as HC.
Standard BSS also appears to deal with non-idealities in the measured RSD data that do not fit
the emissions model. In particular, non-physical CO components generate "noise plumes" that
are then rejected for further processing.
The primary drawback to standard BSS processing occurs when exhaust and evaporative plumes
have a significant spatial overlap. In such cases, standard BSS processing can sometimes create
erroneous results by "over-subtracting" the measured signals from each other, leading to
negative plume regions. This issue was recognized early on in this project, and the BSScov
algorithm, described in the next section, is one possible methodology that can be used to address
the issue when it occurs.
7-28
-------
7.3 Enhanced Blind Source Separation using Correlation Constraints
Processing Rationale - Blind source separation based on independent component analysis uses
the property that the signals being separated - in this case, spatial plume signatures - are
statistically independent of each other. This property is a good one to use for RSD measurements
in this project when the evaporative and exhaust plumes are very different - for example, when
the locations of the plume signatures do not overlap very much. This property is less accurate if
there is a lot of overlap between the exhaust plume and the evaporative plume. For this reason, a
new BSS method was developed under this effort to handle this situation. Called BSScov, the
separation algorithm allows one to specify a parameter, called a correlation parameter and
denoted as p (the Greek symbol "rho"), to specify the amount of overlap between the two
plumes. This parameter should ideally be adjusted based on the amount of overlap of the two
plume types, and it is likely best chosen using the emission location of the evaporative plume.
The BSScov procedure is described in mathematical detail in Section 5.3.
While BSScov provides a potential solution to the issue of overlapping plumes and the proper
assignment of plume structure to EvapPlume and ExhPlume for real-world RSD measurements,
it should be noted that a precise procedure for specifying the value of the correlation parameter p
has not yet been developed. However, evaluations of the candidate procedure on the Westminster
and other similar RSD measurements of vehicle transits indicate that:
The correlation parameter need only be specified for the relationship between EvapPlume
and ExhPlume. Hence, there is only one parameter to be set for each processed transit.
A typical range of the correlation parameter is 0 < p < 0.3. A zero value yields the
standard BSS procedure. Values greater than 0.3 lead to non-physical results.
Since BSScov includes standard BSS processing as a special case, we can obtain our
existing results by selecting p = 0. There is no need to "switch" between algorithms. Non-
switching processing strategies are more robust.
Performance varies smoothly for small changes in the correlation parameter value. In
other words, if the value of the parameter has a small error, then performance will
degrade by a small amount. Thus, adjusting the correlation parameter does not involve
significant risk.
When the correlation parameter p is chosen properly, non-physical artifacts, such as large
negative values in the extracted plume signatures, tend to be suppressed. Thus, the
algorithm has the potential to achieve the desired goal of artifact-free plumes for
estimating both exhaust and evaporative emissions.
Processing Examples - In this section, we illustrate the behavior of the BSScov algorithm
through selected examples. Each of these examples illustrates the separation behavior of the
BSScov algorithm on a specific vehicle transit dataset for a range of correlation parameter
values. The purpose of these examples is to show that a proper value of the correlation parameter
can likely be set once a specific criterion for its design has been chosen. The examples also
illustrate the limitations of the standard BSS approach.
7-29
-------
Figure 7-11 shows the input data channels (top five images) and the results of standard BSS
processing (bottom five images) for an EV-1 vehicle transit with a large evaporative release at
the TANK location. Recall that the TANK release location typically results in an evaporative
plume that significantly overlaps with the exhaust plume, especially for high-level releases.
Standard BSS processing leads to negative values in EvapPlume at the locations where
ExhPlume is large. These can be seen by the two "blue holes" in EvapPlume next to the back
bumper of the vehicle.
Figure 7-12 shows the same input data illustrated in Figure 7-11 (top five images) processed by
the BSScov algorithm with the choice p = 0.1. The "blue holes" in EvapPlume are mitigated
without significant changes in ExhPlume.
Figure 7-13 shows just the HC and CO2 data channels (left two images) as well as the
corresponding EvapPlume and ExhPlume results for standard BSS processing (second-from-the-
left two images) along with the BSScov algorithm's plume outputs for the values of p = 0.05, p =
0.1, and p = 0.15 (right six images). One can see that, as the correlation parameter is increased,
the plume patterns in EvapPlume and ExhPlume slowly change, and a value of this parameter
can be selected to mitigate the "negative holes" issue identified in the previous standard BSS
example. The proper choice of this correlation parameter is currently an open issue. This figure
illustrates, however, that the algorithm's outputs have the desired range of outputs to mitigate the
undesired artifacts of standard BSS processing.
Figures E-l through E-12 show the examples shown in Figures 7-11, 7-12 and 7-13 plus three
more examples comparing standard BSS processing with the results of the BSScov algorithm, in
which differing release locations (DOOR, HOOD) and differing amounts of evaporative releases
are considered. In each case, the BSScov algorithm provides a range of outputs that enable a
reasonable selection of the correlation parameter to obtain physically plausible patterns of
EvapPlume and ExhPlume.
Processing Summary - The BSScov algorithm was designed with the specific goal of
mitigating a known artifact in standard BSS processing when applied to data that does not
precisely fit a statistically-independent source model. It is impossible for two positive-valued
signals that have any degree of overlap to be statistically independent, because they must be non-
negative wherever they overlap, by definition.
The primary issue in using the BSScov algorithm is setting the value of the correlation
parameter. Some exploration has been performed in this direction. It is likely a problem
involving calibration of the processing approach to the specific RSD measurement instrument
collecting the data, where test vehicles with known release amounts can serve as training data. It
is also likely that knowledge of the evaporative emission location, if such information could be
identified on a per-transit basis, could be used to choose a reasonable value of the correlation
parameter. Such procedures are the subject of potential future work.
7-30
-------
Figure 7-11. BSS ICA Separation (p=0) of Example: EV-1, High EvapHC from TANK
HC
5
10
® 15
-Q
E
z 20
c
CT3
O
w 25
30
35
l! I H
I I ill i
¦ ii r i
;
"i
0 100 200
Scan Position
7 20191020 000505 car 001188
C02 CO
1-EV1 6400mg Tank 22mph Y
NO
N02
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
EvapPlume
5
10
® 15
.Q
£
z 20
c
n:
o
w 25
30
35
i III
i | Jill fi
I I I: i III
IlllOII'J ' 111
liliTl ill
I'll!! vi ill!
nir k
¦
mi ii
0 100 200
Scan Position
rho = 0
ExhPlume
Noisel
0 100 200
Scan Position
0 100 200
Scan Position
Noise2
i in
Ij IIin'nl Ii
i i,
mi
V1! In il
H'j illi II I
i i H
i, jiii ii
0 100 200
Scan Position
NoiseS
0 100 200
Scan Position
7-31
-------
Figure 7-12. BSScov Separation (p=0.1) of Example:
7 20191020 000505 car 001188
EV-1, High EvapHC from TANK
1-EV1 64Q0mg Tank 22mph Y
CO
0 100 200
Scan Position
0 100 200
Scan Position
If)
, _
ilia1)!!
1. 01
I'llIVfJfM'/i i
¦
in, 1 "i , i| 1 il W1 II
"" il l ' I1 "lll'l "
ii'il
I1 111
i ,1 1 i1
(ill
Hil
ii i
i_
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
EvapPlume
rho = 0.1
ExhPlume
0 100 200
Scan Position
0 100 200
Scan Position
Noisel
ii ill1'! liil11lib il'll
1,11 il II I , II llllf
11 1 j I ft
IJIrl
ii j, i
i
pi
HI I I, I I
in lrri!|.i«'T«iiii,,
I I 111 III II I!
I | I m Ii ii i,i n
l,n I 'il l III Jin', In L'l
l
i
I ii,Ml i1'
III :I III I: III
0 100 200
Scan Position
Noise2
5
10
15
20
25
30
35
im
ii
,i
' B'M
mill
i
111 r lllll,
111:
' Fill
III V!
t I I
I I
I II
i ; i i idj
,i ii 11 n
uiu
hi
u
0 100 200
Scan Position
5
10
15
20
25
30
35
Noise3
T |
i
1 II 1
If
1
i! 'fli li '11 '}
1 1 i
il
Hi V
J' m
j I |ii ii! I
M |i
.
1
¦
1 ii ' I
i"
j
i i
0 100 200
Scan Position
7-32
-------
Figure 7-13. Evaluation of Plume Outputs while Varying p for Example: EV-1, High EvapHC from TANK
7 20191020 000505 car 001188 1-EV1 6400mg Tank 22mph Y
HC EvapPlume EvapPlume, />=0.05 EvapPlurne, />=0.10 EvapPlume, ,o=0.15
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
" Muuiuii,!,!
lit i
i 11111
in . I u
III I
J !l'"
11, 111 "
nil hiiil ,
,i. In
Mi .
hi ii n.
¦iirin
0 100 200
Scan Position
ExhPlume, p=QA5
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
7-33
-------
7.4 Estimation of EvapHC and ExhHC from Candidate Plumes
Processing Rationale - The Blind Source Separation portion of the Separation/Estimation
Device creates an ExhPlume pattern and an EvapPlume pattern from the improved RSD signals
presented to it. These plumes are effectively unitless. They are spatial patterns, but they do not
specify the amounts of any one compound in any measurement from the RSD. To find these
amounts, an estimation procedure must be performed. The estimation portion of the
Separation/Estimation Device is a mathematical procedure that combines the ExhPlume and the
EvapPlume from the BSS portion with the improved RSD signals to figure out the scaling of
each plume needed to match the portion of that plume in the corresponding RSD signal. The
output of the estimation procedure is a mass image corresponding to the type of emission plume
for each compound that we choose to estimate. In this project, we applied estimation to the
improved HC signal only, although it could be applied to other RSD signals as well.
Using this procedure, we have the following advantages:
1. The plumes produced by the BSS procedure are further "cleaned up" with respect to the
improved RSD signals, so they can be more accurate than any one improved RSD signal.
2. Using an exhaust plume and an evaporative plume allows us to "divvy up" the HC signal
into an ExhHC component and an EvapHC component. The two components are not
obtainable by direct examination of the HC signal alone.
The problem of estimating the height of a signal given a candidate template for this signal is
well-known and involves an estimation procedure called regression. In this project, we used the
structure of the data to better estimate the exhaust HC plume and the evaporative HC plume. In
particular, we used weighting factors determined from the analysis of the entire Westminster
dataset as described in other portions of this document, and as detailed in Section 5.4. This
procedure is called weighted least squares.
Weighted least-squares is a standard approach in regression analysis. How the weighting is
applied to the Westminster data is now explained. This process involves three steps:
1. Generation of one-dimensional weighting functions
2. Extension of these one-dimensional weighting functions to two-dimensional weighting
functions within the transit data
3. Multiplication of the two-dimensional weighting functions with measured data.
These procedures are used to combine the estimated plume patterns, EvapPlume and ExhPlume,
with the measured HC transit data to determine the proper height of the component of each of
these plumes that make up the HC transit data. The end results are scalar values that scale the
"unitless" plumes into estimated components of the measured data, with units of mass that match
the units of the measured data. There is a single scalar value for each particular plume type,
EvapPlume and ExhPlume, for each transit.
7-34
-------
Note that the resulting images of both the ExhHC plume and the EvapHC plume for each transit
look just like the unitless ExhPlume and EvapPlume produced by the BSS procedure. Thus, we
do not show the images of this process here. This estimation procedure is required to leverage
the outcomes produced by the BSS procedure, however.
There are additional choices of least-squares methods that could be made. For example, we could
use a constraint to specify the total amount of hydrocarbon produced in the two estimated HC
plumes. This procedure is known as constrained least squares. We could also use an estimate of
the baseline noise level in the improved HC signal to adjust the estimates to account for the noise
amount. This procedure is known as total least squares. Both constrained least squares and total
least squares could use weighting factors to improve their accuracy, resulting in weighted
constrained least squares and weighted total least squares procedures, respectively. There are
many possible choices for this estimation procedure. We chose weighted least squares due to its
simplicity and the lack of additional knowledge required for the transit data.
Processing Examples - In this section, we illustrate, via several examples, how the one-
dimensional weight functions are extended to two dimensions so that they can be applied to
measured RSD data.
Figure 7-14 shows two identical one-dimensional plots along the first column of the figure.
These one-dimensional plots are the weighting functions determined by the data analysis
procedure described in Section 5.4. They have been rotated by 90 degrees so that they are
oriented along the scan number of the transit data. They have also been shifted to the back
bumper position so that they match the position of the vortex as it appears after the vehicle in the
transit data. The peak of this function is typically at the same position as the first full line of pixel
values after the back bumper of the vehicle.
The second column of Figure 7-14 shows the two-dimensional weighting image generated from
the one-dimensional weighting function. This image is simply the weighting function applied to
every scan position in the image, where we have turned this function into pixel values and
displayed the weighting function as a two-dimensional image. The weighting function is only
non-zero over pixels that are not vehicle pixels. Hence, there are portions of the weighting
function that are "zeroed-out" by the position of the vehicle in the transit data.
The third column of Figure 7-14 shows the HC and CO2 data after pre-processing.
The fourth column of Figure 7-14 shows the HC and CO2 data after it has been weighted by the
two-dimensional weighting function in the second column. In this weighted image, the weighting
suppresses pixels that are far from the back bumper, because these pixels are likely to contain
only noise and very little plume mass of any type. Thus, using weighting functions helps to
improve the robustness of the estimation by suppressing noise pixels that add little to the
accuracy of the assessment.
The fifth column of Figure 7-14 shows the EvapPlume and ExhPlume generated by standard
BSS that is then combined with the weighted data in the fourth column to determine the proper
height of the estimated evaporative and exhaust components. These plumes can be combined
with either the weighted HC or the weighted CO2 image data. If both EvapPlume and ExhPlume
7-35
-------
are combined with weighted HC, one obtains the height of the evaporative HC plume and the
exhaust HC plume, respectively. Similarly, ExhPlume can be combined with the weighted CO2
to determine a more-accurate exhaust CO2 plume. Since CO2 is not an evaporative emission, it
makes no sense physically to combine EvapPlume with weighted CO2 in this application.
Figures 7-15, 7-16, and 7-17 show three additional examples of the weight function used in
estimation of emission components, the corresponding weight images, the measured HC and CO2
after pre-processing, the weighted versions of this data, and the corresponding EvapPlume and
ExhPlume patterns generated from standard BSS for this data in the same ten-image format as
Figure 7-14. In each case, the weighting images applied to the pre-processed data suppress noise
pixels and provide for an improved estimation accuracy. In some cases, the weighting function
suppresses signal energy that is not located right after the vehicle. Such an example can be found
in Figure 7-17 for a high-level hood release for the EV-1 test vehicle.
7-36
-------
Figure 7-14. Use of Weights for Estimation for Example: EV-1, High EvapHC from DOOR, Low Speed
7 20191021 000507 car 001350 1-EV1 6400mg Door 22mph Y
Weight Function
Weight Image
Weighted HC
EvapPlume
0 0.5 1
Weight Value
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
Weight Function
Weight Image
C02
Weighted C02
ExhPlume
0 0.5 1
Weight Value
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
7-37
-------
Figure 7-15. Use of Weights for Estimation for Example: EV-1, Low EvapHC from TANK, Low Speed
7 20191020 000506 car 000244 1-EV1 200mg Tank 22mph Y
Weight Function
Weight Function
7-38
-------
Figure 7-16. Use of Weights for Estimation for Example: EV-1, Medium EvapHC from TANK, Highspeed
Weight Function
7 20191020 000505 car 001347
Weight Image
0 0.5 1
Weight Value
0 100 200
Scan Position
HC
ll I 4
0 100 200
Scan Position
1-EV1 800mg Tank 45mph Y
Weighted HC EvapPlume
0 100 200
Scan Position
0 100 200
Scan Position
Weight Function
Weight Image
CQ2
Weighted C02
ExhPlume
0 0.5 1
Weight Value
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
7-39
-------
Figure 7-17. Use of Weights for Estimation for Example: EV-1, High EvapHC from HOOD, Low Speed
7 20191020 000505 car 001812 1-EV1 6400mg Hood 22mph Y
Weight Function
Weight Image
Weighted HC
EvapPlume
0 0.5 1
Weight Value
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
7-40
-------
Processing Summary - The goal of weighted least-squares is to improve upon standard least-
squares regression by using weightings that emphasize useful signals and suppress noise.
Generally, large weight values are used whenever desirable signal components are much larger
than undesirable noise components. Conversely, small weight values are used if the measured
signals values are dominated by noise. The ideal weighting function is a computed function of
the signal-to-noise-ratio of the measured data, from signal processing theory. In this application,
we do not have a clear idea of where the plume pixel values will be present or absent precisely.
However, we can accurately assume that any plume of interest will be strongest nearest the back
of the vehicle, and its strength will decrease with distance from the back bumper. The one-
dimensional weighting functions used in the estimation procedure have this precise structure.
The one-dimensional weight functions ignore the precise position of the plume as it appears
along the vehicle bumper within the vortex. It is likely that using this lateral position information
could further improve the estimation accuracy of the procedure. Applying a two-dimensional
weight function to the estimation procedure is mathematically straightforward, as a simple two-
dimensional weighting function is already being employed within the software. Two issues
remain: 1) an understanding of the proper weighting template as a function of plume and transit
parameters and b) a mathematical model that encodes this understanding in numerical form.
As a preview of the former issue, Figure 7-18 shows average CO2 plumes generated from EV-1
test vehicle data (top row) and EV-2 test vehicle data (bottom row) as a function of the airspeed
component parallel to the direction of vehicle motion. In this case, each transit has been aligned
to a reference position so that the appropriate pixel average across non-zero transit pixels can be
accurately performed. Each of these figures can be compared to the second column of Figures 7-
14 through Figure 7-17. The images in Figure 7-18 contain plume structure from side-to-side in
the vehicle transit that is not apparent in the weight images of the previous figures. These two-
dimensional plume patterns can potentially be a source of more-accurate weight functions for
estimating plume components.
Figure 7-19 shows average CO2 plumes generated from EV-1 for two different nominal road
speeds - 22.5 mph (top row) and 45 mph (bottom row). Similar data for EV-2 is provided in
Figure 7-20. In these images, the different images from left to right illustrate average plumes as a
function of the airspeed component perpendicular to the direction of vehicle motion. The
structures of these two-dimensional plumes have not been analyzed, but they could represent a
starting point for improved weighting functions as they are easily generated.
7-41
-------
Figure 7-18. 2-Dimensional CO2 Plume Averages for Different Parallel AirSpeed Ranges: EV-1 and EV-2
Top Row: EV-1 Bottom Row: EV-2
AirSpeedPara < 13.5 13.5 < AirSpeedPara < 22.2 22.2 < AirSpeedPara < 36.6 AirSpeedPara > 36.6 mph
CD
•Q
E
cc
o
W
100 200
Scan Position
100 200
Scan Position
100 200
Scan Position
100 200
Scan Position
AirSpeedPara < 13.5 13.5 < AirSpeedPara < 22.2 22.2 < AirSpeedPara < 36.6 AirSpeedPara > 36.6 mph
CD
-Q
E
CO
0 100 200
Scan Position
100 200
Scan Position
100 200
Scan Position
100 200
Scan Position
7-42
-------
Figure 7-19. 2-Dimensional CO2 Plume Averages for Different Perpendicular AirSpeed Ranges: EV-1 at Low and High Speeds
Top Row: EV-1, 22.5 mph Bottom Row: EV-1, 45 mph
7-43
-------
Figure 7-20. 2-Dimensional CO2 Plume Averages for Different Perpendicular AirSpeed Ranges: EV-2 at Low and High Speeds
Top Row: EV-2. 22.5 mph Bottom Row: EV-2, 45 mph
7-44
-------
7.5 Development of Flags to Qualify Processed Detailed Data
The RSD detailed data from some vehicle transits are sometimes contaminated with severe noise
or have low CO2 signal strength. Such events are common in all types of RSD instruments. In
such cases in the Westminster dataset, the noise reduction techniques described above are
insufficient to produce a transit's adjusted data that will result in proper blind source separation
and/or accurate reported emission rates. Just as for traditional RSD outputs, flags that can
identify problematic transits need to be developed for the new methodology to avoid including
such transits in a dataset that will be used to identify high-emitting vehicles or to characterize the
emissions of a fleet sample.
An example will serve to demonstrate how flags can be used to qualify transit data for inclusion
in a dataset. A set of 127 transits from the Westminster dataset were selected to preliminarily
evaluate performance of noise reduction and blind source separation. The transits were not
randomly selected but were selected in groups of light-duty diesels, old gasoline vehicles
operating on warm days, test vehicles, and transits that showed detailed data with various levels
of noise including some with severe noise.
Figure 7-21 shows the set of ten heatmaps for one example transit: Series=515 Transit=1131. In
each heatmap the vehicle is at the bottom of the panel where the color is uniform. The vehicle is
moving downward. The top five panels show the heatmap for each RSD channel after noise
reduction processing, that is, the heatmaps of the adjusted detailed data. Pollutants CO2, CO, NO,
and NO2 can only be exhaust pollutants and therefore, if there is any detectable pollutant present,
they should have similar heatmaps. While no NO2 is visible, CO2, CO, and NO have detectable
signals. The spatial patterns of the heatmaps for CO2 and NO look similar, but CO looks
different. In particular, the high-intensity (red) portion of CO is not at the same location as those
for CO2 and NO. Therefore, the CO heatmap is a concern.
The bottom five panels show the output of the blind source separation. A good separation would
show a heatmap for ExhPlume that is similar to the heatmap for CO2 and heatmaps for Noisel,
Noise2, and Noise3 that are just a field of random speckles. If substantial EvapHC is detected, it
would appear in EvapPlume, as shown in the bottom left panel. The problem is that while Noise2
and Noise3 are predominantly random speckles, Noisel shows a strong signal that looks like the
heatmap for CO. Evidently, BSS "thought" that the heatmaps for CO2 and CO were substantially
different and therefore assigned most of the CO signal to the Noisel heatmap.
One flag that we have begun to develop is designed to determine if the adjusted CO heatmap is
well correlated with the adjusted CO2 heatmap. More specifically, the question is: What is the
probability that the adjusted CO heatmap is the same as the adjusted CO2 heatmap? Based on the
visual examination of the heatmaps in Figure 7-21, we would say that it was a low probability,
but we want to quantify the probability so that we can remove the worst offending transits from
the analysis dataset.
We visually examined the ten heatmaps of the 127 transits in the selected sample. While it is
practical to visually examine the heatmaps of the 127-transit sample set, it is not practical to do
so for the entire 30,000-transit Westminster dataset. Therefore, we need to develop flags that can
be used to identify suspect transits automatically.
7-45
-------
Figure 7-21, Heatmaps for an Example Westminster Transit
7 20191025 000515 car 001131
HC C02 CO NO N02
10
10
J
10
10
11 ii ii>
t '11
20
20
20
20
¦ ¦
30
30
30
30
0 100 200 0 100 200 0 100 200 0 100 200 0 100 200
EvapPlume ExhPlume Noisel Noise2 NoiseS
10
20
30
10
10
10
IlllfP
ill
10
20
20
¦ 1
20
HRliI
11 till
20
i
30
In
30
HH f 1
in it
30
R$
r
V" I
III
30
11 ' I
0 100 200 0 100 200 0 100 200 0 100 200 0 100 200
For each of the 127 transits in the selected set, we visually examined the CO2 and CO heatmaps
and answered the questions: 1) Does the CO heatmap show a detectable signal? and 2) If there is
a detectable CO signal, is the CO heatmap correlated with the CO2 heatmap? Then, we used SAS
to calculate two statistics: 1) the probability that the pixel values used by the CO heatmap were
NOT normally distributed, and 2) the correlation coefficient, R, between the CO pixel values and
the CO2 pixel values.
The results of the exercise are shown in Figure 7-22. The x-axis gives the probability that the
pixel values used by the CO heatmap were NOT normally distributed. Fligh x-values indicate
that a signal was present in the CO heatmap; low x-values indicate that the CO values were
likely just noise. The y-axis gives the correlation between CO pixel values and CO2 pixel values.
High y-values near 1 indicate the pixel values are highly, positively correlated. Y-values near 0
indicate poor correlation.
The symbols in the figure indicate the results of the visual examination of the heatmaps. Blue
triangles indicate that a signal was observed in the CO heatmap and that it appeared to be
correlated with the CO2 heatmap. Black dots indicate that no or a very weak signal was seen in
the CO heatmap and therefore it was not possible to visually determine if a correlation with CO2
was present or not. Red dots indicate that a signal was seen in the CO heatmap and that it was
not correlated with the CO2 heatmap. The figure shows overlap between the "good" (blue and
black) transits from the "bad" (red) transits. The red dot at (0.25, 0.66) is the symbol for the
transit examined in Figure 7-21.
7-46
-------
Figure 7-22. Statistics for Adjusted CO and CO2 for the 127-Transit Sample Set
the normality test statistic, adj_CO_mole
eyecorr_CO • • • -1 ••• 0 1 3 4
/proj1/EDARinDenver-OCT2019/Analysis_MLout/220801 /Anal_MLout/OCT19_make_flags.sas 22AUG22 14:53
We built a model30 to approximate the trend through the black dots and blue triangles, which
were visually judged to be "good" transits with respect to CO and CO2 correlation. The modeling
dataset did not include any red dots and did not include the two black dots that are among the red
dots. All of those transits were judged as "bad." The resulting model is the black solid curve in
the figure. The model also provided an estimate of the standard deviation of the distribution of
measured y-values above and below the black curve. We drew the dashed black line at -1.645
standard deviations, which provides an estimate of the location of the one-tailed 5% probability
curve. That means that a CO heatmap that has its symbol below the dashed line has less than a
5% probability of being correlated with the CO2 heatmap.
The standard deviation can also be used to calculate the correlation probability for each transit.
For example, the probability that the CO heatmap of the transit examined in Figure 7-21 is
correlated with its CO2 heatmap is only 0.015%. These individual probabilities can be used to
choose the probability levels used to cull out suspect transits.
Development of flags is underway but not yet completed. Therefore, the fleet emissions
characteristics that are reported below have not had their underlying transit detailed datasets
screened by flags. Accordingly, the findings below must be regarded as preliminary.
M /proj l/EDARinDenver-OCT2019/Analysis_MLout/220801/Anal_MLout/OCT19_make_flags.sas
7-47
-------
8.0 Exhaust Concentrations Reported by EDAR
The primary goal of this study was to study on-road running loss emissions. However, EDAR
routinely reports exhaust emissions for each vehicle transit. In the subsections below, we use the
test vehicle data and the fleet data to characterize the exhaust emissions performance of the
EDAR instrument.
In Section 2.9, Table 2-5 showed that the EDAR QC flag assigned values of "valid," "interfering
plume," "low CO2," or "no plate" to each transit. Interfering plumes and small CO2 plumes are
likely to cause reported EDAR concentration values that have larger error than otherwise.
Therefore, for the analysis results reported below, transits with EDAR QC flag values of
"interfering plume" or "low CO2" were not used.
8.1 EDAR Exhaust Concentrations on Test Vehicles
As part of the study, the exhaust emissions of the test vehicles were reported by the EDAR
instrument. For most of the test runs, the EV-1, EV-2, F-150, GMC, and Subaru test vehicles
drove past the RSD while releasing real or simulated exhaust emissions and simulated running
loss emissions. But for some planned test runs, they drove past releasing only exhaust emissions.
In addition, the Infiniti, which never released simulated running loss emissions, drove past the
RSD for every convoy transit. We identified 425 individual test vehicle transits when no
artificial running losses were released (ref MeasuredReleaseRate = 0 g/hr), when the test
conditions were satisfied (refQualityFlag = G or Q), and when the EDAR RSD reported that the
result was valid (EDAR QC = valid). The reported exhaust emissions of those runs with date
and time are provided in Appendix A.
Histograms for the reported HC, CO, NO, and CO2 emissions are shown using uniform
concentration axes in Figures 8-1 to 8-4 for EV-1, EV-2, the Subaru, and the Infiniti. These four
vehicles had the largest number of measurements. Table 8-1 gives statistics describing the
eligible exhaust emissions values for all six test vehicles.
As described earlier, the EV-1 and EV-2 test vehicles released a puff of a stoichiometric blend of
dry, simulated exhaust gas mixtures just before each RSD transit. Because these simulated
exhaust gas blends came from a cylinder, the blend concentrations were the same for every
transit. On the other hand, the exhaust from the F-150, GMC, Subaru, and Infiniti was just their
usual, natural exhaust; they did not release artificial exhaust gas mixtures. Consequently, their
exhaust contained water of combustion. Also, emissions concentrations could possibly change
concentrations from run to run depending on engine and catalyst operation. The exhaust gas
concentrations were known for the EV-1 and EV-2 from the labels on the cylinders, but the
exhaust concentrations of the other four test vehicles were not measured except by the EDAR
instrument.
8-1
-------
Figure 8-1, EDAR Exhaust Concentration Measurements on EV-1
CUM.
CUM.
¦a FREC^
po%
po%
0
0
0.00
0.00
0
0
0.00
0.00
0
0
0.00
0.00
0
0
0.00
0.00
0
0
0,00
0.00
0
0
0.00
0.00
1
1
2.38
2.38
2
3
4.76
7.14
8
11
19.05
26.19
14
2b
33.33
59.^2
6
31
14.29
73.81
3
34
7.14
80.95
2
36
4, /6
85./1
1
37
2.38
88.10
3
40
7.14
95.24
0
40
0.00
95.24
0
40
0.00
95.24
0
40
0.00
95.24
0
40
0,00
95.24
1
41
2.38
97.62
0
41
0,00
9/,62
0
41
0.00
97.62
0
41
0,00
97.62
0
41
0,00
9/.62
0
41
0.00
97.62
1
42
2.38
100.00
0
42
0.00
100.00
0
42
0,00
100.00
0
42
0.00
100,00
0
42
0.00
100.00
0
42
0.00
100.00
0
42
0.00
100.00
9 10 11 12 13 14
CUM.
CUM.
FREQ.
0
FREQ.
0
PCT.
PCT.
-100
0.00
0.00
-50
3
3
7.14
7,14
0
37
40
88.10
95.24
50
¦
2
42
4.76
100.00
100
0
42
0.00
100,00
150
0
42
0.00
100.00
200
0
42
0.00
100.00
250
0
42
0.00
100.00
300
0
42
0.00
100.00
350
0
42
0.00
100.00
400
0
42
0.00
100.00
450
0
42
0.00
100.00
500
0
42
0.00
100.00
550
0
42
0.00
100.00
600
0
42
0.00
100.00
650
0
42
0,00
100.00
700
0
42
0,00
100.00
750
0
42
0.00
100.00
800
0
42
0.00
100.00
850
0
42
0.00
100.00
900
0
42
0.00
100.00
950
0
42
0.00
100.00
1000
0
42
0.00
100.00
1050
0
42
0.00
100.00
1100
0
42
0.00
100.00
¦J
200
0
HZ.
42
IUU
0
400
0
42
0
600
0
42
0
800
0
42
0
1000
0
42
0
1200
0
42
0
1400
0
42
0
1600
0
42
0
1800
0
42
0
2000
0
42
0
2200
0
42
0
2400
0
42
0
2600
0
42
0
2800
0
42
0
3000
0
42
0
3200
0
42
0
3400
0
42
0
3600
0
42
0
3800
0
42
0
4000
0
42
0
4200
0
42
0
4400
0
42
0
4600
0
42
0
4800
0
42
0
5000
0
42
0
5200
0
42
CUM.
0
hKLCfe
FRLCfc
PC I.
149000
o.od
1492C0
0
0
0.00
1494CC
0
0
0.00
1496C0
0
0
0.00
149800
0
0
0.00
150000
0
0
0.00
1502CC
0
0
0.00
-—-
150400
0
0
0.00
E
150600
0
0
0.00
CL
150800
0
0
0.00
a.
151000
0
0
0.00
rs
151200
0
0
0.00
O
1514CC
0
0
0.00
U
151600
0
0
0.00
1518CC
0
0
0.00
TS
152000
0
0
0.00
3
_C
152200
0
0
0.00
152400
0
0
0.00
X
152600
0
0
0.00
LU
152800
0
0
0.00
IX
153000
0
0
0.00
<
153200
0
0
0,00
a
1534CC
0
0
0.00
LU
153600
0
0
0.00
153800
0
0
0.00
154000
¦
1
1
2.38
154200
154400
¦
1
18
2
20
2 38
42.86
154600
154800
22
0
42
42
52.38
0.00
155000
0
42
0.00
PCI.
o.od
o.oo
o.oo
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
2.38
4.75
47.62
100.00
100.00
100.00
10
20
30
40
FREQUENCY
/prajl/E DARi n Denvef-OCT20'19,'Ana lys is.'ana lyze_RefVeh_Ex h_1. sas 24 APR20 11:05
8-2
-------
Figure 8-2. EDAR Exhaust Concentration Measurements on EV-2
CUM,
CUM.
:4> FRE4i
po%
po%
0
0
0.00
0.00
0
0
0.00
0.00
0
0
0.00
0.00
0
0
0.00
0.00
0
0
0.00
0.00
0
0
0.00
0.00
0
0
0.00
0.00
1
1
2.04
2.04
0
1
0.00
2.04
4
b
8.16
10.20
1
6
2.04
12,24
7
13
14.29
26.53
16
29
32,6b
59.18
6
35
12.24
71.43
7
42
14.29
85.71
2
44
4,08
89.80
1
45
204
91.84
3
48
6,12
9/.96
0
48
0,00
97.95
0
48
0.00
97.96
1
49
2.04
100.00
0
49
0.00
100.00
0
49
0.00
100.00
0
49
0.00
100.00
0
49
0.00
100.00
0
49
0.00
100.00
0
49
0.00
100.00
0
49
0.00
100.00
0
49
0.00
100.00
0
49
0.00
100.00
0
49
0.00
100.00
0
49
0.00
100.00
0 1 2 3 4 b 6 /
9 10 11 12 13 14 1b 16
CUM.
CUM.
FREQ.
0
FREQ.
0
PCT.
PCT.
-10C
0.00
0.00
-5C
0
0
0.00
0.00
c
0
0
0.00
0.00
5C
0
0
0.00
0.00
1DC
0
0
0.00
0.00
15C
0
0
0,00
0.00
2CC
0
0
0.00
0.00
25C
0
0
0.00
0.00
30C
0
0
0.00
0.00
35G
0
0
0.00
0.00
40C
0
0
0.00
0.00
450
0
0
0.00
0.00
50C
0
0
0.00
0.00
55G
1
1
2.04
2.04
60C
1
2
2.04
4.08
6SC
0
2
0,00
4.08
70C
0
2
0.00
4.08
75C
1
3
2.04
6.12
80C
3
6
6.12
12.24
85G
5
11
10.20
22.4b
900
16
2/
32.6b
bb.10
95C
9
36
18.37
73.47
100G
6
42
12.24
85.71
105C
7
49
14.29
100.00
hoc
0
49
0.00
100.00
CUM.
CUM.
:c%
FREC^
po%
po%
0
0
0.00
0.00
0
0
0.00
0.00
0
0
0.00
0.00
0
0
0.00
0.00
0
0
0,00
0.00
0
0
0.00
0.00
0
0
0.00
0.00
0
0
0.00
0.00
0
0
0.00
0.00
0
0
0.00
0.00
0
0
0.00
0.00
0
0
0.00
0.00
0
0
0.00
0.00
0
0
0.00
0.00
0
0
0.00
0.00
0
0
0.00
0,00
0
0
0.00
0.00
0
0
0.00
0.00
0
0
0.00
0.00
0
0
0.00
0.00
2
2
4.08
4.08
0
2
0.00
4.08
1
3
2.04
6.12
3
6
6.12
12.24
1
7
2.04
14.29
7
14
14.29
28.57
11
25
22.45
51.02
13
38
26.53
77.55
11
49
22.45
100.00
0
49
0.00
100.00
10 11 12 13
£
a.
Q.
CM
O
U
X
LU
<
Q
LU
CUM.
CUM.
FRECl I
FREQ.
PCT
PCT.
149000
V
T)
0.06
0.00
149200
0
0
0,00
0,00
149400
0
0
0.00
0.00
149600 |
1
1
2.04
2.04
149800 |
3
4
6.12
8.16
150000 |
12
16
24.49
32.65
150200 |
19
35
38.78
71.43
150400 |
8
43
16.33
S/./6
150500 |
1
44
2.04
89.80
150800 |
3
47
6.12
95.92
151000 |
1
48
2.04
97.96
151200 |
1
49
2.04
100.00
151400
0
49
0.00
100.00
151600
0
49
0.00
100.00
151800
0
49
0.00
100.00
152000
0
49
0.00
100.00
152200
0
49
0.00
100.00
152400
0
49
0.00
100.00
152500
0
49
0.00
100.00
152800
0
49
0.00
100.00
153000
0
49
0.00
100.00
153200
0
49
0.00
100.00
153400
0
49
0.00
100.00
153500
0
49
0.00
100.00
153800
0
49
0.00
100.00
154000
0
49
0.00
100.00
154200
0
49
0.00
100.00
154400
0
49
0.00
100.00
154500
0
49
0.00
100.00
154800
0
49
0.00
100.00
155000
0
49
0.00
100.00
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
6 8 10 12
FREQUENCY
.''proj1.,'EDA:;linDenver-OCT2019.>'Analysis.,analyze_RefVeh_Exh_1-sas 24APR20 11:05
8-3
-------
Figure 8-3. EDAR Exhaust Concentration Measurements on Subaru
CUM.
CUM.
-5CG
FRCS)
FRD^
pcSo
po%
-45 c
1
1
4.55
4.hh
-4CC
0
1
0.00
+.55
-35 C
0
1
0.00
4.55
-3CC
0
1
0.00
4,»
-25 C
0
1
0.00
4.55
-2CC
0
1
0,00
4.55
to
-15G
1
2
4.55
9.09
-ice
1
3
4,55
13.64
E
¦5C
3
6
13.64
27.27
22
72.73
100.00
Q.
5C
0
22
0.00
100.00
Q.
ICC
0
22
0.00
100.00
u
15C
0
22
0,00
100.00
2CC
0
22
0.00
100.00
X
25 C
0
22
0.00
100-00
in
3CC
0
22
0.00
100.00
35 C
0
22
0.00
100.00
ra
4CC
0
22
0.00
100.00
X
LU
45 C
0
22
0.00
100.00
5CC
0
22
0,00
100.00
55 C
0
22
0.00
100.00
a
6CC
0
22
0.00
100.00
<
65 C
0
22
0,00
100.00
Q
7CC
0
22
0.00
100.00
LU
75 C
0
22
0.00
100.00
8CC
0
22
0.00
100.00
85 C
0
22
0.00
100.00
9CG
0
22
0,00
100.00
95 C
0
22
0.00
100.00
10GG
0
22
0.00
100.00
105C
0
22
0.00
100.00
11CC
0
22
0.00
100.00
0 1 2 3 4 5 6 7
9 10 11 12 13 14 15 16
-5C
2
2
9.09
c
11
13
50.00
c
5C
5
18
22.73
£
IOC
3
21
13.64
c
15C
0
21
0.00
c
20G
0
21
0.00
c
25C
0
21
0.00
c
30C
0
21
0,00
c
35C
0
21
0.00
c
40C
1
22
4.55
1C
45G
0
22
0.00
1C
.500
0
22
0.00
1C
55C
0
22
0.00
1C
600
0
22
0.00
1C
65C
0
22
0.00
1C
70C
0
22
0.00
1C
75C
0
22
0.00
1C
80C
0
22
0.00
1C
85C
0
22
0.00
1C
90C
0
22
0.00
1C
95C
0
22
0.00
1C
1 nnr
n
y>
n nn
1 r
E
Q.
Q.
O
U
-------
Figure 8-4. EDAR Exhaust Concentration Measurements on Infiniti
CUM.
CUM.
-5CC
FRni
FRC4>
poS7o
po%
-45 C
0
0
0.00
0.00
-4CC
0
0
0.00
0.00
-35 G
0
0
0.00
0.00
-3CC 1
1
1
0.34
0,34
-25 C
0
1
0.00
0.34
-2CC 1
1
2
0.34
0.69
-15C 1
2
4
0.69
1.38
y?
r i
-1CC I
4
8
1.38
2./6
w
E
-5C HI
14
22
4.83
7.59
111
87.93
95.52
CL
5C 1
4
281
1.38
96,90
CL
ICC I
3
284
1.03
97.93
U
T
15C
0
284
0.00
97.93
2CC I
1
285
0.34
98.28
J_
25 C
0
285
0.00
98.28
t/J
3CC 1
1
285
0.34
98.62
3
35 C
0
286
0.00
98.62
ro
4CC 1
1
28/
0.34
98.9/
£.
45 C 1
1
288
0.34
99.31
X
LU
5CC 1
1
28Q
0.34
99.66
55 C
0
289
0.00
99.66
cc
6CC
0
289
0.00
99.66
<
65 C
0
289
0.00
99,66
D
7CC
0
289
0.00
99.66
LU
75 C
0
289
0.00
99.66
8CC
0
289
0.00
99.66
85 C
0
289
0.00
99.66
9CC
0
289
0.00
99.66
95 C
0
289
0.00
99.66
10CC
0
289
0.00
99.66
105C 1
1
290
0.34
100,00
11CC
0
290
0.00
100.00
100
200
300
E
Q.
o.
-C
X
UJ
5
G
ID
CUM.
CUM.
FRED
Fn
PCT.
PCT.
-10C
0
0.00
0.00
-5C
1
1
0.34
0.34
C
262
90.00
90.34
5C
^1
24
286
8.28
98.62
10C
3
289
1.03
99.66
15C
0
289
0.00
99.66
20C
0
289
0,00
99.66
25C
1
290
0.34
100.00
30C
0
290
0.00
100.00
35C
0
290
0.00
100.00
40C
0
290
0.00
100.00
45C
0
290
0.00
100.00
50C
0
290
0.00
100.00
55C
0
290
0.00
100.00
60C
0
290
0.00
100.00
65C
0
290
0.00
100.00
70C
0
290
0.00
100.00
75C
0
290
0.00
100.00
80C
0
290
0.00
100.00
85C
0
290
0.00
100.00
90C
0
290
0.00
100.00
95C
0
290
0.00
100.00
100C
0
290
0.00
100.00
105C
0
290
0.00
100.00
110C
0
290
0.00
100.00
E
a.
3
fN
o
u
1a
3
03
.C
X
LU
Sf
Q
LU
100
149000 ¦
149200
149400
149500
149800
150000
150200
150400
150500
150500 |
151000
151200 |
151400
151600
151800
152000
152200
152400
152600 |
152800
153000 |
153200 |
153400 ¦
153500 ¦
153800 m
154000 |
154200 ¦
154400 ¦
154600 ¦
154800
155000
0
20
40
60 80 100 120
FREQUENCY
140
CUM.
CUM.
E%
FRES)
PCX
O.O0
PCT.
0.00
38
38
13.10
13.10
136
174
46.90
60.00
55
229
18.97
78.97
23
252
/.93
86.90
14
266
4.83
91.72
9
275
3.10
94.83
3
2/8
1.03
95.86
3
281
1.03
96.90
0
281
0.00
96.90
1
282
0.34
97.24
2
284
0.69
9/.93
1
285
0.34
98.28
1
286
0.34
98.62
0
286
0.00
98.62
0
286
0.00
98.62
0
286
0.00
98.62
0
286
0.00
98.62
0
286
0.00
98.62
0
286
0.00
98.62
0
286
0.00
98.62
0
286
0.00
98.62
0
286
0.00
98.62
0
286
0,00
98.62
0
286
0.00
98.62
0
286
0.00
98.62
1
287
0.34
98.97
0
28/
0.00
98.9/
0
287
0.00
98.97
0
287
0.00
98.97
3
290
1.03
100.00
¦ FRES
0
0
0
0
0
0
0
0
1
0
1
0
0
0
0
0
0
1
0
2
2
5
4
14
30
61
154
12
0
0
CUM.
FREq.
3
3
3
3
3
3
3
3
4
4
6
8
10
15
19
33
63
124
278
290
290
290
160
PCT,
1.03
0.00
0,00
0.00
0.00
0.00
0.00
0.00
0.00
0.34
0.00
0.34
0.00
0.00
0.00
0.00
0.00
0.00
0,34
0.00
0,69
0.69
1.72
1.38
4.83
10.34
21.03
53,10
4.14
0.00
0.00
CUM.
PCT,
1.03
1.03
1.03
1,03
1.03
1.03
1.03
1.03
1.03
1.38
1.38
1.72
1.72
1.72
1.72
1.72
1./2
1.72
2,0/
2.07
2.76
3.45
5.17
6.55
11.38
21./2
42.76
95.86
1 00.00
1 00.00
100.00
.•'proj l."EDA:liiOenver-0'CT2019.',Analysis'analyze_RefVeh_Exh_l-5as 24APR20 11:05
8-5
-------
Table 8-1. Reported Exhaust Concentrations for the Test Vehicles
a)
EDAR Reported Exhaust HC (ppmC6)
Reference
Vehicle ID
Label HC
(ppmC3)
N
Percentiles
Mean
Std
Dev
Min.
Max.
16th
25th
50th
75th
84th
1-EV1
0
42
-53
-29
2
78
164
55
163
-135
791
2-EV2
402
49
80
123
169
233
271
178
115
-75
554
3-F150
n/a
18
-69
-19
0
2
6
-20
44
-160
17
3-GMC
n/a
4
-1
-1
0
1
1
0
1
-1
1
3-Subaru
n/a
22
-73
-27
-9
4
6
-43
102
-455
15
4-Infiniti
n/a
290
-13
-7
-2
0
1
>¦>
3
86
-300
1047
b)
EDAR Reported Exhaust CO (ppm)
Reference
Vehicle ID
Label CO
(ppm)
N
Percentiles
Mean
Std
Dev
Min.
Max.
16th
25th
50th
75th
84th
1-EVI
()
42
-8
-5
1
7
25
5
22
-48
80
2-EV2
5()43
49
5030
5076
5294
5460
5551
5210
385
4053
5699
3-F150
n/a
18
6
11
44
109
137
80
107
-6
421
3-GMC
n/a
4
58
94
130
345
558
219
229
58
558
3-Subaru
n/a
22
22
79
336
551
608
425
555
-7
2651
4-Infiniti
n/a
290
109
143
251
433
595
778
4071
-91
47348
c)
EDAR Reported Exhaust NO (ppm)
Reference
Vehicle ID
Label NO
(ppm)
N
Percentiles
Mean
Std
Dev
Min.
Max.
16th
25th
50th
75th
84th
1-EVI
0
42
-10
-2
1
7
12
0
18
-57
49
2-EV2
996
49
861
881
922
977
1023
917
101
531
1060
3-F150
n/a
18
7
12
22
42
54
29
23
-12
83
3-GMC
n/a
4
-2
0
•->
3
7
9
4
5
-2
9
3-Subaru
n/a
22
-2
3
17
52
80
42
91
-43
416
4-Infiniti
n/a
290
1
3
6
14
18
10
19
-26
226
d)
EDAR Reported Exhaust CO2 (ppm)
Reference
Vehicle ID
Label C02
(ppm)
N
Percentiles
Mean
Std
Dev
Min.
Max.
16th
25th
50th
75th
84th
1-EVI
15< >5(»>
42
154395
154427
154509
154534
154543
154471
101
154009
154555
2-EV2
I47WIO
49
149998
150051
150215
150319
150425
150234
296
149530
151155
3-F150
n/a
18
154328
154405
154468
154509
154527
154436
98
154166
154532
3-GMC
n/a
4
154136
154295
154454
154472
154489
154383
166
154136
154489
3-Subaru
n/a
22
154067
154112
154245
154431
154486
154177
445
152368
154532
4-Infiniti
n/a
290
154008
154159
154334
154414
154447
153921
3001
119640
154545
8-6
-------
The distributions of the reported exhaust emissions concentrations for EV-1 are shown in Figure
8-1. The distributions for CO (blue), NO (red), and CO2 (green) are relatively tight, but the
distribution for HC (black) has two elevated values that appear to be outliers. Appendix A
indicates that these values, which are shaded with yellow backgrounds, are 498 and 791 ppmC6.
The CO, NO, and CO2 values associated with these two transits do not appear to be outliers.
Table 8-1 shows statistics for the 42 EV-1 transits in the first data row of sub-tables a), b), c),
and d). Comparison of the 50 percentile (median) values (light blue) with the mean values (light
green) provides an indication of the influence of outlier values. For HC, the mean of 55 ppmC6
is somewhat higher than the median 2 ppmC6 value. This difference is presumably caused by the
two high HC values increasing the mean. On the other hand, median and mean values are
comparable for EV-l's CO, NO, and CO2: 1 vs. 5ppm, 1 vs. 0 ppm, and 154,509 vs. 154,471
ppm.
Table 8-1 also shows the comparisons of the statistics with the labeled concentrations for the
cylinders used for EV-1 and EV-2 for the artificial dry exhaust gas mixtures. EV-1 was releasing
a "clean" artificial exhaust gas mixture that had only 15.05 vol% CO2 with balance nitrogen.
Table 8-1 shows percentiles and standard deviations to judge variability. Like means, standard
deviations can be more susceptible to outliers than percentiles - at least for percentiles that are
not near the extremes of distributions. For a normal distribution, the -1 standard deviation point
and the +1 standard deviation points are at approximately the 16 and 84 percentile values.
Table 8-2 helps to focus on the EDAR performance for measuring the "clean" artificial exhaust
gas mixture released from EV-1. The table uses only the medians to describe trends since
medians are less susceptible to extreme measured values. The EDAR medians in the third
column are very close to the cylinder values in the second column. The 95% confidence limits of
the medians are shown in the fourth and fifth columns. The confidence limits are quite close to
the medians for CO, NO, and CO2 and are larger for HC. The CO2 deviations, expressed as
percents with respect to (wrt) the CO2 median, are given in the sixth and seventh columns. The
last column shows that EDAR was reporting CO2 about 2.7% higher than the label concentration
on the gas cylinder.
Table 8-2. Dry, Artificial Exhaust Zero Performance by EV-1 Test Vehicle31
Pollutant
Concentration
(ppm vol)
Confidence Limits
on Median (ppm vol)
Deviation
(A% wrt Median)
Accuracy
(% wrt Cylinder)
Cylinder
Value
EDAR
Median
Lower
95%CL
Upper
95%CL
Lower
95%CL
Upper
95%CL
Median
Cylinder
HC
0
2
-9
40
n/a
n/a
n/a
CO
0
1
-2
3
n/a
n/a
n/a
NO
0
1
-1
3
n/a
n/a
n/a
C02
150500
154509
154469
154523
-0.03%
+0.01%
102.7%
31 P:\EDARinDenver-OCT2019\Analysis/refVeh_out.xlsx
8-7
-------
Test vehicle EV-2 released an artificial, dry exhaust gas mixture that can be called "dirty" since
it had cylinder label concentrations of 402 ppmC3 HC, 5043 ppm CO, 996 ppm NO, and
147,600 ppm CO2, as shown in the second column of Table 8-1. Thus, while exhaust mixture
releases from EV-1 can be used to evaluate EDAR performance at zero, the exhaust mixture
releases from EV-2 can be used to evaluate EDAR performance at high values - essentially a
span evaluation.
Note that EDAR reports HC in units of ppmC6, that is, ppm on a hexane basis. Since the exhaust
gas cylinders used propane as the HC gas, the cylinder HC concentrations are in units of ppmC3,
that is, ppm on a propane basis. The conversion factor between ppmC6 and ppmC3 is roughly a
factor of 2. For example, 402 ppmC3 ~ 201 ppmC6. That means that when evaluating EDAR HC
performance for measuring the dirty mixture, we should compare EDAR HC reported values
against the 201 ppmC6 value.
Figure 8-2 shows histograms for the 49 valid test runs of EV-2 when artificial exhaust was
released but artificial running losses were not released. The following extreme values, although
not necessarily outliers, are observed in the histograms and are shaded in yellow in Appendix A:
-75 ppmC6 HC, 554 ppmC6 HC, 4070 ppm CO, 531 ppm NO, and 624 ppm NO. The appendix
shows that the 4070 ppm CO and 531 ppm NO occurred on the same transit. Similarly, the 554
ppmC6 HC and 624 ppm NO occurred on a different transit. These associated extreme reported
values suggest that some sort of noise threw off the calculations of the reported values for these
transits.
The presence of extreme reported values has an influence on evaluating the EDAR "span"
performance using mean and standard deviation, which are shown in Table 8-1. The median
values for EV-2 are condensed in Table 8-3. Table 8-3 shows that for the exhaust span mixture,
the relative deviations (columns six and seven), expressed as a percent with respect to the
median, were low for CO, NO, and CO2, but were high for HC. The last column shows the
accuracy, as measured by the median, relative to the cylinder concentrations. HC was 16% low,
CO was 5% high, NO was 7% low, and CO2 was 2 % high.
Table 8-3. Dry, Artificial Exhaust Span Performance by EV-2 Test Vehicle32
Pollutant
Concentration
(ppm vol)
Confidence Limits
on Median (ppm vol)
Deviation
(A% wrt Median)
Accuracy
(% wrt
Cylinder)
Cylinder
Value
EDAR
Median
Lower
95%CL
Upper
95%CL
Lower
95%CL
Upper
95%CL
Median
Cylinder
HC
201
169
148
185
-12.6%
+ 9.2%
84.3%
CO
5043
5294
5210
5399
-1.6%
+ 2.0%
105.0%
NO
996
922
898
941
-2.6%
+ 2.1%
92.6%
CO2
147600
150215
150119
150271
-0.06%
+0.04%
101.8%
The four non-electric test vehicles (F-150, GMC, Subaru, and Infiniti) did not release artificial
exhaust gas mixtures. Instead, they released their natural tailpipe exhaust at their usual flow
32 P:\EDARinDenver-OCT2019\Analysis/refVeh_out.xlsx
-------
rates. Also, their exhaust compositions were natural, which means they included water of
combustion and a wide variety of hydrocarbon compounds. Depending on the methods used by
EDAR, these differences could affect the measurements. These potential influences were the
reason that gasoline vehicles were used as part of the test vehicle convoy.
For NO, Table 8-lc shows that for the four non-electric test vehicles, EDAR reported NO mean
and median values that were somewhat (2 to 42 ppm) higher than those reported for EV-1. Such
NO levels can be expected from properly operating current technology vehicles. The 16th and
25th percentile NO values for the non-electric test vehicles were comparable to those for EV-1.
However, for the 75th and 84th percentiles, the NO values tended to be higher than for EV-1. This
trend could be the result of real changes in NO concentrations of the non-EVs as their real
engines and catalyst systems operated.
For HC, Table 8-la shows that the four non-EVs had reported median HC values near 0 ppmC6
and near the 2 ppmC6 median value reported for EV-1. In addition, the non-EV's reported HC
values for the upper and lower percentiles seemed to be substantially tighter (i.e., closer to the
median) than for the EV-1. The tailpipe emissions of gasoline vehicles is a mixture of many HC
compounds, contains water of combustion, and is emitted at flow rates generally larger than
those used on the EVs. Since the EDAR instrument is intended to measure the emissions of real
vehicles, and not artificial exhaust emissions, it is possible that EDAR performance on real
exhaust may be superior to performance on simulated exhaust.
For CO, the EV-1 statistics in Table 8-lb show that EDAR reports CO values with low bias and
good repeatability when challenged with an artificial zero gas. For the non-EVs, Table 8-lc
shows that the non-EVs apparently have CO concentrations somewhat above zero. Additionally,
the variability in the reported median (or mean) values is monotonically increasing with the
median (or mean) values. There are two indistinguishable contributions to this trend in
variability: 1) actual vehicles with higher emissions will tend to have higher emissions
variability, and 2) measured values will tend to have higher variability due to noise in the
measuring process.
Appendix A shows that the Infiniti occasionally had high reported CO values. Of the 290 eligible
transits, the highest four values, which are shaded in yellow, were 5098, 34380, 38009, and
47348 ppm CO. We have no reason to not believe these values and suggest that they may have
occurred because of enrichment while accelerating in traffic.
8.2 EDAR Exhaust Concentrations on Fleet Vehicles
We can also evaluate EDAR's exhaust emissions measurement performance by examining the
reported emissions of fleet vehicles as they drove under the instrument. Of the 33,636 transits
recorded during the testing, 21,398 transits were determined to be from 13,480 private gasoline
vehicles with recognizable Colorado plates, decodable VINs, and with EDAR QC = "valid".
Model years ranged from 1947 through 2020. For calculating emissions statistics, the 1947
through 1989 model year vehicles were combined into a <1989 model year group.
The gray dot symbols in Figures 8-5, 8-6, and 8-7 show EDAR's measured values for HC, CO,
and NO for the 21,398 transits as a function of model year. The plots have constrained upper and
lower y-axis ranges so that some detail in model-year trends can be seen. Consequently, several
8-9
-------
emissions values are "off scale." All three plots show some negative emissions measurements.
Negative measurements can occur because of random noise in the instrument's underlying
optical measurements. Substantially more negative HC values are reported than CO and NO
values. That trend is in agreement with the test vehicle results, which were described in the
previous subsection.
The lines in the plots denote the model-year trends of the following percentiles: 5 (dark blue), 10
(medium blue), 20 (light blue), 50 (black), 80 (orange), 90 (red), and 95 (dark red). All three
figures show a generally downward trend with movement toward newer model years. Figure 8-5
shows the HC downward trend from 1990 to 2004. For model years newer than 2004, no further
HC decrease can be seen. For CO and NO in Figures 8-6 and 8-7, the percentile lines show the
downward trend throughout the entire model year range.
Table 8-4 gives statistics by model year for the HC, CO, and NO emissions measurements
reported by the EDAR instrument. For each pollutant, the table gives the mean and its 95%
confidence limits (tan background) and the median and its 95% confidence limits (green
background). Both mean and median are useful measures of the central tendency of each
distribution. Because the emissions distribution within each model year is positively skewed,
there is a strong tendency for the mean emissions value to be larger than the median emissions
value. For example, for the 96 mean-median pairs in Table 8-4, the mean is larger than the
median 93 times. However, as discussed in the previous subsection, the mean and its confidence
limits are quite susceptible to outliers, while the median and its confidence limits are less
susceptible to outliers. There is also a strong tendency for the 95% confidence interval for the
median to be narrower (and usually much narrower) than the 95% confidence interval for the
mean. For example, for the 96 mean-median pairs of intervals in Table 8-4, the interval for the
mean is larger than the interval for the median 90 times.
8-10
-------
Figure 8-5. Model Year Distribution of Fleet HC Concentration Measurements
1000 H
900
800
1
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
9
9
9
9
9
9
9
9
9
9
9
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
8
9
9
9
9
9
9
9
9
9
9
0
0
0
0
0
0
0
0
0
0
1
1
1
1
1
1
1
1
1
1
2
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
Model Year (dithered)
/proj1/EDARinDenver-OCT2019/Analysis/analyze_Fleet_Exh_1.sas 04MAY20 16:46
Percentile: 5 10 20 50 80 90 95
8-11
-------
Figure 8-6. Model Year Distribution of Fleet CO Concentration Measurements
5000-1
4800
4600
•
*•
•s . •
>r
• •
* "
% •
*'•"* f
•r-." -*
1
1
I
1
I
1
I
1
I
1
i
1
i
1
i
1
I
1
I
1
I
1
1
2
1
2
1
2
1
2
1
2
2
1
2
i
2
1
2
1
2
1
2
1
2
1 ' i
2 2
i
2
1
2
1 1 1
2 2
i 1 i
2 2 2
9
9
9
9
9
9
9
9
9
9
9
0
0
0
0
0
0
0
0
0
0
0
0
0 0
0
0
0 0
0 0 0
8
9
9
9
9
9
9
9
9
9
9
0
0
0
0
0
0
0
0
0
0
1
1
1 1
1
1
1 1
1 1 2
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2 3
4
5
6 7
8 9 0
Model Year (dithered)
/proj1/EDARinDenver-OCT2019/Analysis/analyze_Fleet_Exh_1.sas 04MAY20 16:46
Percentile: 5 10 20 50 80 90 95
8-12
-------
Figure 8-7. Model Year Distribution of Fleet NO Concentration Measurements
2000-
1900-
1800
1700-
1600-
1500-
E
1400
Q.
1300-
Q.
1200-
o
1100-
¦*->
1000
V)
D
900-
re
800-
X
LU
700-
cC
600
<
500-
n
400-
LU
300
200
100-
0-
-100
-200
1
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
9
9
9
9
9
9
9
9
9
9
9
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
8
9
9
9
9
9
9
9
9
9
9
0
0
0
0
0
0
0
0
0
0
1
1
1
1
1
1
1
1
1
1
2
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
Model Year (dithered)
/proj1/EDARinDenver-OCT2019/Analysis/analyze_Fleet_Exh_1.sas 04MAY20 16:46
Percentile: 5 10 20 50 80 90 95
8-13
-------
Table 8-4. Fleet Vehicle Model-Year-Mean and Median Concentration Measurements and Confidence Intervals
HC (ppmC6)
CO
ppm)
NO
[ppm)
Model
Year
Count
Lower
95% CL
Mean
Upper
95% CL
Lower
95% CL
Median
Upper
95% CL
Lower
95%CL
Mean
Upper
95%CL
Lower
95% CL
Median
Upper
95% CL
Lower
95% CL
Mean
Upper
95% CL
Lower
95% CL
Median
Upper
95% CL
<1989
32
52
158
264
15
83
161
3922
10812
17702
1151
3170
8871
688
1031
1374
368
662
1671
1990
15
-30
366
762
0
20
167
1368
8660
15952
320
2516
12157
502
1206
1911
177
993
1485
1991
21
-56
214
484
4
16
70
888
1921
2953
481
847
2095
314
619
924
162
279
1143
1992
18
8
138
268
14
39
91
388
2220
4052
86
643
2746
1126
1568
2011
797
1702
2339
1993
24
-24
280
584
5
28
110
2758
6908
11057
896
2067
6823
660
1135
1609
242
906
1687
1994
81
58
90
122
19
28
42
2769
4505
6241
449
1418
3177
569
773
977
188
413
786
1995
77
30
59
87
7
12
30
1603
3262
4920
571
744
1272
566
751
936
244
427
711
1996
85
-17
58
134
8
14
30
1724
2980
4236
484
660
1511
354
536
718
105
214
388
1997
151
20
32
43
12
16
23
1779
2550
3321
624
1031
1746
480
616
752
128
205
367
1998
195
28
50
73
3
8
12
1790
2874
3958
312
461
782
408
505
603
118
155
240
1999
250
25
46
67
6
10
16
1344
2373
3401
382
545
813
313
387
462
77
105
168
2000
360
7
25
43
2
3
5
1399
1918
2437
321
421
630
299
369
439
67
83
114
2001
390
8
24
41
1
2
3
1242
1840
2438
184
248
338
166
210
254
40
45
59
2002
483
-2
9
20
0
1
2
907
1160
1414
217
267
338
170
215
259
27
35
44
2003
563
10
20
29
0
1
2
891
1159
1427
180
223
273
140
177
214
25
32
40
2004
723
-15
49
114
0
0
1
766
1148
1530
144
177
209
137
170
203
18
23
28
2005
764
-4
3
10
0
0
0
758
1084
1410
144
169
209
93
115
137
17
20
23
2006
816
1
7
13
1
1
1
753
937
1120
164
203
234
69
92
114
16
19
21
2007
980
-4
6
16
0
0
0
637
784
930
146
161
183
64
81
97
13
15
18
2008
1013
-2
6
15
0
0
0
721
861
1002
156
180
216
58
74
91
15
16
18
2009
628
-3
10
23
0
0
0
842
1089
1336
143
176
214
62
84
105
13
16
20
2010
862
-5
2
8
-1
0
0
553
718
883
134
156
178
40
51
62
11
13
15
2011
1104
-3
3
9
0
0
0
614
731
848
143
162
180
45
61
76
12
14
15
2012
1168
-41
42
125
0
0
0
666
811
955
139
153
167
32
39
47
11
12
14
2013
1380
-8
-2
4
0
0
0
697
795
893
157
179
203
47
60
74
12
13
15
2014
1551
-19
26
72
0
0
0
692
810
928
121
138
156
40
49
58
13
14
15
2015
2235
-5
0
4
-1
0
0
553
674
795
134
147
160
31
37
43
9
10
11
2016
1749
-3
6
14
0
0
0
426
514
602
96
109
120
28
34
40
10
11
13
2017
1585
-4
4
12
-1
-1
0
424
514
605
84
94
106
27
34
41
9
10
11
2018
1418
-11
-8
-4
-1
-1
0
315
404
493
72
83
92
23
31
40
8
9
10
2019
672
-4
7
19
-2
-1
-1
274
369
464
81
92
107
15
18
22
8
9
11
2020
5
-10
8
27
0
1
34
-137
295
728
2
247
872
-105
45
196
-29
-1
261
8-14
-------
Figure 8-8. Model Year Distribution of Fleet Mean [HC] Measurements
99999999999000000000000000000000
8999999999900000000001 1 1 1 1 11 1112
90123456789012345678901234567890
Model Year
Legend: 95% Confidence Interval Mean
Figure 8-9. Model Year Distribution of Fleet Median [HC] Measurements
170
160
150
140
130
120
110
100
90
80
70
60
50
40
30
20
10
0
-10
Model Year
/proj1/E DARinDenver-OCT2019/Analysis/analyze_Fleet_Exh_1 .sas 01MAY20 13:58
Legend: 95% Confidence Interval ^—Median
8-15
-------
Figure 8-10. Model Year Distribution of Fleet Mean [CO] Measurements
19000
18000
17000
16000
E
15000
Q.
14000
LL
13000
o
u
12000
*¦»
I/)
11000
3
ro
10000
_c
X
9000
LU
rf
8000
<
G
7000
LU
6000
c
ro
5000
o>
2
4000
3000
2000
1000
0
Legend:
Model Year
¦- 95% Confidence Interval
Mean
Figure 8-11. Model Year Distribution of Fleet Median [CO] Measurements
E
Q.
O
u
*-»
ID
3
ro
-C
X
LU
Ct
<
G
LU
C
re
'•&
Q>
1
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
9
9
9
9
9
9
9
9
9
9
9
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
8
9
9
9
9
9
9
9
9
9
9
0
0
0
0
0
0
0
0
0
0
1
1
1
1
1
1
1
1
1
1
2
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
Model Year
/proj1/EDARinDenver-OCT2019/Analysis/analyze_Fleet_Exh_1 .sas 01MAY20 13:58
Legend: 95% Confidence Interval ^—Median
8-16
-------
Figure 8-12. Model Year Distribution of Fleet Mean [NO] Measurements
E
Q.
3
o
z
+->
I/)
3
ro
_c
X
LU
<
~
LU
C
to
cu
2500
2400
2300
2200
2100
2000
1900
1800
1700
1600
1500
1400
1300
1200
1100
1000
900
800
700
600
500
400
300
200
100
0
-100
Model Year
Legend:
¦- 95% Confidence Interval
Mean
Figure 8-13. Model Year Distribution of Fleet Median [NO] Measurements
2500
2400
2300
2200
2100
2000
q. 1900
3 1800
q 1700
Z 1600
1500
l/l
3 1400
_[2 1300
X 1200
111 1100
Op 1000
Q 900
LU 800
C 700
.5 600
"S 500
2 400
300
200
100
0
-100
99999999999000000000000000000000
8999999999900000000001 1 1 1 1 1 1 11 12
901 2345678901 2345678901 234567890
Model Year
/proj1/E DARinDenver-OCT2019/Analysis/analyze_Fleet_Exh_1 .sas 01MAY20 13:58
Legend: 95% Confidence Interval Median
8-17
-------
In addition to Table 8-4, Figures 8-8 through 8-13 show the model-year trends of the mean and
median HC, CO, and NO values. The dashed green lines in the mean plots give the lower and
upper 95% confidence limits on the mean, and the dashed red lines in the median plots give the
lower and upper 95% confidence limits on the median.
The Table 8-4 median values for HC from 2002 through 2020 are noteworthy. The medians are
all within 1 ppm of zero for those model years. The volatility in the corresponding HC means
makes it difficult to discern a downward trend from 2002 to 2020. For CO and NO, Table 8-4
and the figures show distinct downward trends in means and medians throughout the 2000+
model-year period.
8-18
-------
9.0 Performance of the Emissions Rate Measurement Methodology
In this section, the methodology's performance is evaluated by comparing the method's
calculated Release Rates (g/hr) for the test vehicles with their metered rates. The comparisons
can be made only in those cases where the pollutant releases from vehicles are known because
the pollutants were metered. Table 9-1 shows the corresponding cases.
Table 9-1. Combinations of Test Vehicles and Pollutant Types33
Test
Vehicle
Adjusted
CO
Adjusted
NO
Zeroed
C02
Adjusted
THC
EvapHC
ExhHC
EV-1
Sec 9.1
Sec 9.1
Sec 9.1
Sec 9.2
Sec 9.2
Sec 9.2
EV-2
Sec 9.1
Sec 9.1
Sec 9.1
Sec 9.2
Sec 9.2
Sec 9.2
F150
*
*
*
*
Sec 9.3
*
GMC
*
*
*
*
Sec 9.3
*
Subaru
*
*
*
*
Sec 9.3
*
Infiniti
*
*
*
*
*
*
* Emissions were not metered for these tests.
For EV-1 and EV-2, all exhaust emissions and evaporative emissions were produced by releases
from gas cylinders or tanks. These two all-electric test vehicles had no evaporative emissions of
their own except for the presumably trivial emissions from vehicle construction materials. The
measurement performance of exhaust CO, NO, and CO2 can be evaluated using data only from
EV-1 and EV-2. That analysis is presented in Section 9.1. The measurement performance for
EV-1 and EV-2 of Total HC (i.e., before BSS separation), and EvapHC and ExhHC (i.e., after
BSS separation) is presented in Section 9.2.
The F150, GMC, and Subaru test vehicles can be used to evaluate measurement performance of
EvapHC in those cases where the metered EvapHC release rates are believed to be substantially
larger than the vehicles' natural EvapHC and ExhHC release rates. That analysis is presented in
Section 9.3. The measurements of exhaust CO, NO, and CO2 from these test vehicles cannot be
used since those pollutants were not metered on those vehicles. Finally, no emissions
measurements for the Infiniti test vehicle can be used since neither exhaust nor evaporative
emissions were metered for that vehicle.
For the evaluations below, the measured EDAR detailed data collected in Westminster in
October 2019 was noise-reduced using the methods described in Section 7.1 to produce so-called
adjusted arrays (adjTHC, adjCO, adjNO, adjN02, adjC02) for each EDAR channel and each
transit. All five adjusted arrays were then used by BSS (standard FastICA) as described in
Section 7.2 to produce relative arrays assigned to EvapPlume and ExhPlume plus the three Noise
arrays. EvapHC and ExhHC arrays were determined by weighted regression of the adjTHC array
against the EvapPlume and ExhPlume arrays. The weights were the ScanSum weights described
in Section 6.5. Then, the baseline of adjC02 was additionally adjusted in SAS, as described in
33 P:\EDARinDenver-OCT2019\Analysis_MLout\220113\Anal_MLout/
OCT 19 VSPbins 2 EV RefVeh.xlsx
9-1
-------
Section 6.5, to produce ZeroedC02. The Vortex Entrainment Time (VET) values for the tailpipe
location described in Section 6.4 were used to convert the RSD-measured vortex masses to
release rates. Using tailpipe VETs is appropriate for exhaust emissions when the tailpipe exit is
near the vehicle rear. In the calculations below, rear-vehicle VETs were used for all tests - even
those where releases were from farther forward on the vehicle. This approach allows the
deviations between RSD-measured release rates and metered release rates to be evaluated in
Section 9.2 in terms of the known release locations of EvapHC on the test vehicles.
9.1 Exhaust CO, NO, and CO2 Release Rates from Reference EVs
The only vehicles in the Westminster study for which the emission rates of the exhaust emissions
were measured were the two all-electric test vehicles EV-1 and EV-2. As described earlier, the
exhaust emissions were entirely artificial by conducting metered releases of dry gas from gas
cylinders. The artificial dry exhaust from EV-1 simulated clean emissions, which contained
15.05 % CChin nitrogen. EV-2 artificial exhaust emissions simulated a dirty vehicle with 402
ppm propane, 5043 ppm CO, 996 ppm NO, and 14.76 % C02in nitrogen. For both vehicles, the
simulated exhaust gas was released at 30 scfm. Since the artificial exhaust gas releases were
replicated independently of the various release rates used for artificial evaporative emissions,
many replicate measurements of the exhaust emissions are available for analysis.
To obtain exhaust emission rates using the methodology, the detailed RSD data for the NO, CO,
and CO2 data undergo the processing steps shown in Figure 5-1. Specifically, no BSS separation
or estimation is needed to convert the raw RSD detailed data to Release Rates (g/hr) and
Emission Rates (g/mile).
Table 9-2 shows the Metered release rates in Column 3 and the corresponding Mean RSD-
measured release rates in Column 4 with standard error, lower 95% confidence limit on the
mean, upper 95% confidence limit on the mean, and the t-value in subsequent columns. The last
column gives the percent recovery, which is the ratio of Mean divided by Metered. The
recoveries range from 66% to 87%.
9-2
-------
Table 9-2. Comparison of Metered and RSD-Measured Exhaust CO, NO, and CO2
Release Rates for EV-1 and EV-2
Vehicle ID
Nobs
Exh NO (g/hr)
Metered
Mean
Std Err
LCLM
UCLM
t
Recovery
EVl
159
0.00
0.37
0.25
-0.13
0.86
1.5
EV2
260
63.1
41.9
1.08
39.8
44.1
39.0
66%
Vehicle ID
Nobs
Exh CO (g/hr)
Metered
Mean
Std Err
LCLM
UCLM
t
Recovery
EVl
159
0.00
1.74
2.20
-2.60
6.08
0.79
EV2
260
298
227
6.04
215
239
37.6
76%
Vehicle ID
Nobs
Exh C02 (g/hr)
Metered
Mean
Std Err
LCLM
UCLM
t
Recovery
EVl 159 13991 12239 361 11526 12953 33.9 87%
EV2 260 13721 10715 253 10218 11213 42.4 78%
9.2 Total HC, Evaporative HC, Exhaust HC Release Rates from Reference EVs
The performance of the methodology for unambiguously determining the methodology's
performance for measuring HC Release Rates (g/hr) can be done using only data from the EV-1
and EV-2 test vehicles. Only for those two vehicles were all EvapHC and ExhHC release rates
metered. During testing, the other test vehicles (F150, GMC, Subaru) emitted their natural
EvapHC and ExhHC emissions; however, as shall be shown in the next subsection, those
emissions appear to be quite low.
For EV-1, the artificial dry exhaust gas was released at 30 scfm, but the gas contained no HC.
Therefore, the EV-1 Metered ExhHC release rate was 0 g/hr. For EV-2, the artificial dry exhaust
gas was released at 30 scfm, and the gas contained 402 ppm propane, which produces a release
rate of 37.4 g/hr HC. For both EV-1 and EV-2, propane was released from either the fuel fill
door (DOOR), under the hood (HOOD), or at the center of the rear axle (TANK) and at either 0,
1.1, 2.3, 4.5, 9, 18, 36, 72, 144, or 288 g/hr. Additionally, the test vehicles were driven at either
22.5 or 45 mph past the RSD instrument.
The processing described by Figure 5-1 was used to convert the raw RSD HC channel data into
THC (Total HydroCarbon) release rates. Specifically, BSS separation and estimation is not
needed and was not used to calculate the THC release rates. On the other hand, since the raw
RSD HC channel must be split to arrive at the separate release rates for EvapHC and ExhHC, the
processing described by Figure 5-2 was used to provide the separation (by ICA) and estimation.
9-3
-------
Because of the different processing paths used for THC vs. EvapHC and ExhHC, comparison of
Measured vs. Metered THC provides evaluation of different processing steps than the
comparison of Measured vs. Metered EvapHC and Measured vs. Metered ExhHC. Calculated
values of THC are affected by the processing steps in Figure 5-1, namely, pre-processing, vortex
shape calculation, and emission calculation. Calculated values of EvapHC and ExhHC are
additionally affected by the additional processing step in Figure 5-2, namely,
separation/estimation.
Because test vehicles EV-1 and EV-2 were driven in every transit of the convoy past the RSD
instrument, many replicate RSD measurements are available for analysis. The analysis begins by
examining the trends of averages of replicate transits. Tables 9-3 and 9-434 show the average
results for THC, EvapHC, and ExhHC for EV-1 and EV-2. Later, we will examine plots of the
data from the table, and in the next section, we will examine data tables and plots for similar
calculations for the gasoline test vehicles.
Because the tables and plots for all five test vehicles have the same format, we will describe and
discuss the results for EV-1 in some detail.
34 P:\EDARinDenver-OCT2019\Analysis_MLout\220817\Anal_MLout\RefVehs/
OCT19_perf_RefVeh.xlsx, which was derived from P:\EDARinDenver-
OCT2019\Analysis_MLout\220817\Anal_MLout\RefVehs / OCT19_perf_RefVeh.sas and * .1st
9-4
-------
Table 9-3. Metered HC and Measured HC Average Responses for Test Vehicle EV-1
Vehicle ID
Evap Release
Location
N
obs
Metered
fake
EvapHC
Evap HC (g/hr)
Measured
Metered
fake
ExhHC
ExhHC (g/hr)
Measured
Metered
fake
THC
THC (g/hr)
Measured
(g/hr)
Mean
Std Err
t
(g/hr)
Mean
Std Err
t
(g/hr)
Mean
Std Err
t
EV-1
61
0
-1.2
1.9
-0.6
0.0
1.6
0.6
2.8
0.0
1.1
2.0
0.5
5
1.1
7.6
2.0
3.8
0.0
7.9
3.2
2.4
1.1
15.6
4.0
3.9
10
2.3
7.8
6.5
1.2
0.0
13.0
5.5
2.4
2.3
21.2
6.7
3.2
10
4.5
14.0
8.9
1.6
0.0
5.8
2.3
2.5
4.5
20.1
7.7
2.6
EV1
Door
10
9
26.9
6.9
3.9
0.0
6.8
1.4
5.0
9.0
34.0
7.3
4.7
10
18
14.5
2.6
5.7
0.0
16.1
4.0
4.0
18.0
31.0
4.8
6.4
9
36
36.0
9.2
3.9
0.0
14.1
4.4
3.2
36.0
50.8
8.8
5.8
9
72
56.4
9.6
5.9
0.0
22.5
7.0
3.2
72.0
79.9
14.3
5.6
10
144
136.5
13.2
10.4
0.0
28.4
14.1
2.0
144.0
165.7
11.7
14.1
5
288
250.6
35.9
7.0
0.0
79.5
35.8
2.2
288.0
334.7
48.2
6.9
5
1.1
3.6
1.8
2.0
0.0
1.0
0.4
2.4
1.1
4.5
1.6
2.8
10
2.3
4.9
4.6
1.1
0.0
2.9
1.5
1.9
2.3
8.9
5.2
1.7
10
4.5
13.1
5.8
2.3
0.0
4.2
1.4
3.1
4.5
17.4
5.1
3.4
EV1
Hood
10
9
9.7
5.5
1.8
0.0
3.6
2.1
1.7
9.0
14.1
6.3
2.2
10
18
8.0
4.1
2.0
0.0
6.9
1.8
3.9
18.0
15.5
4.9
3.2
9
36
14.1
6.1
2.3
0.0
5.5
2.3
2.4
36.0
20.1
6.7
3.0
10
72
29.8
4.0
7.5
0.0
10.0
3.3
3.1
72.0
39.9
4.7
8.4
10
144
75.1
10.6
7.1
0.0
11.9
3.3
3.7
144.0
89.0
10.5
8.5
5
288
104.1
18.0
5.8
0.0
38.0
18.5
2.1
288.0
143.8
28.7
5.0
5
1.1
5.5
3.3
1.7
0.0
4.7
1.1
4.4
1.1
10.5
3.7
2.9
11
2.3
6.5
2.8
2.4
0.0
5.2
1.0
5.3
2.3
11.9
3.4
3.5
10
4.5
-0.5
7.0
-0.1
0.0
10.9
2.1
5.2
4.5
16.9
3.1
5.5
EV1
Tank
10
9
13.4
5.2
2.6
0.0
10.0
1.7
5.7
9.0
23.4
5.2
4.5
10
18
5.7
8.6
0.7
0.0
10.4
2.9
3.6
18.0
17.5
9.7
1.8
12
36
23.6
8.9
2.7
0.0
19.2
3.5
5.4
36.0
43.0
8.5
5.0
10
72
37.1
11.4
3.2
0.0
23.8
5.2
4.6
72.0
69.2
8.9
7.8
12
144
65.8
11.8
5.6
0.0
58.8
16.9
3.5
144.0
125.3
8.2
15.4
5
288
118.7
8.8
13.5
0.0
62.2
17.0
3.7
288.0
181.6
11.7
15.6
9-5
-------
Table 9-4. Metered HC and Measured HC Average Responses for Test Vehicle EV-2
Vehicle ID
N
obs
Metered
fake
EvapHC
Evap HC (g/hr)
Metered
fake
ExhHC
ExhHC (g/hr)
Metered
fake
THC
THC (g/hr)
Evap Release
Location
Measured
Measured
Measured
(g/hr)
Mean
Std Err
t
(g/hr)
Mean
Std Err
t
(g/hr)
Mean
Std Err
t
EV-2
62
0
-0.9
2.0
-0.5
37.4
19.9
1.8
10.9
37.4
20.6
2.6
7.8
5
1.1
-0.6
4.5
-0.1
37.4
33.1
2.8
11.9
38.5
32.5
5.7
5.7
10
2.3
6.5
5.8
1.1
37.4
30.9
4.7
6.5
39.7
42.6
5.8
7.3
10
4.5
8.7
4.3
2.0
37.4
27.5
5.7
4.9
41.9
42.9
6.0
7.1
EV-2
Door
10
10
9
18
6.9
10.0
4.2
3.3
1.6
3.1
37.4
37.4
30.0
35.0
4.8
2.7
6.3
13.0
46.4
55.4
38.3
45.6
6.3
4.0
6.1
11.5
10
36
43.4
8.9
4.9
37.4
29.7
4.9
6.1
73.4
73.7
11.2
6.6
10
72
56.1
10.0
5.6
37.4
36.6
5.4
6.7
109.4
97.4
8.8
11.1
10
144
155.2
19.2
8.1
37.4
47.6
10.8
4.4
181.4
203.4
22.2
9.2
5
288
131.2
25.8
5.1
37.4
65.3
34.9
1.9
325.4
245.3
36.2
6.8
5
1.1
3.1
3.4
0.9
37.4
26.1
5.6
4.7
38.5
30.4
7.2
4.2
10
2.3
1.6
3.1
0.5
37.4
30.2
3.9
7.8
39.7
32.2
3.1
10.4
10
4.5
1.9
8.6
0.2
37.4
35.1
5.6
6.2
41.9
44.0
8.8
5.0
EV-2
Hood
10
10
9
18
9.2
11.4
5.7
6.4
1.6
1.8
37.4
37.4
29.3
34.1
4.7
3.4
6.3
10.1
46.4
55.4
45.4
50.4
7.3
6.1
6.2
8.3
10
36
13.4
2.3
5.8
37.4
34.7
4.7
7.3
73.4
48.5
5.5
8.8
10
72
17.3
4.2
4.1
37.4
45.8
6.9
6.6
109.4
69.5
5.0
13.9
10
144
59.9
4.4
13.5
37.4
55.8
13.2
4.2
181.4
118.9
12.0
9.9
5
288
53.4
14.6
3.7
37.4
72.3
19.0
3.8
325.4
151.6
14.7
10.3
5
1.1
8.3
2.8
3.0
37.4
30.7
5.2
5.9
38.5
39.5
6.1
6.5
12
2.3
8.0
3.0
2.6
37.4
35.0
3.1
11.4
39.7
44.9
4.6
9.7
10
4.5
4.3
5.9
0.7
37.4
37.5
4.2
8.9
41.9
42.6
3.6
11.8
EV-2
Tank
12
10
9
18
5.9
7.1
3.6
3.2
1.6
2.3
37.4
37.4
39.3
38.3
5.4
2.7
7.3
14.4
46.4
55.4
46.0
46.0
5.5
4.8
8.3
9.5
12
36
31.9
4.6
6.9
37.4
32.5
4.1
7.9
73.4
69.1
6.1
11.2
10
72
21.0
6.5
3.3
37.4
44.7
10.0
4.5
109.4
86.2
6.6
13.0
11
144
72.2
12.2
5.9
37.4
73.5
7.7
9.5
181.4
146.7
11.6
12.7
5
288
76.2
22.2
3.4
37.4
131.6
37.9
3.5
325.4
225.9
22.8
9.9
9-6
-------
Table 9-3 shows the average results for EvapHC, ExhHC, and THC for EV-1. The table has
fourteen columns in three groups of columns for EvapHC, ExhHC, and THC. Columns 3, 7, and
11 show the Metered release rates. Column 3 shows the various EvapHC release rates. Column 7
shows the constant ExhHC release rate of 0.0 g/hr. Column 11 is just the sum of Columns 3 and
7. The means, standard errors, and t-values for each average measured release rate are shown in
Columns 4, 5, 6; 8, 9, 10; and 12, 13, 14. The standard error can be viewed as the standard
deviation of the mean value. The t-value is just the mean divided by the standard error. Column 2
gives the number of replicates that are used to calculate the statistics in each row.
The table has four horizontal sections (groups of rows). The top section (first row) shows the
results when no EvapHC was released. The remaining sections show results for EvapHC releases
from Door, Hood, and Tank locations. Throughout the table, it is useful to compare a measured
value with the metered value just to the left of it. This provides information on the accuracy of
the methodology for that particular test condition represented by the row.
The first row shows that for EvapHC the measured mean plus or minus the standard error of -1.2
± 1.9 is in agreement with the metered value of 0 g/hr. This is an important result because it
means that the separation/estimation processing produces good EvapHC values when no
EvapHC is present. In the same row, the measured mean ExhHC value of 1.6 ± 0.6 g/hr is higher
than the metered ExhHC value of 0.0 g/hr, but it is not substantially higher.
As we move down the EvapHC columns, the measured values (Column 4) tend to increase but
not necessarily at the same rate of increase as the metered values (Column 3). The reasons for the
different rates of increase arise from the EvapHC release locations and will be considered later.
The same exercise can be done for the ExhHC columns. However, for ExhHC, the measured
values (Column 8) should stay constant as the metered values (Column 7) do. The table shows
that this is only somewhat true. At each of the three release locations (Door, Hood, Tank), there
is a tendency for "leakage" or "crosstalk" of some of the EvapHC into the ExhHC channel. This
tendency is a measure of the performance of the separation/estimation processing.
The features of Table 9-4 for EV-2 are similar to those of Table 9-3 for EV-1. One important
feature to be pointed out is the result in Table 9-4 in the first row for transits when no EvapHC
was released. For these transits, the release rate of ExhHC was quite high at 37.4 g/hr (402 ppm
exhaust HC). In spite of this large ExhHC release rate, the separation/estimation processing
produced an average EvapHC value of -0.9 ± 2.0 g/hr, which is not significantly different from
the metered EvapHC value of 0.0 g/hr.
The plots in Figures 9-1 and 9-2 use the data from Tables 9-3 and 9-4 to show the trends for EV-
1 and EV-2 more clearly.
In Figure 9-lb, which is for the fuel-fill-Door releases, the blue line shows that the measured
EvapHC increases close to linearly with the metered EvapHC values on the x-axis. The slope of
the blue line is also close to the ideal value of 1. The red, dashed line shows the much milder
increase of ExhHC as the metered EvapHC on the x-axis increases. If the separation/estimation
process were perfect, the red, dashed line would stay horizontal and be at the metered ExhHC
value of 0 g/hr, which is shown by the red solid horizontal reference line.
9-7
-------
a) EV-1: THC Overview
Figure 9-1. HC Performance (Average) for Test Vehicle EV-1
b) EV-1: Evap release at DOOR
Ul
-a
-------
a) EV-2: THC Overview
Figure 9-2. HC Performance (Average) for Test Vehicle EV-2
b) EV-2: Evap release at DOOR
300
300
3
u
x
a
ra
>
LU
¦a
<1>
~
I/I
ro
Q
-------
The plots in Figures 9-lc and 9-ld show the same plots for releases from Hood and Tank
locations for the same vehicle EV-1. The notable difference of these plots, compared with Figure
9-lb for Door, is that the slope of the blue lines is much lower for Figure 9-lc and 9-ld. This is a
consequence of the lower entrainment efficiency (VET) for releases farther forward on the
vehicle (Hood, Tank) in comparison with those at the vehicle rear (Door) in the case of EV-1.
Figure 9-la shows an overview of the effect of increasing the metered EvapHC release rate (x-
axis) on the THC release rate (y-axis) as a function of release location. The trend of the slopes is:
Door > Tank > Hood. Again, this is a consequence of the lower entrainment efficiency for
EvapHC release locations farther forward on the vehicle. The portion of THC that is contributed
by ExhHC is not greatly affected by its tailpipe release location since the VET factor for tailpipe
releases (if they are near the rear of the vehicle) of 0.96 is close to 1 (see Table 6-4).
The corresponding ExhHC and EvapHC plots for EV-2 are shown in Figure 9-2. The ExhHC
dashed curves (red) in Figure 9-2b, c, and d are compared to the red solid horizontal reference
line at 37.4 g/hr since that was the metered ExhHC release rate. Just as for the ExhHC curves for
EV-1 in Figure 9-1, these ExhHC curves show some crosstalk of EvapHC signal into the
deduced ExhHC signal. The blue curves for EV-2 EvapHC in Figure 9-2 are quite similar to the
corresponding curves for EV-1 EvapHC in Figure 9-1. And just as for EV-1, the EV-2 EvapHC
curves show the effects of release location.
Figures 9-3 and 9-4 show additional details for the performance of the methodology as applied to
the EV-1 and EV-2 test vehicles.
Figures 9-3a and 9-4a show a comparison of the average RSD-measured EvapHC release rates,
to which the release-location factors of Table 6-4 have been applied, against the metered
EvapHC release rates. For these plots, all curves should be on the 1:1 parity trend. Except for the
highest metered release rate at 288 g/hr, the location-corrected, RSD-measured trends of
EvapHC release rate are reasonably close to parity.
Plots in panels b, c, and d for Figures 9-3 and 9-4 show the individual transit measurements of
EvapHC vs. the metered values. These values are not corrected for release location and are the
values used to get the averages in Figures 9-1 and 9-2 for the corresponding panels. The data
points at each test condition show the considerable amount of scatter, which we attribute to
plume variability from turbulence behind the moving vehicle. The plots also show that the
variability increases with increasing levels of metered EvapHC.
9-10
-------
Figure 9-3. HC Performance
a) EV-1: Location-corrected EvapHC Overview
(Details) for Test Vehicle EV-1
b) EV-1: Evap at DOOR (scatter)
300
400
C)
50 100 150 200 250
Metered (fake) EvapHC (Propane) (g/hr)
Evap Location -"Door -"Hood *~Tank
300
0 50 100 150 200 250 300
Metered (fake) EvapHC (Propane) (g/hr)
d)
e) EV-1: Evap at HOOD (scatter)
f) EV-1: Evap at TANK (scatter)
4001
350
O)
J 300
I
Q. 250
(O
l5 200
-o
cu 150
3
l/l
ro
cu
100
50
(/) 0>
I ;
-501
0 50 100 150 200 250 300
Metered (fake) EvapHC (Propane) (g/hr)
yprojl /EDARinDenver-OCT2019/Analysis_MLouU22!l817/Anal_M Lout/RefVehsTOCTI 9_perf_Reft/eh_scatter.sss 29AUG22 15:17
400
^ 350
O)
Z 300
I
Q. 250
(O
|jj 200
T3
CU 150
3
m
fo
CD
S 50
I/)
-50
50
100
150
200
250
300
Metered (fake) EvapHC (Propane) (g/hr)
yprojl/EDARinDeriver-OCT2019/Analysis_MLout'220ai7/Anal_MLout/RefVehs/OCTI9_perf_RefVeh_scatter.sas 29AUG22 15:17
9-11
-------
Figure 9-4. HC Performance (Details) for Test Vehicle EV-2
g) EV-2: Location-corrected EvapHC Overview h) EV-2: Evap at DOOR (scatter)
300 r
o>
250
300
Metered (fake) EvapHC (Propane) (g/hr)
3
u
x
Q.
re
>
LU
T3
re
Q)
5
Q
(/)
100
50
If
-50-
0 50 100 150 200 250 300
Metered (fake) EvapHC (Propane) (g/hr)
yproj1/EDARinDenver-OCT2019Mrialysis_MLoiiU220817Mnal_MLout/Re1Veh£/OCT19_perf_Refveh_seattersas 29AUG22 15:17
350
U
X
Q.
re
>
LU
"O
0)
i-
3
W
re
a)
300
250
200
150
100
50
w
a:
-50
0 50 100 150 200 250 300
Metered (fake) EvapHC (Propane) (g/hr)
yprojl/EDARinDeriver-OCT2019/Analysis_MLout/220817/Anal_MLout/RefVehs/OCT19_perf_Re1Veh_scatter.sas 29AUG22 15:17
9-12
-------
9.3 Evaporative HC Release Rates from Reference Gasoline Vehicles
The performance of the methodology to determine EvapHC release was examined in the
previous subsection by examining the responses for EV-1 and EV-2, which are all-electric
vehicles. The advantage of the test conditions for those vehicles was that the composition and
release rates were well known. However, there were a few differences that could have an
influence on the performance results. First, the composition of the artificial exhaust gas was
relatively simple. HC was simulated using only propane, and NOx was simulated using only NO.
Second, the simulated exhaust did not contain any gaseous water of combustion, which makes up
a huge fraction of real exhaust. Third, because the simulated exhaust came from gas cylinders,
the exhaust flow rate used was 30 scfm, which is quite low compared to exhaust flow rates from
engines under moderate to high loads. Fourth, the EV test vehicles were Chevrolet Bolts, which
have bodies substantially smaller than many light-duty vehicles that are typically on the road.
Thus, it is also important to examine the performance of the methodology for test situations that
come closer to real-world operating environments of conventional gasoline-fueled vehicles -
even if some of the advantages of testing all-electric vehicles need to be relaxed. The three
gasoline vehicles tested were the F150, GMC, and Subaru. Their exhausts were natural, and their
flow and composition were not metered or measured, but since the vehicles were all 2016 model
year or newer, we expected that their exhaust and evaporative emissions would be well
controlled. Since the exhaust was a real product of combustion, its composition was complex
with a mixture of HC compounds, NOx compounds, and water of combustion. The exhaust flows
were likely typical of a variety of light-duty vehicles passing by the RSD instrument during data
collection. Finally, these vehicles had larger bodies than the EVs. The F150 was a pick-up with a
full-size cap over the bed; the GMC was a pick-up with an open bed; the Subaru Outback was a
small SUV with no distinct rear deck over a trunk.
Tables 9-5, 9-6, and 9-7 show the average measured release rates for EvapHC, ExhHC, and THC
for the three gasoline test vehicles using the same format as Tables 9-3 and 9-4 for the two EVs.
The values in Tables 9-5, 9-6, and 9-7 can be examined just as those for EV-1 and EV-2 in
Tables 9-3 and 9-4. Note that for the F150 and GMC pick-ups, the fuel fill door and therefore the
artificial EvapHC release point were located just aft of the driver's door. In the tables and
figures, this location is called Side.
The first row of all three tables shows that when no EvapHC was released, the methodology
produced average EvapHC release rates that were not significantly different from zero. We
conclude that the natural EvapHC of the three vehicles was basically 0 g/hr. Also, from the first
row, the mean ExhHC release rates were near 13.4, 9.6, and 6.2 g/hr, which are low values but
statistically significantly above zero (since their t-values are greater than 1.96). These values can
be viewed as the typical ExhHC release rates for these vehicles.
9-13
-------
Table 9-5. Metered HC and Measured HC Average Responses for Test Vehicle F150
Vehicle ID
Evap
N
obs
Metered
fake
EvapHC
(g/hr)
Evap HC (g/hr)
Measured
Metered
fake
ExhHC
ExhHC (g/hr)
Measured
Metered
fake
THC
THC (g/hr)
Measured
Release
Location
Mean
Std
Err
t
(g/hr)
Mean
Std
Err
t
(g/hr)
Mean
Std
Err
t
F150
19
0.0
-6.1
7.0
-0.9
Natural
13.4
4.6
2.9
n/a
12.2
7.9
1.6
1
1.1
-41.6
Natural
-0.1
n/a
-33.1
4
2.3
-2.1
2.9
-0.7
Natural
18.6
10.6
1.8
n/a
20.3
8.8
2.3
2
4.5
0.0
9.5
0.0
Natural
2.2
3.3
0.7
n/a
26.6
24.4
1.1
F150
Side
4
9
5.3
6.1
0.9
Natural
18.5
7.6
2.4
n/a
37.0
12.4
3.0
2
18
8.8
34.7
0.3
Natural
5.7
2.1
2.7
n/a
32.5
19.8
1.6
4
36
4.3
6.9
0.6
Natural
5.6
8.3
0.7
n/a
19.5
16.0
1.2
2
72
60.0
56.5
1.1
Natural
11.8
13.3
0.9
n/a
95.8
47.9
2.0
4
144
94.8
13.2
7.2
Natural
13.0
7.2
1.8
n/a
150.8
38.0
4.0
1
288
171.8
Natural
36.4
n/a
242.0
1
1.1
-24.6
Natural
0.1
n/a
-18.8
4
2.3
19.4
20.9
0.9
Natural
12.9
4.4
2.9
n/a
36.1
23.7
1.5
2
4.5
17.6
4.5
3.9
Natural
7.0
3.0
2.3
n/a
26.6
0.1
355.3
F150
Hood
4
9
-7.0
17.5
-0.4
Natural
18.4
9.0
2.0
n/a
13.7
22.5
0.6
2
18
9.3
16.7
0.6
Natural
24.8
7.5
3.3
n/a
52.7
15.6
3.4
4
36
-7.1
22.6
-0.3
Natural
19.8
10.4
1.9
n/a
13.5
19.5
0.7
2
72
11.4
0.7
17.1
Natural
30.8
7.5
4.1
n/a
60.5
6.8
8.9
4
144
65.1
24.8
2.6
Natural
55.1
16.0
3.4
n/a
131.3
28.9
4.5
1
288
215.9
Natural
163.9
n/a
385.6
1
1.1
23.2
Natural
5.8
n/a
31.3
4
2.3
-27.5
14.4
-1.9
Natural
22.2
13.4
1.7
n/a
17.9
13.2
1.4
2
4.5
-23.3
48.5
-0.5
Natural
15.5
12.6
1.2
n/a
-5.3
32.7
-0.2
F150
Tank
4
9
4.0
18.9
0.2
Natural
12.5
7.4
1.7
n/a
26.4
10.2
2.6
2
18
-13.6
34.2
-0.4
Natural
16.9
2.8
6.0
n/a
14.4
28.1
0.5
4
36
39.7
13.4
3.0
Natural
18.8
11.2
1.7
n/a
61.7
20.3
3.0
2
72
-9.7
45.7
-0.2
Natural
48.0
19.8
2.4
n/a
39.7
26.6
1.5
4
144
98.7
38.4
2.6
Natural
54.8
25.0
2.2
n/a
178.1
4.9
36.6
1
288
391.9
Natural
23.2
n/a
417.1
9-14
-------
Table 9-6. Metered HC and Measured HC Average Responses for Test Vehicle GMC
Vehicle ID
Evap
N
obs
Metered
fake
EvapHC
(g/hr)
Evap HC (g/hr)
Measured
Metered
fake
ExhHC
ExhHC (g/hr)
Measured
Metered
fake
THC
THC (g/hr)
Measured
Release
Location
Mean
Std
Err
t
(g/hr)
Mean
Std
Err
t
(g/hr)
Mean
Std
Err
t
GMC
12
0.0
-4.2
5.9
-0.7
Natural
9.6
2.7
3.5
n/a
10.7
5.8
1.8
1
1.1
-54.0
Natural
-2.2
n/a
-54.8
2
2.3
5.0
19.5
0.3
Natural
23.5
8.7
2.7
n/a
29.1
27.7
1.0
2
4.5
-15.3
3.1
-5.0
Natural
18.3
2.9
6.2
n/a
15.9
12.5
1.3
GMC
Side
2
9
-37.8
37.5
-1.0
Natural
46.7
36.1
1.3
n/a
10.3
0.2
55.8
2
18
-1.3
5.4
-0.2
Natural
11.3
5.4
2.1
n/a
45.0
17.5
2.6
2
36
6.0
12.3
0.5
Natural
5.1
2.4
2.1
n/a
11.2
10.0
1.1
2
72
58.6
58.1
1.0
Natural
12.4
3.4
3.7
n/a
78.8
53.7
1.5
2
144
66.5
18.5
3.6
Natural
50.0
46.3
1.1
n/a
117.8
26.7
4.4
1
288
-15.4
Natural
45.7
n/a
38.5
1
1.1
2.4
Natural
3.8
n/a
8.2
2
2.3
26.3
18.7
1.4
Natural
3.4
3.9
0.9
n/a
35.1
27.6
1.3
2
4.5
-16.5
22.6
-0.7
Natural
24.3
19.7
1.2
n/a
9.3
3.1
3.0
GMC
Hood
2
9
-21.4
29.8
-0.7
Natural
6.0
0.4
14.3
n/a
9.4
4.7
2.0
2
18
8.6
10.0
0.9
Natural
10.5
2.6
4.1
n/a
18.4
8.2
2.3
2
36
61.0
53.8
1.1
Natural
5.5
2.2
2.5
n/a
66.4
56.0
1.2
2
72
-7.6
3.3
-2.3
Natural
6.8
10.8
0.6
n/a
0.2
15.2
0.0
2
144
-44.9
57.1
-0.8
Natural
25.1
0.4
64.9
n/a
111.6
74.0
1.5
1
288
13.0
Natural
91.7
n/a
146.6
1
1.1
21.0
Natural
7.7
n/a
28.5
2
2.3
-2.7
0.6
-4.6
Natural
2.6
4.3
0.6
n/a
2.7
4.9
0.6
2
4.5
-33.5
34.2
-1.0
Natural
17.8
2.7
6.5
n/a
13.6
2.4
5.6
GMC
Tank
2
9
-24.8
9.0
-2.7
Natural
15.3
8.2
1.9
n/a
-0.5
6.9
-0.1
2
18
4.4
9.8
0.5
Natural
44.1
18.2
2.4
n/a
50.4
10.3
4.9
2
36
0.7
12.6
0.1
Natural
25.4
3.7
6.9
n/a
26.3
16.2
1.6
2
72
18.5
20.9
0.9
Natural
24.1
4.7
5.2
n/a
42.7
25.6
1.7
1
144
-8.8
Natural
69.6
n/a
128.0
1
288
71.3
Natural
96.6
n/a
263.0
9-15
-------
Table 9-7. Metered HC and Measured HC Average Responses for Test Vehicle Subaru
Vehicle ID
Evap
N
obs
Metered
fake
EvapHC
(g/hr)
Evap HC (g/hr)
Measured
Metered
fake
ExhHC
ExhHC (g/hr)
Measured
Metered
fake
THC
THC (g/hr)
Measured
Release
Location
Mean
Std
Err
t
(g/hr)
Mean
Std
Err
t
(g/hr)
Mean
Std
Err
t
Subaru
23
0.0
-1.0
4.5
-0.2
Natural
6.2
2.6
2.4
n/a
5.2
2.7
2.0
2
1.1
-3.9
0.8
-5.0
Natural
4.6
4.5
1.0
n/a
0.7
3.7
0.2
4
2.3
-2.6
8.4
-0.3
Natural
2.3
1.3
1.8
n/a
16.3
6.3
2.6
4
4.5
11.8
2.4
5.0
Natural
8.6
4.4
2.0
n/a
21.7
5.7
3.8
Subaru
Door
4
9
8.3
8.6
1.0
Natural
4.9
2.2
2.3
n/a
14.5
8.9
1.6
4
18
21.9
4.7
4.7
Natural
16.1
1.0
16.6
n/a
39.1
5.0
7.8
4
36
30.0
8.1
3.7
Natural
19.9
5.7
3.5
n/a
57.3
1.7
33.5
4
72
49.9
10.3
4.8
Natural
14.9
7.4
2.0
n/a
71.0
3.0
23.6
4
144
159.4
33.4
4.8
Natural
22.6
10.7
2.1
n/a
183.3
38.1
4.8
2
288
198.3
21.1
9.4
Natural
98.8
17.9
5.5
n/a
296.6
3.0
99.0
2
1.1
1.7
9.7
0.2
Natural
5.4
2.3
2.3
n/a
13.3
1.3
10.2
3
2.3
-5.0
7.1
-0.7
Natural
4.9
3.7
1.3
n/a
2.7
7.3
0.4
4
4.5
-5.6
6.0
-0.9
Natural
15.2
4.8
3.1
n/a
15.0
2.4
6.3
Subaru
Hood
4
9
10.9
4.7
2.3
Natural
8.4
2.4
3.6
n/a
20.7
7.3
2.8
4
18
8.5
5.6
1.5
Natural
10.3
4.8
2.1
n/a
15.8
6.2
2.5
4
36
4.9
8.6
0.6
Natural
12.8
2.2
5.8
n/a
33.7
8.3
4.0
4
72
23.2
6.3
3.7
Natural
24.4
7.0
3.5
n/a
47.8
11.1
4.3
4
144
59.9
8.9
6.7
Natural
34.5
8.0
4.3
n/a
99.4
8.6
11.5
2
288
127.2
3.1
41.4
Natural
51.7
3.5
14.7
n/a
179.5
7.0
25.5
2
1.1
6.5
1.5
4.2
Natural
6.3
4.0
1.6
n/a
13.1
5.2
2.5
4
2.3
-10.1
5.6
-1.8
Natural
6.0
1.8
3.4
n/a
2.5
4.6
0.5
4
4.5
-11.2
7.0
-1.6
Natural
13.4
9.7
1.4
n/a
13.1
5.4
2.4
Subaru
Tank
4
9
5.1
2.7
1.9
Natural
6.6
2.6
2.5
n/a
23.2
5.7
4.1
4
18
28.4
25.9
1.1
Natural
15.9
4.8
3.3
n/a
58.5
43.0
1.4
4
36
9.9
17.9
0.6
Natural
13.0
5.6
2.3
n/a
35.8
9.3
3.9
4
72
68.2
25.7
2.7
Natural
34.3
7.6
4.5
n/a
105.0
26.5
4.0
4
144
108.5
16.6
6.5
Natural
41.9
27.6
1.5
n/a
154.5
12.9
12.0
2
288
171.8
29.3
5.9
Natural
63.2
8.3
7.6
n/a
237.5
40.2
5.9
9-16
-------
The mean and metered values in Table 9-5, 9-6, and 9-7 are plotted in Figures 9-5, 9-6, and 9-7.
The solid purple lines provide a comparison of the RSD-measured EvapHC (g/hr) on the y-axis
with the metered EvapHC (g/hr) on the x-axis. Since for these test vehicles the ExhHC was
neither measured nor metered, we used, as a best estimate, the RSD-measured ExhHC (g/hr)
values from the first row of Tables 9-5, 9-6, and 9-7 to draw the solid red horizontal reference
lines that designate the ExhHC release rates in panels b, c, and d. The RSD-measured trends for
ExhHC, as the metered EvapHC increases, are shown by the dashed red lines. Generally, these
dashed red lines rise monotonically above the solid red reference lines indicating the presence of
crosstalk between the EvapHC and ExhHC signals.
The trends seen in Figures 9-5 and 9-7 for the F150 and the Subaru agree with the trends seen for
EV-1 and EV-2:
a) The response of measured EvapHC tends to increase linearly with the metered EvapHC.
b) The calculated values of ExhHC tend to increase as metered EvapHC increases, which
suggests that the separation/estimation processing is not operating optimally.
c) For the F150, the responses to EvapHC releases from the Side and Hood locations are smaller
than to releases from the Tank location.
d) For the Subaru, the responses to EvapHC releases from the Tank and Hood locations are
smaller than to releases from the Door location.
For the GMC pick-up, the mean EvapHC responses shown in Figure 9-6 do not show reliable
linear trends. We attribute this to the relatively small number of transits driven by the GMC.
Figures 9-8, 9-9, and 9-10 show details of the trends for these vehicles. The trends for the F150
and the Subaru show good evidence of increasing measured EvapHC trends as the metered
EvapHC is increased.
9-17
-------
Figure 9-5. HC Performance (Average) for Test Vehicle F150
a) F150: THC Overview b) F150: Evap release at SIDE
-50
0 50 100 150 200 250 300
Metered (fake) EvapHC (Propane) (g/hr)
/proj1/EDARinDenver-OCT2019/Analysis_MLout/220817/Anal_MLout/RefVehs/OCT19_perf_RefVeh.sas 28AUG22 09:44
d) F150: Evap release at TANK
3
U
x
Q.
TO
>
LU
-o
L_
3
l/l
ro
0)
200
Metered (fake) EvapHC (Propane) (g/hr)
/projl/EDARinDenver-OCT2019/Analysis_M Lout/220817/Anal_MLout/RefVehs/OCT19_perf_RefVeh.sas 28AUG22 09:44
c) F150: Evap release at HOOD
D)
U
I
Q.
ro
>
LU
¦o
-------
a) GMC: THC Overview
Figure 9-6. HC Performance (Average) for Test Vehicle GMC
b) GMC: Evap release at SIDE
300
300
300
0 50 100 150 200 250 300
Metered (fake) EvapHC (Propane) (g/hr)
Evap Location -—Side Hood '""Tank
0 50 100 150 200 250
Metered (fake) EvapHC (Propane) (g/hr)
U)
U
X
.c
X
LU
T3
<0
a*.
3
1/1
fO
(C
CD
Q
V)
a.
9-19
-------
Figure 9-7. HC Performance (Average) for Test Vehicle Subaru
a) Subaru: THC Overview
b) Subaru: Evap release at DOOR
300
250
3
u
X
H
-o
Q)
k.
3
in
ro
CD
D
(/)
a:
200
150
300
100
50 100 150 200 250 300
Metered (fake) EvapHC (Propane) (g/hr)
Evap Location • Door "-Hood '""Tank
U)
u
x
Q.
ra
>
LU
¦o
(U
3
l/l
ra
0)
5
6
w
a.
250
200
150
100
-50
1300
250
200
150
100
J-50
0 50 100 150 200 250 300
Metered (fake) EvapHC (Propane) (g/hr)
3
u
x
-E
X
LU
¦o
01
1-
3
l/l
ra
LU
"O
-------
Figure 9-8. HC Performance (Details) for Test Vehicle F150
a) F150: Location-corrected EvapHC Overview b) F150: Evap at SIDE (scatter)
300
400-
300
Metered (fake) EvapHC (Propane) (g/hr)
350
3
u
x
a
re
>
LU
T3
a>
d
i/>
ra
a>
2
6
U)
a
300
250
200
150
100
50
-50
Evap Location
Hood
¦ Side
* Tank
50 100 150 200 250
Metered (fake) EvapHC (Propane) (g/hr)
300
c) F150: Evap at HOOD (scatter)
d) F150: Evap at TANK (scatter)
400
.C 350
J 300
x
Q. 250
03
>
LU
T3
Q>
3
in
ra
-------
Figure 9-9. HC Performance (Details) for Test Vehicle GMC
a) GMC: Location-corrected EvapHC Overview b) GMC: Evap at SIDE (scatter)
c) GMC: Evap at HOOD (scatter)
d) GMC: Evap at TANK (scatter)
400
0 50 100 150 200 250 300
Metered (fake) EvapHC (Propane) (g/hr)
/projl/EDARinDenver-OCT2019/Analysis_MLoutt22Q817/Anal_M Lout/RefVehs/OCTI 9_perf_RefVeh_scatter.sas 29AUG22 15:17
350
O)
U 300
x
Q. 250
(B
11J 200
¦a
0) 150
I/I
ro
0)
-------
Figure 9-10. HC Performance (Details) for Test Vehicle Subaru
a) Subaru: Location-corrected EvapHC Overview b) Subaru: Evap at DOOR (scatter)
c) Subaru: Evap at HOOD (scatter)
d) Subaru: Evap at TANK (scatter)
400
400
150
300
Metered (fake) EvapHC (Propane) (g/hr)
/projl /EDARinDenver-OCT2019/Analysis_MLout/220817/Anal_MLout/RefVehs/OCT19_perf_RefVeh_scatter.sas 29AUG22 15:17
3
U
350
300
Q. 250
re
lij 200
"O
-------
9.4 Recommendations for Development of the RSD Emission Rate Method
This study shows that, with different processing of detailed RSD data, direct estimations of
vehicle release rate (g/hr) and emission rate (g/mile) are possible. The traditional RSD method
for exhaust concentration (ppm) and fuel-specific mass emission rates (g/gFuel) cannot obtain
emission release and emission rates, and exhaust HC values are subject to potentially large errors
introduced by evaporative emissions. However, the traditional RSD method has a great
advantage in that it is straight-forward and minimizes certain types of interferences and vehicle-
specific dependencies. The traditional method has these attractive properties because its
calculations are based on the ratio of each pollutant's RSD signal to the simultaneous CO2RSD
signal. In contrast, the new RSD emission rate method is based, not on a ratio of pollutant
signals, but on the single, absolute RSD signal of each pollutant. Consequently, the new method
will benefit from additional development. This subsection briefly describes sixteen suggested
areas to improve the method in general and to extend it to medium- and heavy-duty on-road
vehicles. These are ordered with the most important areas for improvement at the top of each list.
To improve the method in general:
1. Poor Correlations among Exhaust Pollutant Detailed Data
2. RSD Signal Dependence on Laser Pathlength
3. Vortex Entrainment Time (VET)
4. Release Location Detection of Light-Duty EvapHC
5. Vortex Shape (Weights)
6. RSD Signal Accuracy
7. RSD Signal Attenuation
8. Evaporative Plume Signal-to-Noise Improvement
9. Drag Area
10. Enhanced Blind Source Separation
16. Interfering Plumes
To extend the method to medium-duty and heavy-duty vehicles
3. Vortex Entrainment Time (VET) - specific to MDVs and HDVs
5. Vortex Shape (Weights) - specific to MDVs and HDVs
11. Release Location Detection of Medium- and Heavy-Duty Exhaust
12. Diesel Engine Load
13. Particulate Material Pollutant Correction Factor
14. Trailer Configuration Detection
15. Emissions of Vehicles with Trailers
1. Poor Correlations among Exhaust Pollutant Detailed Data - The investigation of flags (see
Section 7.5) seems to indicate that correlations of CO and NO with CO2 are relatively good for
the test vehicles and for diesel vehicles but are frequently poor for the vehicles that dominate the
fleet - gasoline vehicles. Since CO, NO, and CO2 can originate only from exhaust sources, their
RSD detailed data should always be highly correlated. So, poor correlation among them is a
major concern. The BSS separations help identify poor correlations by displaying the remnant
9-24
-------
signals seen in the BSS output heatmaps of Noise 1, 2, and 3. Our analyses do not indicate that
the problem arises from our signal improvement processing.
Since the vortex contains releases from the last 5 to 10 seconds of engine operation, our
hypothesis is that recent changes in engine operation cause shifts between CO and CO2 mass that
is reflected in the spatial distributions of pollutants in the vortex. For gasoline vehicles, changes
in engine operation in response to load causes a time-varying shift of carbon between CO and
CO2. Thus, unless the engine has been operating under chemically steady-state conditions over
the last 5 to 10 seconds, none of the pollutant signals precisely represent the exhaust plume.
Nevertheless, it may be possible to determine the combustion-independent exhaust plume signal
by accounting for the stoichiometry differences seen in all pollutant channels for a given transit.
2. RSD Signal Dependence on Laser Pathlength - The EDAR instrument is positioned above
the lane, and the laser scans back and forth. This geometry causes the optical pathlength to be
longer for pixels near the edge of the lane and shorter for pixels near the center of the lane. This
geometry is not a problem for the traditional RSD concentration method. While the 400ppm
ambient CO2 produces about 20% of the total CO2 signal, HEAT has a proprietary method to
correct for the ambient CO2.
For the RSD emission rate method, the varying pathlength needs to be corrected for every
pollutant channel. We believe that the proposed O2 channel, discussed below, could be used to
make this correction in addition to correcting for signal attenuation.
Another possibility is to use both the CO2 and O2 channels to monitor and correct for the
changing pathlength. The O2 and CO2 optical masses at each pixel are simply a proportional
blend of the O2 and CO2 concentrations in ambient air and the O2 and CO2 concentrations at the
tailpipe exit. By taking advantage of that fact, it should be possible to use O2 and CO2 pixel
measurements to verify that the signals of all RSD channels have been consistently corrected to
changing laser pathlengths in each scan.
3. Vortex Entrainment Time (VET) - The VET of a transit is just the ratio of the Mass in
Vortex (g) to the Release Rate (g/hr). Thus, if the RSD measures a pollutant's Mass in Vortex,
the VET can be conveniently used to estimate the pollutant release rate. For the light-duty
vehicles in this study, the VET was found to depend on vehicle air velocity, vehicle drag area,
and emission release location, and VET was independent of pollutant and release rate. In future
data collection efforts, we would like to 1) confirm the dependencies of light-duty VETs and 2)
collect data that can be used to extend VET knowledge to medium- and heavy-duty vehicles.
To determine generalized values and dependencies of VETs, repeated RSD measurements of
representative test vehicles must be made while releases of a pollutant are being measured as the
vehicles are driven under carefully controlled conditions. These requirements suggest test
vehicles in a staging area out of traffic.
For confirming light-duty VETs, we suggest releasing 30,000 ppm NO in nitrogen from
gasoline-fueled test vehicles with a wide range of drag areas while they drive at two speeds
under a variety of wind conditions. We suggest using the high NO concentration so that the
natural NO emissions of the test vehicles will only trivially affect the RSD signal - thus avoiding
9-25
-------
procuring electric test vehicles. Using NO as the metered pollutant will produce strong RSD
signals. To determine EvapHC release location dependencies, the NO can be released from Door,
Tank, Hood, Side, and Well locations. To simulate exhaust emissions location dependencies, the
NO can be injected into the exhaust system between the catalyst and upstream of the muffler.
Light-duty test vehicles will have to operate at constant speed and release at constant conditions
for at least 15s (= ~3 VETs) before arriving at the RSD.
For discovering medium- and heavy-duty VETs, we suggest using the natural CO2 emissions of
test vehicles with a wide range of sizes while they drive at two speeds and a variety of wind
conditions. The test vehicles would also be chosen so that the fuel rate could be monitored with
an OBD data stream. During each test transit the engine would have to be operated so that the
fuel rate remains constant for 30s (= ~3 VETs) before passing the RSD so that the C02Mass in
Vortex (box trucks have a large vortex to fill) is substantially in dynamic equilibrium with the
CO2 release rate. Test vehicles should also be chosen to cover the range of exhaust release
locations (e.g., under-chassis tailpipe, over-cab stacks, rear bumper) seen in the fleet.
4. Release Location Detection of Light-Duty EvapHC - This Westminster study found that
light-duty VETs were dependent on emission release location. Therefore, to promote more
accurate EvapHC emission rate values, a method to get a general idea of EvapHC emission
release location is desirable. We suggest staged testing of test vehicles out of traffic to generate
transits with detailed data that can be used to develop a method.
One of the main challenges of EvapHC release location detection is the need for an improved
signal-to-noise ratio of the HC signal. Specifically, even with large EvapHC release rates (g/hr)
from test vehicles, finding the EvapHC plume is difficult. To judge the performance of candidate
signal processing release location detection methods using the HC channel data alone, we need
an EvapHC plume tracer gas that produces a strong signal. Therefore, we suggest releasing a
mixture of butane and 30,000 ppm NO. The butane simulates the EvapHC since it is a major
component of EvapHC emissions. Butane will be metered over a wide range. The release rate of
the NO will be kept high so that the true location of the EvapHC plume can be easily seen in the
RSD NO detailed data stream. Then, signal processing techniques for estimating EvapHC release
location can be developed on the butane detailed data and tested against the NO detailed data,
which defines the true location of the EvapHC plume.
The effort will focus on determining release location for moderate and high EvapHC release
rates since knowing the release location of low EvapHC releases is less important and is
expected to be quite difficult to achieve. Because we will not attempt low butane release rates,
the test vehicles can be gasoline-fueled vehicles, whose natural ExhHC and EvapHC release
rates will be lower than the butane release rates. Three different vehicle shapes (sedan, sport-
utility vehicle, and pick-up with open bed) will be a minimum to generate widely different
EvapHC plume dispersion patterns from a few locations (Door, Tank, Hood, Side). As usual,
tests will be conducted at two speeds, under a variety of wind conditions, and with RSD
approaches at constant release and operating conditions for at least 15s (= ~3 VETs).
5. Vortex Shape (Weights) - The RSD emission rate method uses the expected shape of the
vortex that follows the vehicle to weight the measured optical masses at each pixel. The shape is
the relative magnitude of optical absorption of emissions at different scans and/or pixel positions
9-26
-------
in the vortex. The weights are used to effectively improve the signal-to-noise ratio of the RSD
signal of each pollutant channel and thereby improve the detection limit of the calculated
emission rates (g/mile). In this report's analysis, we developed one-dimensional (ID) weights
with one weight for each RSD scan behind the vehicle. We developed these weights primarily
for light-duty vehicles. These light-duty ID weights were found to be a function of time after the
vehicle rear, the parallel component of the airspeed, and vehicle length. Light-duty ID weights
were found to be independent of pollutant. We expect that two-dimensional (2D) weights, which
are effectively weights on individual pixels, will be additionally dependent on the perpendicular
component of the airspeed.
In future work, we would like to confirm the ID weights on light-duty vehicles, extend the ID
weights to medium- and heavy-duty vehicles, and collect EDAR detailed data for and develop
2D weights for light-, medium-, and heavy-duty vehicles. These extensions to weights can be
made by analyzing the CO2 in vortexes of thousands of fleet vehicles in traffic. The light-duty ID
weights did not seem to be a function of vehicle drag area or pollutant release location. However,
because exhaust release locations vary considerably among individual medium- and heavy-duty
vehicles, it would be prudent to plan to examine future field data for these possible influences.
Metered releases are not needed to determine vortex shape (weights) since only relative, not
absolute, magnitude is determined. And any pollutant can be used to determine weights. This
Westminster study used the CO2 in the vortexes of the 30,000 fleet vehicles to determine the ID
weights and their dependencies. We also found that we needed at least this many transits to
determine the dependencies. When a small subset of the 30,000 transits was analyzed for
dependencies, the trends became confused. We expect that this was a consequence of the
turbulence that is always present around vehicles moving through air.
6. RSD Signal Accuracy - In its detailed data files, the EDAR instrument reports the optical
mass measurements at each pixel of each pollutant channel in moles/m2. Unlike the traditional
RSD concentration method, which uses ratios, the RSD emission rate method uses absolute RSD
optical mass values and therefore requires accurate reported detailed data values. This
Westminster study assumed that the reported optical mass values were accurate. To be certain
that EDAR's reported optical masses are accurate, we need to have a field method to verify or
measure the accuracy of the reported optical mass values for each EDAR pollutant channel.
HEAT may already have a method such as a gas cell within the EDAR instrument. Using a gas
cell to check calibration at the RSD site could provide data to correct for any variations in
reflection efficiencies of different areas of the retro-reflective pavement tape.
7. RSD Signal Attenuation - Also because the RSD emission rate method uses the absolute
RSD optical mass values, the RSD signals on which the method's reported values of emission
release rate are based, are subject to attenuation caused by wear of and dirt on the retro-reflective
tape and by particulate material in the air between the EDAR instrument and the pavement. The
traditional RSD concentration method, which is used by EDAR, is not affected by such
attenuation sources. We need to develop a way to monitor and to correct for these attenuations so
that the calculated emission rates (g/mile) are not biased low.
The attenuation of the EDAR optical path could be monitored continuously by getting an RSD
signal from gases that occur naturally in the atmosphere. Ambient CO2 could be used since its
9-27
-------
global concentration is about 400 ppm. However, the ambient CO2 concentration around a busy
roadway is likely higher since the vehicles on the road emit large exhaust volumes of CO2 with
concentrations around 15% (=150,000 ppm). This method might work in out-of-traffic staging
areas, on roadways with light traffic, or roadways that have non-polluted air blowing across
them, but it will not work on busy roadways since ambient CO2 levels are greater than 400ppm
there.
A procedure for correcting for signal attenuation might be tested by artificially attenuating the
RSD signals by placing and removing an attenuation grid on top of the retro-reflective tape.
8. Evaporative Plume Signal-to-Noise Improvement - The noise that is present in the BSS
output channel called Evap Plume causes uncertainty in the calculated EvapHC emission rate
(g/mile). In turn, this causes the detection limit of the EvapHC emission rate to be higher than it
would otherwise be. Therefore, reduction of the noise in Evap Plume is desirable. Generally,
most of the noise in Evap Plume is from the noise in the adjusted HC channel, which is one of
the inputs to the BSS. Even though the noise in raw detailed HC data is reduced by the signal
analysis techniques described in this report, because the typical EvapHC emission rates (g/mile)
of fleet vehicles are small, the S/N ratio of EvapHC is small. Therefore, to improve the detection
limit of EvapHC emission rates, either the noise in Evap Plume or in adjusted HC needs to be
reduced and/or the EvapHC signal in the RSD HC channel needs to be increased.
One method that might be used to increase the EvapHC signal in the RSD HC channel is to
select an EDAR infrared (IR) wavelength that focuses on the IR spectrum of butane, which is the
dominant gas in EvapHC emissions. Reducing noise and artifacts in the Evap Plume signal, that
is, the BSS output, might also yield improved S/N ratios.
Another candidate for improving the evaporative plume S/N ratio is to use ethanol vapor as a
tracer for the EvapHC plume. Our modeling of gasoline blends with 10 wt% ethanol, which is
now universally used in the United States, indicates that its headspace is about 20 mole% ethanol
and the whole evaporated blend contains about 18 mole% ethanol. Thus, whether an evaporative
emission is produced by an evaporative emission control system vapor leak or by an evaporated
gasoline liquid leak, the emission is about 20 mole% ethanol. On the other hand, only about 3
mole% of the NMHC of ExhHC is ethanol. The infrared spectrum contains about eight sharp
spectral lines that are candidates for evaluation as targets for measurements of ethanol by the
DiAL method. If relatively free of interferences from other compounds, one of the ethanol lines
might be able to increase the evaporative S/N compared to using the EDAR HC channel's signal
as done in this study.
9. Drag Area - The analysis of the September 2016 (TTI) data and the October 2019
(Westminster) data indicates that the drag area of a vehicle mildly affects the VET. Therefore,
knowledge of the drag area of the vehicle in each transit would help improve the calculated
emissions rate. For some, but not all, vehicles, drag areas can be obtained from a look-up table
using the VIN obtained from the license plate. In a further development, it might be possible to
apply machine learning to the vehicle footprint (when the laser beam is occluded by the vehicle)
and perhaps an image of the side profile of the vehicle to estimate the drag area of the vehicle.
Such a method would be able to the estimate drag area of any vehicle without using a look-up
table.
9-28
-------
10. Enhanced Blind Source Separation - The Independent Component Analysis (ICA) method
using FastICA was the BSS method used in this first analysis of the Westminster dataset to split
the RSD HC signal into an EvapHC signal and an ExhHC signal. This method produces very
good separations when the EvapHC and ExhHC plumes do not overlap. However, we found that
in circumstances where the EvapHC and ExhHC plumes substantially overlap, the standard ICA
separations produce only mediocre "first-order" separations, which are usable but are not
optimal. In these situations in the portion of the vortex just behind the tailpipe, too much HC
mass is assigned to ExhHC and not enough mass to EvapHC. In some cases, the mass assigned
to EvapHC at the tailpipe exit location is a slightly negative - seen in heatmaps as the "blue
hole" - a physical impossibility.
To address the problem, we developed a new BSS technique called BSScov (see Section 7.3) in
which one of the constraints in the matrix algebra is relaxed to produce better separation results
when plumes overlap, which is the usual case for vehicles in light-duty fleets. However, we have
not yet developed a method to determine when the adjustable parameter, rho, is at an optimum.
Without the BSScov method, the standard ICA method produces usable results, but developing a
method to optimize BSScov's rho will produce superior results.
11. Release Location Detection of Medium- and Heavy-Duty Exhaust - Just as for EvapHC,
the release location of exhaust emissions can affect the VET of exhaust plumes. For light-duty
exhaust emissions, the release location will always be from the vehicle rear, and therefore
determination of release location for light-duty exhaust is not needed. On the other hand, on
medium- and heavy-duty vehicles, exhaust exit locations vary. For example, medium-duty trucks
may release exhaust from just behind and below the cab or below the rear of the cargo box. We
expect that these different locations will have different release location fingerprints.
A method for detecting different exhaust release locations of medium- and heavy-duty vehicles
can be developed from the same data that is collected for the VET effort described above. That
is, we suggest using the natural CO2 emissions of test vehicles with a wide range of sizes while
they drive at two speeds and a variety of wind conditions. The test vehicles would also be chosen
with an OBD data stream so that the fuel rate can be monitored. During each test transit the
engine would have to be operated so that the fuel rate remains constant for 30s (= ~3 VETs)
before passing the RSD so that the CO2 Mass in Vortex (box trucks have a large vortex to fill) is
substantially in dynamic equilibrium with the CO2 release rate. Test vehicles should also be
chosen to cover the range of exhaust release locations seen in the fleet.
Stewart Hager of HEAT indicates that for large trucks releasing exhaust emissions near the front
of the truck 1) vehicle speeds of at least 17 mph are needed to produce a significant vortex
behind the box or trailer, and 2) exhaust emissions "hug" the sides of the box (Bernoulli) on their
way to the vortex. The second of these observations might be verified by collecting EDAR data
using a set of two EDAR instruments placed above the left and right edge of the lane but looking
across the full width of the lane. Such a set-up might allow seeing emissions move past the side
of the box or trailer and establish the point at which transverse air movement causes a reduction
of emissions entrainment in the vortex behind the vehicle.
12. Diesel Engine Load - The load on an in-use diesel engine is important to know to judge if
the engine emissions are elevated with respect to the emission certification standards of the
9-29
-------
engine. An emission-compliant engine operating under high load might have the same high
emissions as a non-emissions-compliant engine operating under light load. While road grade
contributes to engine load, for cargo-hauling diesel vehicles the weight of the cargo is an
important factor. While we do not have a method to remotely determine vehicle weight, we
believe that the RSD emission rate method applied to CO2 can provide an estimate of diesel
engine load.
Diesel engines tend to have similar CO2 emission rates (g/mile) when operating under similar
loads - pretty much regardless of engine displacement, design, RPM, and rated horsepower. This
is a consequence of the heat content of the fuel. If we assume that all diesel engines have about
the same efficiency, the gross load on the engine (including internal frictional losses) is
proportional to the fuel rate (g/hr). Since the carbon in the fuel becomes primarily exhaust CO2,
the CO2 release rate (g/hr) is directly proportional to the fuel rate. Because the RSD emission rate
method can determine CO2 release rate, it can determine gross engine load. Thus, the RSD
method can not only measure pollutant release rates (g/hr) and emission rates (g/mile), it should
be able to estimate the gross engine load. Knowledge of the engine load would allow the
distinction between an emission-compliant engine operating under high load and a non-
emissions-compliant engine operating under light load.
By looking up the license plate in the vehicle registration database, decoding the heavy-duty
VIN, and determining the engine's displacement, the estimated minimum and maximum non-
boost fuel rates could be determined and compared to the RSD-measured fuel rate to estimate
relative engine load at the instant of the RSD transit. The minimum fuel rate can be determined
by assuming that an idling diesel will have an airfuel ratio near 100:1. The maximum fuel rate
can be determined by assuming that a diesel engine operating near 100% non-boosted load will
have an equivalence ratio35 near 0.6. Thus, dividing the RSD-estimated absolute fuel rate by the
displacement-estimated maximum fuel rate will provide an estimate of the relative load on the
engine. We would expect that turbo-boosted operation would simply produce relative loads
greater than 100%.
The ability of the RSD emission rate method to determine diesel engine load and relative load
using this technique could be evaluated during staged testing. The diesel test vehicles that were
equipped with a monitor of OBD fuel rate would be used to generate load and fuel rate data for
comparison with values determined by the RSD emission rate method. The diesel test vehicles
would be operated under a variety of engine loads and RPMs.
13. Particulate Material (PM) Pollutant Correction Factor - The exhaust PM emission rates
of diesel vehicles could be measured by the RSD emission rate method if the RSD optical signals
produced by the PM can be used to estimate the PM Mass in Vortex. The conversion from RSD
optical signal to PM mass depends on 1) the method used by the RSD and on 2) the Mie light
scattering of diesel particles. EDAR reports PM in units of particles/m2.
35 Equivalence ratio is the ratio of the actual air: fuel ratio to the stoichiometric air: fuel ratio. For diesel
fuel combustion, the stoichiometric airfuel ratio is about 14.4:1. Therefore, diesel engines are usually
designed to operate at actual airfuel ratios no less than about 25:1 to avoid excessive generation of
smoke.
9-30
-------
The sensitivity of EDAR to diesel PM could be determined on a diesel test vehicle in an out-of-
traffic staging area. The PM release rate (g/hr) and ideally size distribution would be determined
using a PEMS installed on the vehicle while the vehicle drove past the RSD under a variety of
engine loads to produce a variety of PM release rates.
An accompanying Mie light scattering modeling effort of reported diesel particle size
distributions would be able to determine the sensitivity of EDAR's measurement and the
calculated PM emission rate (g/mile) to the literature-reported variations in size distribution,
refractive index, and particle morphology.
14. Trailer Configuration Detection - In this Westminster study, the analysis eliminated
transits that appeared to be from vehicles pulling trailers since quantifying the emissions from
those vehicles was not the focus of the study. We developed SAS code to identify such vehicles.
However, if the efforts to quantify emissions from combination tractor-trailers is successful, we
would want to distinguish them from medium-duty pick-ups pulling work trailers. We think that
applying machine learning techniques to the RSD detailed data would be more productive for
solving this problem than using further in-depth SAS efforts.
15. Emissions of Vehicles with Trailers - With the determination of VETs, vortex shapes, and
an exhaust release location method, we expect that the exhaust emissions rate (g/mile) of
medium-duty box trucks and heavy-duty combination tractor-trailers could be determined using
RSD. However, quantification for medium-duty pick-ups pulling a wide variety of work trailers
may be more difficult. The Westminster data shows that vortexes for these transits are formed
behind the vehicle and behind the trailer. Accordingly, at this point, we do not propose work to
quantify the emissions of these types of vehicle assemblages.
16. Interfering Plumes - The EDAR QC flag assigned "interfering plume" to 7.2%
(=2302/31763) of the transits in the dataset of fleet vehicles at the relatively light-traffic
Westminster site. Interfering plumes can occur when a transit has emissions from another vehicle
detected by EDAR in front of the subject vehicle. Such emissions cause incorrect reported
emission concentrations when the standard calculations are used. So, standard protocol is to
throw out measurements where an interfering plume is detected. However, we think that it may
be possible to use signal processing to process the transit's raw data so that the interfering plume
can be separated from the subject vehicle's data and thereby obtain good emissions
measurements on the subject vehicle. This would reduce the number of transits that are thrown
out and would allow use of the EDAR instrument in heavier-traffic situations.
9-31
-------
10.0 Application of the RSD Emission Rate Method to the Fleet
Sample
In this section, we evaluate the RSD emission rate method by comparing the exhaust release
rates measured in Westminster in October 2019 with MOVES model predictions of exhaust
release rates for the same time period. To be able to make that comparison, we need to describe
the bases of the release rate values for the MOVES predictions and the RSD measurements.
Portable emissions measurement systems (PEMS) are sometimes used in field studies to measure
the rapid up and down changes on emissions concentrations, release rates, and flows on a
second-by-second basis. The exhaust gas is sampled using a probe that is inserted in the tailpipe.
The gas sample is conducted via a small-diameter, heated sample line, without dilution, to the
PEMS instrument package installed on board the vehicle. The PEMS analyzers typically have
fast response times to attempt to preserve the rapidly varying time traces.
MOVES predictions for newer model year light-duty vehicles are based primarily on
dynamometer tests that use a constant-volume sampling (CVS) procedure. The exhaust gases are
conducted through a heated pipe to the CVS system for dilution and concentration measurement.
During this procedure, the rapid ups and downs of concentrations and flows are smoothed
slightly. Thus, the dynamometer CVS measurements do not reflect the rapid second-by-second
changes that a PEMS instrument would have, but still CVS measurements preserve the rapid
changes relatively well.
Rather than conducting a sample of gas from the tailpipe exit, as PEMS and CVS systems do,
RSDs measure pollutants in-situ in the vortex without taking a sample of the gas. Because RSDs
measure optically, the RSD measurement process for one vehicle transit takes only about 0.5s.
While the measurement is fast, the RSD results reflect the mass of pollutants in the vortex - not
the pollutant flows from or concentrations in the tailpipe. The reason for this is that the mass of
pollutants in the vortex at any given instant is the result of entrainment of emissions released
from the vehicle and stripping of pollutants from the vortex by air moving across the surface of
the vortex as described by Figure 5-4. In this study, we have found that for light-duty vehicles
the vortex entrainment time (VET) is around 4 seconds. The mass of pollutants in the vortex is
influenced by the emissions released from the vehicle over at least the last 2 VETs or about 8
seconds. Thus, even though RSDs measure almost instantaneously, the masses in the vortex that
they measure are influenced by the changes in vehicle and engine operation over a comparatively
long time.
So, MOVES predictions are based on relatively rapid changes in exhaust, and RSD
measurements of the vortex mass reflect the relatively slowly changing effects of the entrainment
process. Thus, evaluating the RSD emission rate method by comparing MOVES predictions with
the Westminster RSD results must be done with the differences of time scales in mind.
10.1 Simulation of Vortex Entrainment using PEMS Data
An RSD measures the mass in a vehicle's vortex at one instant, but, of course, vortex
entrainment of pollutant releases is occurring constantly as the vehicle drives on the road. Thus,
the RSD is simply getting one measurement of a long time series of vortex masses. This study
10-1
-------
has given us an understanding of the entrainment process. Now, we can simulate that process to
determine the validity of comparing RSD measurements with MOVES predictions. We use
historical PEMS data to evaluate the validity.
We retrieved PEMS data collected from two vehicles that were procured for the Kansas City PM
Characterization Study. Table 10-1 shows information taken from Table 4-14 of the report.36 The
test notes for the Forester say that it was malfunctioning with abnormally high exhaust
temperature. The vehicle had NOx emissions of 5.764 g/mile - a quite elevated value. The Trail
Blazer had relatively low NOx emissions.
Table 10-1. Kansas City PEMS Data on Two Vehicles
CTR_TST_ID
D_KS2_904_1
D_KS2_795_1
Test Date
3/28/2005
3/9/2005
Vehicle
Description
2001 Subaru Forester
2002 Chevrolet Trail Blazer
Displacement
2.5 L
4.2 L
Test Distance
23.4 miles
56.9 miles
Composite C02
362.2 g/mile
479.5 g/mile
Composite CO
3.774 g/mile
2.272 g/mile
Composite NOx
5.764 g/mile
0.255 g/mile
Composite THC
0.136 g/mile
0.047 g/mile
PEMS Data File
Location
P:\KansasCity\SEMTECH
\Round2\Dri veaway s/
pp_MO_690SB GDRIVE AW AY
TestO M0-M3.csv
P:\KansasCity\SEMTECH
\Round2\Dri veaway s/
pp_MO_TBL970_DRIVEAWAY
TestO M0-M5.csv
Figure 10-137 demonstrates the simulation procedure using a 250-second portion of the PEMS
data for the Forester from Second 3050 to Second 3300 of the PEMS file.
36 S. Kishan, A.D. Burnette, S.W. Fincher, M.A. Sabisch, B. Crews, R. Snow, M. Zmud, R. Santos, S.
Bricka, E. Fujita, D. Campbell, P. Arnott, "Kansas City PM Characterization Study, Final Report,"
prepared for U.S. Environmental Protection Agency, prepared by Eastern Research Group, BKI, NuStats,
and Desert Research Institute, EPA-061027, October 27, 2006.
37 C:\Users\TDeFries\Documents\EPAWA5-13 (MAR22-FEB23)\8_Reports/
pp_MO_690SBG_DRIVEAWAY_Test0_M0-M3_THDplay.xlsx
10-2
-------
Figure 10-1. Vortex/RSD Simulation using PEMS Data for the Forester
40000
800
— NO measured by PEMS
3150 , x 3200
Elapsed Time (s)
10-3
-------
The top panel of Figure 10-1 shows the road speed (black, solid line). To calculate the VET
(purple, dashed line), we used the proportionality constant for the Subaru test vehicle at the
tailpipe location, which is given in Table 6-3 as 23.9. To calculate airspeed, we assumed that
there is always a lmph light breeze. This prevents the VET from going to infinity when the
vehicle is stopped. Therefore, the VET line in the top panel of Figure 10-1 is given by:
VET(s) = 23.9 / sqrt(RoadSpeed_mph + lmph)
We used the second-by-second VET to simulate the vortex entrainment of released emissions to
determine the release rate values that would be measured by an RSD instrument that
incorporated the RSD emission rate method. Note that the VET is independent of exhaust
pollutant.
The middle panel of Figure 10-1 shows the PEMS-measured CO2 release rate (g/hr) (light-blue
line), and the bottom panel shows the PEMS-measured NO release rate (g/hr) (orange line).
The entrainment is easily simulated in a spreadsheet using this procedure:
Initialization:
1) In Second 1, the release rate is given by the PEMS Release Rate,
2) In Second 1, the Mass in Vortex is an arbitrary value (which will converge after a few
iterations)
3) In Second 1, the Stripping Rate is the Mass in Vortex divided by the VET
Iteration:
4) For Second 2, the Mass in Vortex is calculated as Second l's Mass in Vortex plus
Second l's PEMS Release Rate, minus Second l's Stripping Rate,
5) For subsequent seconds, iterate Step 4.
The simulated RSD release rate time series is just the Mass in Vortex time series divided by the
VET time series, as described by Equation 5-17. This is equivalent to the stripping rate in the
spreadsheet.
This procedure was used to calculate the simulated release rate as would be measured by an RSD
instrument using the RSD emission rate method. The green dots in the middle panel give the
simulated CO2 release rate values, and the red dots in the bottom panel give the simulated NO
release rate values. Any one of these red dots is the expected release rate value that an RSD
instrument would obtain at a given second.
There are several things to notice about the plots in Figure 10-1:
1) The top panel shows that at speeds higher than about 15 mph, the changes in VET are
relatively small.
2) A comparison of the blue CO2 PEMS and orange NO PEMS data with the black Road Speed
data shows that relatively minor changes in speed produce large, sharp spikes in the PEMS
10-4
-------
tailpipe CO2 and NO release rates. This is a consequence of high engine load and the
corresponding high fuel rate needed to produce power at the wheels.
3) When the engine is idling with the vehicle at rest or when the vehicle decelerating, the PEMS
CO2 release rate is about 4000 g/hr, which is generally the lowest CO2 release rate in the
vehicle's dataset.
4) The simulated release rates (green CO2 dots and red NO dots) follow the PEMS release rate
values, but the simulated values are smoother and have rounded peaks that are delayed compared
to the PEMS peaks. In general, the simulated release rates do not go as high or as low as the
PEMS release rates.
5) As a consequence of the time-dependent entrainment of emissions in the vortex, second-by-
second PEMS and RSD measurements almost never agree, as demonstrated by the second and
third plots in Figure 10-1. Thus, matching a PEMS emissions measurement with an RSD
measurement on the same second cannot be a reliable method for checking RSD accuracy -
unless the emissions release rates are verified as constant over at least 2 VETs before the RSD
transit.
Given the substantial observed differences between the PEMS-measured release rate time series
and the simulated RSD emission rate method release rate time series, we begin wondering if the
RSD emission rate method's release rates capture the emission trends that the PEMS data does.
To help address that concern, we show the 250-second Forester snippet's PEMS NO vs. CO2
release rate in Figure 10-2 and the simulated RSD NO vs. CO2 release rate in Figure 10-3.
The log-log plot of the Forester PEMS NO vs. CO2 release rates in Figure 10-2 shows a locus of
1-second points that forms a "dog leg." The vertical branch of the dog leg is produced by
conditions when the engine is idling since the CO2 release rate values are around 4000 g/hr. For
these conditions, the NO release rate varied from 0 to about 60 g/hr. The angular branch of the
dog leg is formed by conditions with CO2 release rates greater than about 5000 g/hr, which
occurs when the engine is under at least some load greater than at idle. The key feature of the
angular branch is that most of the points are within a relatively tight clump with a slope
somewhat greater than 1. The angular branch indicates that, when the engine is operating under
some load even just slightly greater than at idle, the NO release rate usually increases with the
CO2 release rate.
Figure 10-3 plots NO vs. CO2 release rates for the same Forester data snippet, but for release
rates in the vortex as simulated for the RSD emission rate method using the PEMS data. Clearly,
the plot has a shape quite similar to Figure 10-2. The points in the vertical branch are more
scattered than for the PEMS data in Figure 10-2, but the angular branches are quite similar.
10-5
-------
Figure 10-2. NO v CO2 Release Rates as Measured by PEMS for Forester Snippet
Figure 10-3. NO v CO2 Release Rates as Simulated for RSD for Forester Snippet
10-6
-------
Figure 10-4. NO v CO2 Release Rates as Measured by PEMS for Forester
1000
100
1000 10000
C02 Release Rate measured by PEMS (g/hr)
100000
Figure 10-5. NO v CO2 Release Rates as Simulated for RSD for Forester
10-7
-------
Figures 10-4 and 10-5 show the same plots for the entire 1843-second Forester dataset when the
vehicle was moving faster than 15 mph. The figures show the same trends for this large dataset
as for the snippet plots. Figure 10-5 shows that the simulated RSD release rates have the same
trend as the PEMS data in Figure 10-4 when the CO2 release rate is larger than about 10,000
g/hr, which is about twice the idle CO2 release rate. One of the benefits of the accumulation of
emissions releases in the vortex is a decrease in the scatter of the simulated RSD release rates
relative to the measured PEMS release rates. Specifically, the scatter of points in the region
above 10,000 g/hr CO2 is less in Figure 10-5 than in Figure 10-4.
Figures 10-6 and 10-7 show the corresponding figures for the lower-NO-emitting Trail Blazer.
The symbols are shown only for the 6,836 observations when the vehicle speed was greater than
15 mph. The simulated RSD values in Figure 10-7 show lower scatter than for the PEMS
measurements in Figure 10-6. Comparison of Figure 10-5 for the Forester with Figure 10-7 for
the Trail Blazer shows that the clump of simulated RSD release rates for the Forester (red points)
are much higher on the plot field than for the Trail Blazer (green points). Consequently, the RSD
emission rate method can clearly distinguish these high- and low-emitting vehicles.
10-8
-------
Figure 10-6. NO v CO2 Release Rates as Measured by PEMS for Trail Blazer
Figure 10-7. NO v CO2 Release Rates as Simulated for RSD for Trail Blazer
10-9
-------
10.2 Comparison of RSD Emission Rates and MOVES Release Rates
The previous sections of this report describe the RSD emission rate method that uses RSD
detailed data to calculate mass release rate (g/hr) and, by dividing by vehicle speed, the mass
emissions rate (g/mile). Section 9 compared the method's RSD-determined release rates (g/hr) of
the test vehicles that we imbedded in the Westminster traffic with their metered emissions
releases. In this section, we compare the method's calculated release rates of the fleet vehicles
with MOVES (MOtor Vehicle Emission Simulator) modeled estimates of the release rates of the
vehicle fleet. Although the method has estimated release rates for all combinations of fleet
vehicle fuel type and regulatory class, the comparisons focus only on gasoline LDVs and LDTs
since they are by far the most common vehicle types in the Westminster fleet sample.
For the comparison, we selected MOVES3, EPA's well developed mobile source modeling
system, whose emission release rate information is derived from laboratory dynamometer
measurements and inspection/maintenance program dynamometer with some validation using
PEMS measurements. The purpose of the comparison is not to judge the accuracy of MOVES.
Instead, we anticipate that if both the RSD emission rate method and MOVES are reasonably
accurate, then emission release rates from both methods should compare well. We do not
necessarily expect to find a close match in emission rates between these two sources of data.
From the Westminster data, we have observed RSD signal strength to be correlated with factors
such as position within the road lane and wind speed - factors that are not present within
MOVES. Although we have taken steps towards accounting for such factors that increase or
decrease the measured RSD signal strength, an additional understanding of the problem is
needed so that further improvements to the RSD emission rate method can be made.
Release rate information with high specificity can be acquired from within MOVES, that is, from
MOVES internal tables, not from making MOVES runs. The latest MOVES3 database version
(movesdb20220802) was used to extract MOVES emission factors for the exhaust pollutants of
interest - exhaust CO2, NOx, CO, and Total HC38. The NOx, CO, and Total HC exhaust
emission rates are expected to change over the lives of the modeled vehicles; these rates are
found in the emissionratebyage table.
The emissionratebyage table uses the variable sourceBinID. This variable is 19 digits long and
can be parsed to decode specific fuel types, engine technologies, regulatory classes, and model
years. This Westminster field study occurred in 2019, so we kept only the model year and age
group combinations relevant for our data. For instance, for the 0-to-3-year-old age group
contained only model years 2020 through 2017. Exhaust emission rates for NOx, CO, and Total
HC were separated into distinct fuel type, regulatory class, vehicle specific power (VSP), and
age groups. These extracted rates show the expected behavior of different vehicle classes as their
emissions systems degrade over time.
Comparison of the RSD emission rate method and MOVES for LDTs and LDVs and for exhaust
CO2, NOx, CO, and HC are explored by the log scale plots in Figures 10-8 to 10-15. We use log
38 Here, Total HC refers to the MOVES context. Total HC means the sum of all hydrocarbon species
coming from the exhaust. Thus, Total HC includes exhaust methane and exhaust non-methane
hydrocarbon (NMHC). Total HC does not include EvapHC.
10-10
-------
plots so that we can see trends over a wide range of release rate values. On each plot, the exhaust
release rates (g/hr) are displayed on the vertical axis. The vehicle specific power (VSP) is
displayed in bins on the horizontal axis. While the bin labels are evenly spaced on the x-axis,
please note that the ranges within each bin are expanded on the left and right ends of the x-axis
and compressed in the middle of the x-axis. As shown in the legends, each age group is given its
own color, but the MOVES results are shown with colored lines, and the RSD emission rate
method results are shown with colored dots.
Exhaust CO2-Unlike for exhaust NOx, CO, and Total HC, MOVES does not store CO2
emission factors in the emissionratebyage or emissionrate tables (i.e. CO2 g/hr rates are not
directly available from MOVES table). However, total energy consumption, which is found in
the emissionrate table of the MOVES database, can be used to produce MOVES estimates of
CO2 release rates. The CO2 release rates are calculated from total energy consumption, fuel
carbon content, the fraction of carbon oxidized to form CO2, and the ratio of the molecular
weight of C02to the atomic mass of carbon. The resulting emission factors represent CO2 release
rates in g/hr.
Figures 10-8 and 10-9 show the comparisons of the RSD emission rate method CO2 release rates
(dots) and the MOVES CO2 release rates (lines) for LDTs and LDVs, respectively. The two
figures have similar trends. Overall, the RSD emission rate method's CO2 release rates are
roughly in same range as the MOVES rates. Also, the vertical scatter of the dots and lines among
the age groups are similar and indicate that the influence of vehicle age is about the same for the
two methods across the range of age groups. MOVES indicates that CO2 rates are about 50%
higher for 20-years-and-older vehicles than for 0-to-3-year-old vehicles. While the trend of
MOVES CO2 rates shows an increase as age increases, the RSD trend is too scattered to notice a
trend with age group. The differing VSP trends of the RSD emission rate method and MOVES is
a concern. While both have linear, upward trends with respect to the VSP bins, the RSD
emission rate method has substantially higher rates in the low VSP bins (e.g. an RSD/MOVES
ratio of 3/2 in the 3-to-6 kW/Mg bin) and substantially lower rates in the high VSP bins (e.g., an
RSD/MOVES ratio of 2/3 in the 24-to-30 kW/Mg bin) than MOVES does.
Exhaust NOx - For gasoline engines, 95% or more of NOx is NO. Therefore, we expect
minimal differences between exhaust NO and NOx emission rates. Consequently, Figures 10-10
and 10-11 compare NOx g/hr rates from MOVES against NO g/hr rates from the Westminster
dataset.
Just as for the CO2 comparisons, the figures show similar trends for MOVES and the RSD
emission rate method. The general range of MOVES NOx and RSD NO rates are the same. The
NOx/NO comparison additionally shows that the age-group order is the same for MOVES NOx
and RSD NO. The vertical scatter by age group is similar with NOx/NO release rates about 50
times lower for 0-to-3-year-old vehicles compared to 20-years-and-older vehicles. The VSP
trends for MOVES and RSD are both generally upward and linear. However, just as for CO2, the
VSP slopes are lower for the RSD emission rate method than for MOVES.
10-11
-------
Figure 10-8. LDT Exhaust CO2 Release Rates v. VSP Bin and Age Group
10000CH
3
0)
+->
re
Ct.
£ 10000
(/)
re
_a>
re
£
£ 10000
re
<0
CH
(N
O
U
10001
VSP Bin (kW/Mg)
Age Group (years) • • • rsd 00-03 • • • rsd 04-05 • • • rsd 06-07 • • • rsd 08-09
•••RSD 10-14 •••RSD 15-19 RSD 20-99 MOVES 00-03
MOVES 04-05 MOVES 06-07 MOVES 08-09 MOVES 10-14
MOVES 15-19 MOVES 20-99
/projl/EDARinDenver-OCT2019/Analysis_MLout/220113/Anal_MLout/OCT19_VSPbins_3.sas 23DEC22 15:42
10-12
-------
Figure 10-10. LDT Exhaust NOx/NO Release Rates v. VSP Bin and Age Group
1000.0
Age Group (years) • • rsd 00-03 • • • rsd 04-05 • •1 rsd 06-07 • • • rsd 08-09
•• RSD 10-14 •••RSD 15-19 RSD 20-99 MOVES 00-03
MOVES 04-05 MOVES 06-07 MOVES 08-09 MOVES 10-14
MOVES 15-19 MOVES 20-99
/proj1/EDARinDenver-OCT2019/Analysis_MLout/220113/Anal_MLout/OCT19_VSPbins_3.sas 23DEC22 15:42
VSP Bin (kW/Mg)
Figure 10-11. LDV Exhaust NOx/NO Release Rates v. VSP Bin and Age Group
1000.0
Age Group (years) • • rsd 00-03 • • • rsd 04-05 • •1 rsd 06-07 • • • rsd 08-09
•• RSD 10-14 •••RSD 15-19 RSD 20-99 MOVES 00-03
MOVES 04-05 MOVES 06-07 MOVES 08-09 MOVES 10-14
MOVES 15-19 MOVES 20-99
/projl/EDARinDenver-OCT2019/Analysis_MLout/220113/Anal_MLout/OCT19_VSPbins_3.sas 23DEC22 15:42
VSP Bin (kW/Mg)
10-13
-------
Figure 10-12. LDT Exhaust CO Release Rates v. VSP Bin and Age Group
10000
1000
Age Group (years)
VSP Bin (kW/Mg)
RSD 00-03 • ¦ • RSD 04-05 • • • RSD 06-07
RSD 10-14 • • • RSD 15-19 RSD 20-99
MOVES 04-05 — MOVES 06-07 MOVES 08-09
MOVES 15-19 MOVES 20-99
• • • RSD 08-09
MOVES 00-03
MOVES 10-14
/projl/EDARinDenver-OCT2019/Analysis_MLout/220113/Anal_MLout/OCT19_VSPbins_3.sas 23DEC22 15:42
Figure 10-13. LDV Exhaust CO Release Rates v. VSP Bin and Age Group
100001
1000
VSP Bin (kW/Mg)
Age Group (years) • • • rsd 00-03 • • • rsd 04-05 • • • rsd 06-07 • • • rsd 08-09
••¦RSD 10-14 •••RSD 15-19 RSD 20-99 MOVES 00-03
MOVES 04-05 MOVES 06-07 MOVES 08-09 MOVES 10-14
MOVES 15-19 MOVES 20-99
/projl/EDARinDenver-OCT2019/Analysis_MLout/220113/Anal_MLout/OCT19_VSPbins_3.sas 23DEC22 15:42
10-14
-------
Figure 10-14. LDT Exhaust Total HC Release Rates v. VSP Bin and Age Group
100.01
0.1
£ d?
O) y
& <§>
cy
<§>
oj
o
V
y
&
*
O)
9>
Age Group (years)
VSP Bin (kW/Mg)
RSD 00-03
RSD 10-14
MOVES 04-05
MOVES 15-19
' RSD 04-05
' RSD 15-19
MOVES 06-07
MOVES 20-99
' • RSD 06-07
RSD 20-99
MOVES 08-09
' RSD 08-09
MOVES 00-03
MOVES 10-14
/projl/EDARinDenver-OCT2019/Analysis_MLout/220113/Anal_MLout/OCT19_VSPbins_3.sas 23DEC22 15:42
Figure 10-15. LDV Exhaust Total HC Release Rates v. VSP Bin and Age Group
VSP Bin (kW/Mg)
Age Group (years) • • rsd 00-03 • • • rsd 04-05 • •1 rsd 06-07 ~ • • rsd 08-09
•• RSD 10-14 •••RSD 15-19 RSD 20-99 MOVES 00-03
MOVES 04-05 MOVES 06-07 MOVES 08-09 MOVES 10-14
MOVES 15-19 MOVES 20-99
/projl/EDARinDenver-OCT2019/Analysis_MLout/220113/Anal_MLout/OCT19_VSPbins_3.sas 23DEC22 15:42
10-15
-------
Exhaust CO - Figures 10-12 and 10-13 show the comparisons for CO for LDTs and LDVs.
Again, the general range of CO rates and the size of the age-group scatter are similar for
MOVES and the RSD emission rate method. Also, there is evidence that the order of age-groups
is the same for MOVES and for the RSD emission rate method, for example, the 0-to-3-year-olds
(light blue) are at the bottom and the 20-year-and-olders (pink) are at the top. The size of the
vertical scatter associated with vehicle age is similar for MOVES and RSD with CO rates about
10 times lower for 0-to-3-year-old vehicles compared to 20-year-and-older vehicles.
Again, the biggest concern with the comparison of CO rates between MOVES and the RSD
emission rate method is the relative twist between VSP dependencies. The evidence of the twist
is seen at high VSPs by the substantially lower RSD release rates for the RSD emission rate
method compared to MOVES. In terms of VSP trends, the MOVES lines show increasing slopes
at high VSP bins, while the RSD emission rate method dots show a more muted increase with
VSP.
Exhaust Total HC - For the comparisons of exhaust Total HC shown in Figures 10-14 and 10-
15, the RSD g/hr values are substantially above the MOVES values. In general, the MOVES
versus RSD emission rate method comparisons that we made for other pollutants are poor for
exhaust Total HC. A large part of this difference may be due to noise in the RSD HC channel.
Noise in the RSD HC channel tends to increase the quantified Total HC release rate. We have
used signal processing techniques to reduce the noise in all RSD pollutant channels, but
additional work is needed for the HC channel.
Overall, the release rate trends by the RSD emission rate method and by MOVES correlate and
indicate that the RSD emission rate method shows promise. Still, there are some caveats. In
Figures 10-8 to 10-13 for CO2, NOx/NO, and CO, and for each combination of vehicle type and
age, there is a twist in the VSP trends. That is, low-VSP RSD values tend to be higher than low-
VSP MOVES values, and high-VSP RSD values tend to be lower than high-VSP MOVES
values. At middle-VSP values (9-18 kW/Mg), there is usually good agreement between RSD
values and MOVES values.
Why the twist? - We have an explanation for the worrisome "twist" in the VSP dependencies of
the RSD-based release rates vs. the MOVES-based release rates. Since all of the graphical
comparisons of all four pollutants in Figures 10-8 through 10-13 exhibit the twist, the twist is
likely independent of pollutant. A VSP is determined for each second of vehicle operation -
whether that operation is on the dynamometer for MOVES or on the road for RSD. Additionally,
the actual measurements of the emissions by the laboratory instruments or by the RSD
instrument are also both relatively fast. The CVS quickly aspirates a sample of the tailpipe, and
therefore the measurement is an almost instantaneous measure of tailpipe concentrations and
flows. But on the road, the entrainment of tailpipe releases (and any other source of releases) into
the vortex is slow because the mass of pollutants in the vortex at any given instant comes from
the releases over about at least 2 VETs, which for a typical light-duty vehicle in a 30-mph air
flow is about 8 seconds.
For example, for a vehicle moving with a constant 4-second VET and releasing a pollutant at a
constant rate (g/hr), Figure 10-16 shows the fraction of the pollutant mass in the vortex
contributed during each previous second's release. The first previous second's release
10-16
-------
contributes only 25% of the vortex pollutant mass. Similarly, the five previous seconds together
contribute 76% of the pollutant mass, and the remaining 24% of the vortex mass is contributed
by all seconds prior to the fifth previous second. Clearly, if the release rate of a pollutant changes
during previous seconds, the portion of the mass in the vortex for each previous second will vary
from that shown in the figure. Also, if the VET is substantially larger, for example for the 53-
foot box trailer of an 18-wheeler, the vortex will contain contributions from many more previous
seconds. The point is that the mass measured by an RSD during a vehicle transit is not a measure
of the instantaneous (or even almost instantaneous) emissions release rate, emission rate, or
tailpipe concentrations. It is a measure of the weighted emissions over the most recent many
seconds.
Figure 10-16. Portions of Pollutant Mass in a 4-Second-VET Vortex
So, the mass of pollutants in the on-road vortex has "memory" of the engine operation, including
fuel rate, over a much longer time than for the dynamometer or PEMS testing.
In Figures 10-8 through 10-15 , the MOVES release rate "lines" have VSP coordinates that are
the almost instantaneous VSP of the MOVES release rate measurement. But for the RSD release
rate "dots," the VSP coordinate is simply the VSP at the time of the RSD measurement, while
the RSD release rate value is a measurement of pollutant mass in the vortex accumulated over
the previous many seconds. Since for high VSP values, the VSPs of these previous many
seconds are most likely less than the VSP coordinate value, the RSD-measured release rate is
likely to be lower than for the corresponding MOVES release rate value. The opposite is true of
low VSP values. Because of this difference of VSP basis between the MOVES and RSD release
rate values, Figures 10-8 through 10-15 artificially introduce the apparent twist when comparing
the two data sources. This twist might cause the analyst to incorrectly conclude that the MOVES
and the RSD emission rate method do not agree.
10-17
-------
If vortex entrainment explains all of the apparent MOVESvRSD twist, it is likely that the
middle-level VSP values (perhaps the 9-12 kW/Mg bin) do not have much VSP bias between the
MOVES and RSD release rate values. Accordingly, if we compare the MOVES and RSD release
rate values for just the 9-12 kW/Mg bins in Figures 10-8 through 10-15, we conclude that the
values and the range of by-age values of MOVES and RSD release rate agree well for CO2, NO,
and CO. However, for Total Hydrocarbon, the RSD values are substantially higher than the
MOVES values.
Now, we consider an alternative method for comparing MOVES and RSD release rates while
using fuel rate to account for VSP. The blue dots in the log-log plot in Figure 10-17 show the
RSD NO vs. RSD CO2 release rates calculated for 15-to-19-year-old LDVs in the Westminster
dataset. Because both the RSD NO and the RSD CO2 are obtained from RSD, they both are
measures of those pollutants in the vortex at the same instant of the RSD transit and therefore
they both are influenced in exactly the same way by vortex entrainment. Also, because the
measures are optical - not via extraction of a gas sample (as CVS or PEMS do) - the RSD NO
and RSD CO2 are perfectly time-aligned with each other, and with all other pollutants in the
vortex.
Unless the CO release rate is high, the CO2 release rate is directly proportional to the engine fuel
rate. Therefore, the highest fuel rates, and therefore, the highest engine loads are usually on the
right side of the figure. An examination of the blue dots at around 40,000 g/hr CO2 release rate
shows that some vehicles were releasing around 1 gNO/hr while others were releasing around
600 gNO/hr. Keep in mind that those NO release rates are for vehicles operating at high fuel
rates. The top edge of the blue-dot field in the figure shows the NO-release trend of the highest
releasing vehicles in this data subset.
We can use the MOVES trends of release rates as a reference for the RSD release rates.39 The
basis of the MOVES values for NOx and CO2 release rates is the same, that is, both are based on
dynamometer testing. While the basis of the MOVES values is different than the basis of RSD
values, the trends for MOVES vs. RSD can be compared. The black line in Figure 10-17 shows
the trend of the MOVES NOx vs. MOVES CO2 release rates for the MOVES opModelDs
averaged over the four LDV model years in the 15-19-year-old age group in Calendar Year 2019.
The opModelDs for the highest CO2 release rates, VSPs, and fuel rates are on the upper right end
of the black line. Note that the upper edge of the blue-dot field of RSD transit release rates is
roughly parallel to the MOVES line.
The values of the RSD NO and RSD CO2 release rates, which determine the location of each
blue dot in Figure 10-17, are calculated as described earlier using each transit's EDAR signal
divided by the estimated VET. The value of the VET is independent of pollutant, and therefore,
the same value of the VET is used to calculate both the RSD NO and the RSD CO2 release rate -
that is, both coordinates of each blue dot.
39 Note that we are not trying to evaluate MOVES quality. The idea is that if RSD trends somehow agree
with MOVES trends, then we gain comfort with the RSD results.
10-18
-------
Figure 10-17. Comparison of NO v CO2 Release Rates by RSD for Westminster 15-to-19-Year-Old LDVs
10001
1000 10000 100000
C02 Release Rate (g/hr)
Source: • • • Westminster: 2000-2004 LDVs in 2019 MOVES: Average 2000-2004 LDV in 2019
/projl/EDARinDenver-OCT2019/Analysis_MLout/220113/Anal_MLout/OCTI9_VSPbins_4_CP.sas 03MAR23 10:51
10-19
-------
Because of turbulence, the estimated VET for a given transit has some uncertainty. Figures 6-10,
6-11, and 6-14 to 6-17 indicate that the approximate size of the uncertainty in VET might be at
most -50% and +100% of the estimated value. For example, the true value of an estimated VET
of 4s might be somewhere between 2s and 8s.
Since the uncertainty affects both blue-dot coordinates by the same amount, the true location of a
blue dot would be on a 1:1 line passing through the blue-dot location in Figure 10-17. To
demonstrate the effect of VET uncertainty, Figure 10-17 is annotated with a red circle around a
selected blue dot. The blue-dot location for the release rates using the true VET would be
somewhere on the red line with the arrowheads. Thus, even taking the uncertainty of the VET
into account, the location of the selected blue dot will remain in the same general location in the
blue-dot field, that is, near the top-left angled edge of the blue-dot field. Importantly, regardless
of wherever the true blue-dot location might be along the red line, the location will be
approximately the same distance from the black MOVES reference line - because the MOVES
reference line and the red 1:1 line have approximately the same slope.
Forty-two plots comparing CO2, NO, CO, and THC, for LDVs and LDTs, and for all seven age
groups are shown in Appendix F. Overall, the results for NO vs. CO2, and CO vs. CO2 suggest
that the RSD emission rate method provides results that are consistent with the trends modeled
by MOVES by vehicle class, vehicle age, and fuel rate. The THC results for the RSD results are
not consistent with MOVES modeled THC values. We suspect that the RSD values are elevated
either by noise or by a calculation error. We do not suspect the MOVES values.
Because the MOVES and RSD results in Figure 10-17 are consistent with each other, we
wondered whether the vertical distance of a transit's blue point from the MOVES black line
could be used to determine if the vehicle was a high or low emitter. To evaluate that possibility,
we overlaid the simulated second-by-second RSD NO vs. RSD CO2 results from the Kansas City
2001 Forester high-emitter (5.8 gNO/mile) from Figure 10-5 and the 2002 Trail Blazer low-
emitter (0.3 gNO/mile) from Figure 10-7 onto the plot shown in Figure 10-17 to create Figure
10-18. The overlay is appropriate since both the Forester and the Trail Blazer would be 15-19-
year-olds in 2019.
Figure 10-18 shows that most of the red pluses for the 2001 Forester fall at the top edge of the
blue-dot field of Westminster 15-19-year-old LDVs. The lime-green pluses for the Trail Blazer
fall much lower in the figure and are near the MOVES black line. The idea is that if either the
Forester or the Trail Blazer would pass under the EDAR instrument at any second of their PEMS
time series, then the simulated RSD NO vs. RSD CO2 release rates would almost always fall in a
location in Figure 10-18 that would correctly characterize them as either a high or low NO
emitter.
10-20
-------
Figure 10-18. Comparison of NO v CO2 Release Rates for Datasets of 15-to-19-Year-Old LDVs
1000
U)
£
"fS
£
++
+ +++++
+ + +^-rM&- -h. * +++J. + + + + + +
+v * V*++^+ltT + •
+. +- +• *
+++ + .+ + * • * *
++ + ++ • + + +++
t+ +U.+ + + . + H+ +. +++
-+* * * • + +J..+ +* * + 4+ %
+*+ +,% 4
4- +#. ++ + ++ ++ +
+ + 4- ++ /*+ **-
+++n+*++ ++t < V
+ \++ + + ++ + +
4- *+ + + + <- + + ^ *
~ + ^i+++++\+* ¦ ' v.
+ +4- + + + ++•
\ + + t+4. + .
+ + ++ +++ •+.
+ 44 4- + 44 4- 4- .
v • +
+ + • +++ • + <¦ • 1 ¦ .
+- + + -t- + *•
+¦ +¦ +
+ -1- + * • * «
4- +• T + • ¦
+¦ + +¦ + + + ¦
+ •)- 4 + + -pl-
+ + +
+\+;.. v-A+\ /Cfn
+• + *v% . + ++ #+ +++ + + + +++++ v-
•+ +* " +*44^* +, ++±* * /
. + 4- + . + \ * *+* + J
+ . + t±lv£ +++ +*-/ #+++
h ++ . • • +4- 4- + + +J "+ *»W4>*fr + + « +
; • +^ +
*• + ' ++^
•++ 4-+* + ileitis*+v*
• +«-• -t*+f I**Ss Vy^+H4++#f+++f , +
+ 1 ++ + ++"1" + + +
+; - •. * -t
* + + + ++a+-i-+ + J+
+¦ + 4+ ++ + +
++ + ++ + +
* + + + " +#++J"_L+ej
+ • + + "^i
+ + •+*••++ +++ +Jr\
• i . • *'
+ • • + + *+ - * + * y \ ++H
+ **..+/ VV*1
* ++ + +/
•+ +*+¦>¦'+ ^ S+tX£+
• m • m • + +Y1F-+++ #"
• • % + ,+ -1/ ±7++. • -%
f + +. V * •;. •
« • • a % 1 • + • +4 -i®"
• • • * * ii+++
• * * • .t+ !%?¦
* * * • + + + ++ + * +++ +rfc^S»« •
A**\ +»S^4-^+i-#Cw+^tri^4 * ++ ** +
i+:^**4.+V++ •••• * ' v "++
v%H V.. \ ; .
¦+*|- «.•++ +H-4* t1/. « ++ -?+ ® •
*^^4-+ • + "
• ',/^T ' •« if1-$S * +
1000
10000
C02 Release Rate (g/hr)
100000
Source: • • • Westminster: 2000-2004 LDVs in 2019
+ + + Kansas City: 2001 Forester (5.8 gNO/mile)
MOVES: Average 2000-2004 LDV in 2019
+ + + Kansas City: 2002 Trail Blazer (0.3 gNO/mile)
/projl/E DARinDenver-OCT2019/Analysis_M Lout/220113/Anal_M Lout/OCTI 9_VS Pbins_4_CP.sas 03MAR23 10:51
10-21
-------
Careful examination of Figure 10-18 shows that, for a small fraction of the seconds, some red
pluses fall below where the bulk of the Forester symbols are located. This suggests that
occasionally, the history of engine operation just before the RSD transit could cause the RSD-
observed NO release rate to be somewhat lower than what the general NO release rate for the
vehicle might more typically be. Similarly, but in the other direction, a few lime-green pluses
appear at higher NO release rates than the bulk of symbols for the Trail Blazer. If the vehicle got
an RSD during one of those seconds, it would appear to be a somewhat higher NO emitter than it
is most of the time. Further analysis of the PEMS time series might help eliminate some of these
classification errors through selected siting of the RSD instrument.
10.3 Comparison of the Traditional RSD Concentration Method with the RSD
Emission Rate Method
This subsection uses the detailed data obtained during an example Westminster transit to
compare the traditional RSD method used to calculate exhaust emissions concentrations with the
new RSD method to calculate mass emissions rate (g/mile).
As an overview, traditional calculations can be described by the schematic in Figure 10-19. The
calculations start with the RSD detailed data for CO2, CO, NO, and HC on the left. The method
then assumes that all pollutants are exhaust pollutants emitted at the tailpipe, which leads to the
expectation that all exhaust pollutants are well mixed and diffuse and disperse together from the
tailpipe exit.40 Consequently, the optical masses of all pollutants should be directly proportional
to each other, which means that plots of the detailed data for pairs of pollutants will produce
lines passing through the origin.
To demonstrate and compare the RSD methods, we will use the transit (Series=515,
Transit=2469, 10/25/19:13:27:23.6) ofa2001 pickup truck that happened to drive by. Figure 10-
20 shows the pairwise detailed data plots against CO2 and regression fits for this transit. The blue
and orange trends in the top two plots for NO vs. CO2 and CO vs. CO2 are close to linear, have
positive slopes, and pass near the origin. The green points in the lower left plot show the trend
for the RSD HC channel, which reports measurements of Total HC, that is, for ExhHC plus
EvapHC. The apparent41 noise in this plot makes it difficult to determine if the trend is linear or
not, but the linear regression indicates that the trend passes near the origin, and if the trend is
linear, it has a negative slope.
40 Gases with different molecular weights actually would diffuse at different rates. For example, C02
(MW=44) would diffuse more slowly than CO (MW=28). However, given the time scale of each transit's
RSD detailed data collection (about 0.5s) and the accuracy of RSD detailed data measurements, the
assumption of equal diffusion rates for all gases is reasonable.
41 We say "apparent" because we will see that a large part of the apparent noise is actually from EvapHC.
10-22
-------
Figure 10-19. Diagram of Traditional RSD Method for Calculating Exhaust Concentration from Detailed Data
10-23
-------
Figure 10-20. Pairwise Optical Mass Plots for the Example Transit (Series=515, Transit=2469)
0.009
fM
E
o
a>
_D
>
O
U
U
><
before HC separation:
total HC vs C02 has
negative slope
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
Pixel C02 Value (mole/m2)
-------
The traditional RSD method then uses the regression slopes of CO, NO, and HC against the CO2
to estimate exhaust emissions concentrations - if the vehicle is assumed to be a
stoichiometrically operating engine, for example, a gasoline engine. To perform the
concentration calculations for this example, we will use the algebraic equation that ESP has used
in the past42 for its Accuscan 4600 RSD instrument:
ppmCO = Slope C0vC02 * ppmC02
ppmNO = Slope N0vC02 * ppmC02 Equation 10-1
ppmHC = Slope HCvC02 * ppmC02
where ppmC02 =
150538
(1 + 0.7168 * Slope COvC02 + 0.3584 * Slope NOvC02 + 0.3011 * Slope HCvC02)
Following the traditional RSD method, the slopes of the regression lines shown in the first three
subplots (blue, orange, green) of Figure 10-20 were used with Equation 10-1 to calculate the
estimated exhaust concentrations shown in the right column of Figure 10-21. The negative
estimated concentration for ExhHC is a concern. An incorrect interpretation of this value would
be that the vehicle has a near-zero exhaust HC emissions concentration. The negative value is the
result of the negative regression slope of the green points in Figure 10-20.
The left column of Figure 10-21 shows the plume heatmaps of the four RSD channels as
observed by the RSD from above the pavement. Those heatmaps clearly show that CO2, CO, and
NO have plumes of the same shape, but the HC heatmap shows that the HC plume is
substantially different from the others. We conclude that most of the HC mass cannot be exhaust
HC, otherwise the HC heatmap would have the same shape as the heatmaps for CO2, CO, and
NO. Thus, although there may be some evidence of ExhHC in the HC heatmap, we claim that
most of the HC must be EvapHC. Since the heatmaps indicate that EvapHC, as well as some
ExhHC, is present, the traditional RSD assumption that all pollutants are from the tailpipe is
violated. We cannot be certain that the calculated value of-184 ppm ExhHC is correct.
42 T.H. DeFries, J.H. Lindner, C.F. Palacios, S. Kishan, "Investigation of RSD for High Evaporative
Emissions Vehicle Detection: Denver Summer 2008 Pre-Testing Study," Version 1, EPA-090306,
prepared for U.S. Environmental Protection Agency, March 6, 2009.
10-25
-------
Figure 10-21, Traditional RSD Method Exhaust Concentration Results
for the Example Transit
Detailed Data
CO.
CO
NO
Appears to be Clean HC
Est. Exhaust Cone.
14.95%
0.13%
353 ppm
-184 ppm
10-26
-------
The next step in the analysis of the example transit is to use the Blind Source Separation method
to split the RSD HC detailed data into an ExhHC portion and an EvapHC portion. This BSS
converted the green Total HC heatmap at the bottom left of Figure 10-21 into the purple ExhHC
heatmap and the red EvapHC heatmaps at the bottom left of Figure 10-22. The re-calculated
exhaust concentrations are estimated using the same Equation 10-1 but now the regression slopes
are obtained from the blue NO vs. CO2, the orange CO vs. CO2, and the purple ExhHC vs. CO2
shown in Figure 10-20. The resulting exhaust concentrations are listed in the right column of
Figure 10-22. The re-calculated CO2, CO, and NO concentrations are virtually the same values
as originally calculated and shown in Figure 10-21 and the second column of Figure 10-22.
However, the new value of ExhHC is +74 ppm, which is a much more reasonable value than the
original -184 ppm value. Overall, we believe that the concentration results shown in Figure 10-
22 are reasonably accurate concentration results for the exhaust emissions for the example
transit.
Of course, since concentration has no meaning for EvapHC, the value is listed as "n/a" in Figure
10-22. Still, we want to have some measure of the EvapHC emissions - especially since the
green HC heatmap in Figure 10-21 indicates that the EvapHC emissions are substantially larger
than the ExhHC emissions. We will use the RSD emission rate method to estimate the magnitude
of the EvapHC emissions rate (g/mile).
The schematic shown in Figure 10-23 shows an overview of the calculation using the RSD
emission rate method. Just as for the schematic for the traditional RSD concentration method
shown in Figure 10-19, the flow begins on the left with the detailed RSD data. The first step is
the application of the BSS separation method to effectively split the Total HC detailed data into
ExhHC detailed data and EvapHC detailed data. Then, following Equation 5-19, the RSD Signal
(g) from each of the RSD channels is simply divided by the RSD instrument's 100%
Illumination Speed (mile/hr) and the Vortex Entrainment Time (hr) to produce each pollutant's
Emission Rate (g/mile).
The last column of Figure 10-24 shows the resulting mass emission rates (g/mile) for the five
pollutants. The estimated EvapHC emission rate of 8.6 g/mile is over 100 times the EvapHC
running loss emission standard of 0.05 g/mile and more than 10 times the estimated ExhHC
emission rate of 0.8 g/mile. This is an example where the traditional RSD method had failed to
identify this vehicle as an elevated HC emitter. The fundamental cause of the erroneous
conclusion was that the emission circumstances did not satisfy the assumptions made by the
traditional RSD method.
It's important to recognize that the new RSD emission rate method and the improved results use
the same detailed, internal data that all RSD instruments are collecting.
10-27
-------
Figure 10-22. Re-Calculation of Exhaust Concentrations after separating ExhHC from EvapHC Plumes
for the Example Transit
Detailed Data
CO.
CO
NO
ExhHC
EvapHC
Traditional
ReCalc Result
14.94 %
0.13%
353 ppm
74 ppm
n/a
11
10-28
-------
Figure 10-23. Diagram of RSD Method for Calculating Mass Emission Rates (g/mile) from Detailed Data
RSD
Detailed Data
C02
CO
NO
HC
BSS
-h7oo%is I
split
ExhHC
EvapHC
VET
Emission Rate
(a/mile)
C02
CO
NO
ExhHC
EvapHC
10-29
-------
Figure 10-24. Calculation of Mass Emission Rates (g/mile)
for the Example Transit
t ~ 1
|A j j
Massive EvapHC
Detailed Data
Traditional
ReCal^Result
Emission Rate
C02
14.95%
14J^ %
546. g/mile
CO
0.13%
0.13^0
12.3 g/mile
NO
353 ppiri
353 pprrr
, 3.1 g/mile
ExhHC
-184 ppm
74 pprn
\ 0.8 g/mile
n/a
n/a
\
¦0 EvapHC
\8.6 g/mile
10-30
-------
10.4 Characterization of NO Mass in the Westminster Fleet Sample
Figures 10-10 and 10-11 in Section 10-2 showed that MOVES NOx release rates (g/hr) and the
RSD emission rate method NO release rates (g/hr) were comparable over a wide range of VSP
bins and vehicle age bins. Accordingly, this subsection uses the Westminster sample's NO
measurements to demonstrate the type of fleet analysis that can be performed using mass
emission rate (g/mile) measures that the RSD emission rate method can provide.
We used the RSD emission rate method to calculate the NO emission rates (g/mile) for the
transits of fleet vehicles that drove past the RSD instrument. The left side of Figure 10-25 shows
that the distribution of the number of vehicles has a peak at around 0.03 g/mile. In this plot,
equal areas represent equal numbers of vehicles. Therefore, the plot shows that the youngest
vehicles (blue) have low NO emission rates and the oldest vehicles (orange + red) have higher
average emission rates. The areas also indicate that a large part of the fleet is made up of vehicles
less than 16 years old (blue + green + yellow). The vertical reference line at 1 gNO/mile shows
that only a small fraction of the fleet vehicles has NO emission rates above 1 gNO/mile.
The right side of Figure 10-25 shows the distribution of NO mass for the same analysis set of
transits. Equal areas on this plot represent equal NO mass. The plot was produced by summing43
the calculated NO emission rate (g/mile) for every transit in the analysis set. The right plot shows
that even though large numbers of vehicles are present at low NO emission rates on the left plot,
they do not contribute much NO mass to the fleet emissions. For example, the fraction of the
fleet NO mass below 1 gNO/mile of the right plot is much smaller than the fraction of vehicles
with NO emission rates below 1 gNO/mile on the left plot.
These two plots help us focus on the portion of the fleet that produces the largest part of the fleet
emissions problem. In Figure 10-26 the shaded areas on the left and right plots show that
vehicles 11 years and older (yellow + orange + red) make up only about 8% of the fleet but
contribute 64% of the fleet NO mass. On the other hand, vehicles less than 11 years old (blue +
green) contribute a small portion of the fleet NO mass.
These observations are really nothing new to the mobile sources emissions research community.
What is new is the ability of RSD measurements to provide data that can quantify the mass
distributions.
It is important to recognize that, just like for any RSD measurement, using measured emissions
values from 1-second snapshots of many vehicles to infer the mass of emissions of the entire
fleet is an extrapolation. The extrapolation is affected by using a single RSD instrument location
and by how the visual presence of the RSD equipment may affect how drivers operate their
vehicles as they drive past the RSD. For example, some drivers may take their foot off the
accelerator, which will produce a small exhaust plume and thereby greatly affect the emission
rates (g/mile) of all pollutants during the transit.
43 We assumed that all vehicles in the analysis set have equal annual mileage accumulation.
10-31
-------
Figure 10-25. Distributions of Number of Vehicles and NO Mass by NO Emission Rate and Age
0.01
Number of
Vehicles
0.10
10
0.01
Vehicle In-Use NO (g/mile)
Age Group
0-3
i4-10
11-16
17-24
>24
10-32
-------
Figure 10-26. Fleet Fraction that Produces Most NO Mass
Number of
Vehicles
0.01
Vehicle In-Use NO (g/mile)
Age Group hO-3 h4-10 11-16 17-24 h>24
10-33
-------
Appendix A:
Test Vehicle EDAR Exhaust Emissions Measurements1
1 P:\EDARinDenver-OCT2019\Analysis/refVeh_out.xlsx
A-0
-------
Convoy
Position -
Reference
Vehicle ID
Nominal
Vehicle
Speed
(mph)
Exhaust
HC
(ppmC6)
Exhaust
CO
(ppm)
Exhaust
NO
(ppm)
Exhaust
CO2
(ppm)
Transit
Date
Transit
Time
1-EV1
22.5
32
-3
-2
154535
10/20/2019
11:54:55 AM
1-EV1
22.5
-35
-7
0
154542
10/20/2019
12:38:32 PM
1-EV1
22.5
46
-27
-19
154496
10/20/2019
1:18:45 PM
1-EV1
22.5
-29
-1
2
154543
10/20/2019
3:51:05 PM
1-EV1
22.5
-71
-9
10
154527
10/20/2019
4:09:15 PM
1-EV1
22.5
-42
4
1
154541
10/21/2019
9:37:09 AM
1-EV1
22.5
-135
-4
-1
154512
10/21/2019
10:21:06 AM
1-EV1
22.5
-3
3
8
154552
10/21/2019
11:25:10 AM
1-EV1
22.5
-9
2
0
154554
10/21/2019
12:37:08 PM
1-EV1
22.5
-33
9
9
154534
10/21/2019
1:03:07 PM
1-EV1
22.5
40
-6
0
154522
10/22/2019
10:49:03 AM
1-EV1
22.5
4
1
2
154555
10/22/2019
11:43:22 AM
1-EV1
22.5
-87
3
12
154520
10/22/2019
12:30:50 PM
1-EV1
22.5
16
3
1
154543
10/22/2019
3:29:19 PM
1-EV1
22.5
30
3
-2
154526
10/22/2019
4:05:13 PM
1-EV1
22.5
234
38
16
154372
10/23/2019
9:00:15 AM
1-EV1
22.5
78
25
-32
154468
10/23/2019
9:46:50 AM
1-EV1
22.5
0
67
-21
154398
10/23/2019
10:30:44 AM
1-EV1
22.5
185
-4
4
154414
10/23/2019
11:56:44 AM
1-EV1
22.5
0
3
7
154427
10/23/2019
12:30:47 PM
1-EV1
22.5
498
-12
-15
154231
10/23/2019
1:48:51PM
1-EV1
22.5
-111
15
-1
154469
10/23/2019
2:34:46 PM
1-EV1
22.5
-14
-2
-5
154546
10/24/2019
2:16:35 PM
1-EV1
22.5
-60
2
-1
154533
10/24/2019
2:36:48 PM
1-EV1
22.5
19
1
-1
154546
10/24/2019
3:09:12 PM
1-EV1
45
-73
-14
16
154516
10/20/2019
12:13:42 PM
1-EV1
45
54
2
-2
154515
10/20/2019
12:55:51PM
1-EV1
45
24
-8
-9
154523
10/20/2019
1:36:09 PM
1-EV1
45
-14
-5
24
154516
10/20/2019
3:09:38 PM
1-EV1
45
226
0
-57
154378
10/20/2019
3:28:05 PM
1-EV1
45
47
2
45
154494
10/21/2019
11:04:02 AM
1-EV1
45
0
44
4
154378
10/21/2019
11:44:02 AM
1-EV1
45
-11
-5
7
154506
10/21/2019
12:16:02 PM
1-EV1
45
0
-12
2
154481
10/22/2019
11:07:49 AM
1-EV1
45
88
7
0
154494
10/22/2019
12:03:51PM
1-EV1
45
109
25
16
154465
10/22/2019
12:49:50 PM
1-EV1
45
164
-1
-10
154450
10/22/2019
2:33:49 PM
1-EV1
45
158
-48
-10
154411
10/22/2019
2:53:50 PM
1-EV1
45
791
0
49
154009
10/23/2019
3:44:37 PM
1-EV1
45
-53
27
-36
154479
10/24/2019
1:31:43 PM
1-EV1
45
0
80
3
154395
10/24/2019
1:55:44 PM
1-EV1
45
232
8
1
154384
10/24/2019
3:28:04 PM
2-EV2
22.5
178
4340
772
150967
10/20/2019
11:54:58 AM
2-EV2
22.5
112
5399
922
150173
10/20/2019
12:38:36 PM
A-l
-------
Convoy
Position -
Reference
Vehicle ID
Nominal
Vehicle
Speed
(mph)
Exhaust
HC
(ppmC6)
Exhaust
CO
(ppm)
Exhaust
NO
(ppm)
Exhaust
CO2
(ppm)
Transit
Date
Transit
Time
2-EV2
22.5
149
5294
965
150214
10/20/2019
1:18:49 PM
2-EV2
22.5
147
5237
948
150262
10/20/2019
3:51:12 PM
2-EV2
22.5
193
5246
932
150229
10/20/2019
4:09:20 PM
2-EV2
22.5
159
5041
881
150425
10/21/2019
9:37
13 AM
2-EV2
22.5
171
4818
833
150596
10/21/2019
10:21
11 AM
2-EV2
22.5
170
5085
886
150379
10/21/2019
11:25
14 AM
2-EV2
22.5
148
5411
919
150147
10/21/2019
12:37:14 PM
2-EV2
22.5
169
5304
898
150215
10/21/2019
1:03:13 PM
2-EV2
22.5
150
5357
895
150194
10/22/2019
10:49:08 AM
2-EV2
22.5
271
5558
941
149938
10/22/2019
11:43:26 AM
2-EV2
22.5
218
5529
925
150010
10/22/2019
12:30:54 PM
2-EV2
22.5
241
5674
940
149886
10/22/2019
3:29:26 PM
2-EV2
22.5
160
5565
1057
149998
10/22/2019
4:05:20 PM
2-EV2
22.5
141
5440
1025
150119
10/23/2019
9:00:19 AM
2-EV2
22.5
185
5585
1032
149983
10/23/2019
9:46:55 AM
2-EV2
22.5
-3
5505
1023
150051
10/23/2019
10:30:47 AM
2-EV2
22.5
226
5517
1020
150016
10/23/2019
11:56:51 AM
2-EV2
22.5
80
5349
977
150232
10/23/2019
12:30:53 PM
2-EV2
22.5
143
5492
1052
150068
10/23/2019
1:48:55 PM
2-EV2
22.5
409
5551
1017
149880
10/23/2019
2:34:51PM
2-EV2
22.5
98
5607
1060
150017
10/23/2019
3:21:18 PM
2-EV2
22.5
205
5339
940
150154
10/24/2019
2:16:42 PM
2-EV2
22.5
112
5232
932
150292
10/24/2019
2:36:54 PM
2-EV2
22.5
144
5051
885
150424
10/24/2019
3:09:17 PM
2-EV2
45
163
4558
812
150794
10/20/2019
12:13:45 PM
2-EV2
45
55
4583
816
150832
10/20/2019
12:55:54 PM
2-EV2
45
76
5166
970
150346
10/20/2019
1:36:12 PM
2-EV2
45
424
5069
861
150242
10/20/2019
3:09:42 PM
2-EV2
45
302
5110
898
150271
10/20/2019
3:28:11PM
2-EV2
45
247
5210
907
150225
10/21/2019
9:56:06 AM
2-EV2
45
233
5132
873
150309
10/21/2019
11:04:05 AM
2-EV2
45
0
4070
531
151155
10/21/2019
11:44:05 AM
2-EV2
45
183
5069
861
150335
10/21/2019
12:16:07 PM
2-EV2
45
0
4519
805
149530
10/21/2019
2:00:03 PM
2-EV2
45
123
5235
875
150288
10/22/2019
11:07:52 AM
2-EV2
45
163
5699
961
149907
10/22/2019
12:03:55 PM
2-EV2
45
237
5460
909
150051
10/22/2019
12:49:54 PM
2-EV2
45
331
5647
907
149852
10/22/2019
2:33:54 PM
2-EV2
45
0
5359
869
150146
10/22/2019
2:53:55 PM
2-EV2
45
285
5371
1029
150076
10/23/2019
9:21:55 AM
2-EV2
45
175
5428
1006
150107
10/23/2019
10:05:54 AM
2-EV2
45
233
5443
1012
150070
10/23/2019
10:49:53 AM
2-EV2
45
554
4053
624
150777
10/23/2019
12:33:54 PM
2-EV2
45
-75
5236
1029
150319
10/23/2019
1:25:53 PM
A-2
-------
Convoy
Position -
Reference
Vehicle ID
Nominal
Vehicle
Speed
(mph)
Exhaust
HC
(ppmC6)
Exhaust
CO
(ppm)
Exhaust
NO
(ppm)
Exhaust
CO2
(ppm)
Transit
Date
Transit
Time
2-EV2
45
94
5076
900
150425
10/24/2019
1:31:48 PM
2-EV2
45
175
5249
925
150247
10/24/2019
1:55:49 PM
2-EV2
45
381
5030
896
150289
10/24/2019
3:28:08 PM
3-F150
22.5
1
246
36
154278
10/23/2019
9:00:24 AM
3-F150
22.5
-13
-6
17
154527
10/23/2019
9:47:00 AM
3-F150
22.5
-160
45
4
154412
10/23/2019
10:10:59 AM
3-F150
22.5
2
134
38
154411
10/23/2019
10:30:52 AM
3-F150
22.5
2
18
20
154461
10/23/2019
11:56:56 AM
3-F150
22.5
0
6
18
154530
10/23/2019
1:49:00 PM
3-F150
22.5
-10
421
83
154166
10/23/2019
2:34:55 PM
3-F150
22.5
-19
10
59
154499
10/23/2019
2:59:01PM
3-F150
22.5
5
5
20
154532
10/23/2019
3:21:22 PM
3-F150
45
17
137
42
154405
10/23/2019
9:22:00 AM
3-F150
45
6
19
54
154511
10/23/2019
9:26:02 AM
3-F150
45
-46
109
35
154415
10/23/2019
10:49:57 AM
3-F150
45
-83
86
50
154404
10/23/2019
12:33:58 PM
3-F150
45
1
28
-12
154328
10/23/2019
1:25:58 PM
3-F150
45
-69
44
23
154494
10/23/2019
2:07:57 PM
3-F150
45
0
11
7
154509
10/23/2019
2:11:55 PM
3-F150
45
6
43
9
154491
10/23/2019
2:53:56 PM
3-F150
45
0
73
12
154474
10/23/2019
3:44:44 PM
3-GMC
22.5
-1
58
2
154489
10/22/2019
10:49:13 AM
3-GMC
22.5
0
558
5
154136
10/22/2019
4:05:26 PM
3-GMC
45
1
131
9
154454
10/22/2019
11:07:56 AM
3-GMC
45
-1
129
-2
154453
10/22/2019
11:11:59 AM
3-Subaru
22.5
4
2651
416
152368
10/20/2019
11:55:01 AM
3-Subaru
22.5
-112
96
13
154431
10/20/2019
12:38:39 PM
3-Subaru
22.5
-9
250
23
154335
10/20/2019
1:18:52 PM
3-Subaru
22.5
6
16
9
154486
10/20/2019
3:51:18 PM
3-Subaru
22.5
1
22
7
154532
10/20/2019
4:09:24 PM
3-Subaru
22.5
-14
639
100
153949
10/21/2019
9:37:15 AM
3-Subaru
22.5
-45
315
57
154244
10/21/2019
10:21:14 AM
3-Subaru
22.5
6
-7
-2
154478
10/21/2019
11:01:13 AM
3-Subaru
22.5
-9
458
80
154167
10/21/2019
11:25:18 AM
3-Subaru
22.5
-23
39
3
154518
10/21/2019
12:37:20 PM
3-Subaru
22.5
-27
266
35
154317
10/21/2019
1:03:19 PM
3-Subaru
45
-24
551
81
154100
10/20/2019
12:13:48 PM
3-Subaru
45
15
439
22
154143
10/20/2019
12:17:26 PM
3-Subaru
45
-174
352
52
154259
10/20/2019
12:55:56 PM
3-Subaru
45
-73
401
52
154221
10/20/2019
1:36:14 PM
3-Subaru
45
4
376
3
154127
10/20/2019
3:09:46 PM
3-Subaru
45
-22
-2
-27
154499
10/20/2019
3:28:15 PM
3-Subaru
45
1
556
43
154112
10/21/2019
9:56:09 AM
3-Subaru
45
1
79
9
154425
10/21/2019
10:00:08 AM
A-3
-------
Convoy
Position -
Reference
Vehicle ID
Nominal
Vehicle
Speed
(mph)
Exhaust
HC
(ppmC6)
Exhaust
CO
(ppm)
Exhaust
NO
(ppm)
Exhaust
CO2
(ppm)
Transit
Date
Transit
Time
3-Subaru
45
7
608
-43
154067
10/21/2019
11:04:07 AM
3-Subaru
45
-8
918
-1
153867
10/21/2019
12:16:12 PM
3-Subaru
45
-455
321
-10
154246
10/21/2019
2:00:07 PM
4-Infiniti
22.5
0
48
6
154481
10/20/2019
11:40:01 AM
4-Infiniti
22.5
-1
414
36
154197
10/20/2019
11:47:20 AM
4-Infiniti
22.5
9
176
15
154372
10/20/2019
11:50:58 AM
4-Infiniti
22.5
-2
54
14
154446
10/20/2019
11:55:04 AM
4-Infiniti
22.5
0
114
1
154344
10/20/2019
12:22:09 PM
4-Infiniti
22.5
-1
17
10
154536
10/20/2019
12:25:51PM
4-Infiniti
22.5
-7
15
5
154512
10/20/2019
12:29:30 PM
4-Infiniti
22.5
8
282
5
154334
10/20/2019
12:33:12 PM
4-Infiniti
22.5
4
136
3
154414
10/20/2019
12:38:42 PM
4-Infiniti
22.5
4
275
4
154349
10/20/2019
1:00:42 PM
4-Infiniti
22.5
1
399
4
154251
10/20/2019
1:04:23 PM
4-Infiniti
22.5
0
595
5
154101
10/20/2019
1:08:08 PM
4-Infiniti
22.5
0
230
5
154375
10/20/2019
1:11:44 PM
4-Infiniti
22.5
0
261
1
154334
10/20/2019
1:15:20 PM
4-Infiniti
22.5
-62
32
0
154519
10/20/2019
1:18:54 PM
4-Infiniti
22.5
6
2025
8
153051
10/20/2019
2:08:38 PM
4-Infiniti
22.5
-4
190
2
154414
10/20/2019
2:27:17 PM
4-Infiniti
22.5
9
656
5
154062
10/20/2019
2:36:01PM
4-Infiniti
22.5
0
222
4
154385
10/20/2019
2:45:50 PM
4-Infiniti
22.5
2
254
14
154354
10/20/2019
2:54:24 PM
4-Infiniti
22.5
0
-10
-26
154527
10/20/2019
3:01:40 PM
4-Infiniti
22.5
0
88
1
154490
10/20/2019
3:21:32 PM
4-Infiniti
22.5
4
1140
6
153710
10/20/2019
3:25:35 PM
4-Infiniti
22.5
1
64
4
154498
10/20/2019
3:44:03 PM
4-Infiniti
22.5
-80
439
3
154220
10/20/2019
3:47:38 PM
4-Infiniti
22.5
0
236
1
154384
10/20/2019
3:51:21PM
4-Infiniti
22.5
-8
135
4
154450
10/20/2019
4:02:13 PM
4-Infiniti
22.5
0
464
2
154207
10/20/2019
4:05:50 PM
4-Infiniti
22.5
4
248
4
154350
10/20/2019
4:09:29 PM
4-Infiniti
22.5
1
144
7
154399
10/21/2019
9:21:17 AM
4-Infiniti
22.5
-9
156
6
154416
10/21/2019
9:25:18 AM
4-Infiniti
22.5
-1
205
8
154389
10/21/2019
9:29:18 AM
4-Infiniti
22.5
-7
320
15
154308
10/21/2019
9:33:15 AM
4-Infiniti
22.5
-1
105
5
154457
10/21/2019
9:37:18 AM
4-Infiniti
22.5
-7
219
6
154304
10/21/2019
10:05:13 AM
4-Infiniti
22.5
-4
168
8
154421
10/21/2019
10:13:13 AM
4-Infiniti
22.5
-1
346
14
154288
10/21/2019
10:17:16 AM
4-Infiniti
22.5
0
393
4
154267
10/21/2019
10:21:17 AM
4-Infiniti
22.5
-2
470
3
154205
10/21/2019
11:01:16 AM
4-Infiniti
22.5
-1
494
31
154134
10/21/2019
11:09:20 AM
4-Infiniti
22.5
-3
565
8
154135
10/21/2019
11:13:20 AM
A-4
-------
Convoy
Position -
Reference
Vehicle ID
Nominal
Vehicle
Speed
(mph)
Exhaust
HC
(ppmC6)
Exhaust
CO
(ppm)
Exhaust
NO
(ppm)
Exhaust
CO2
(ppm)
Transit
Date
Transit
Time
4-Infiniti
22.5
0
160
7
154429
10/21/2019
11:17:17 AM
4-Infiniti
22.5
-8
103
-1
154467
10/21/2019
11:21:21 AM
4-Infiniti
22.5
-8
405
6
154247
10/21/2019
11:25:20 AM
4-Infiniti
22.5
0
231
5
154376
10/21/2019
12:09:17 PM
4-Infiniti
22.5
-8
894
7
153891
10/21/2019
12:13:22 PM
4-Infiniti
22.5
0
594
4
154109
10/21/2019
12:29:23 PM
4-Infiniti
22.5
-3
7
5
154542
10/21/2019
12:33:08 PM
4-Infiniti
22.5
0
522
4
154169
10/21/2019
12:37:23 PM
4-Infiniti
22.5
0
207
28
154379
10/21/2019
12:51:21PM
4-Infiniti
22.5
0
123
-1
154465
10/21/2019
12:55:23 PM
4-Infiniti
22.5
-4
215
4
154387
10/21/2019
1:03:22 PM
4-Infiniti
22.5
-15
589
10
154088
10/21/2019
1:11:24 PM
4-Infiniti
22.5
-1
178
3
154424
10/21/2019
1:19:20 PM
4-Infiniti
22.5
0
104
7
154458
10/21/2019
1:47:19 PM
4-Infiniti
22.5
-4
1001
27
153694
10/21/2019
1:57:27 PM
4-Infiniti
22.5
0
734
3
154013
10/21/2019
2:05:18 PM
4-Infiniti
22.5
0
76
-2
154462
10/22/2019
10:33:15 AM
4-Infiniti
22.5
0
152
0
154008
10/22/2019
10:37:12 AM
4-Infiniti
22.5
0
99
3
154467
10/22/2019
10:41:12 AM
4-Infiniti
22.5
-2
140
6
154429
10/22/2019
10:45:16 AM
4-Infiniti
22.5
-8
98
3
154457
10/22/2019
10:49:16 AM
4-Infiniti
22.5
-2
126
1
154393
10/22/2019
11:23:05 AM
4-Infiniti
22.5
-7
307
8
154314
10/22/2019
11:27:04 AM
4-Infiniti
22.5
-4
331
-4
154282
10/22/2019
11:31:21 AM
4-Infiniti
22.5
0
179
1
154423
10/22/2019
11:39:17 AM
4-Infiniti
22.5
-16
172
18
154415
10/22/2019
11:43:34 AM
4-Infiniti
22.5
-1
128
3
154463
10/22/2019
12:09:13 PM
4-Infiniti
22.5
-1
72
0
154495
10/22/2019
12:14:57 PM
4-Infiniti
22.5
-11
141
6
154442
10/22/2019
12:19:00 PM
4-Infiniti
22.5
-2
269
-2
154355
10/22/2019
12:22:58 PM
4-Infiniti
22.5
-9
38
6
154523
10/22/2019
12:26:59 PM
4-Infiniti
22.5
-1
759
19
153967
10/22/2019
12:31:02 PM
4-Infiniti
22.5
-5
326
3
154288
10/22/2019
1:51:02 PM
4-Infiniti
22.5
-12
124
3
154458
10/22/2019
1:59:03 PM
4-Infiniti
22.5
-3
461
4
154215
10/22/2019
2:07:08 PM
4-Infiniti
22.5
-5
161
12
154412
10/22/2019
2:15:03 PM
4-Infiniti
22.5
-40
636
36
154066
10/22/2019
2:23:02 PM
4-Infiniti
22.5
0
212
0
154388
10/22/2019
2:31:07 PM
4-Infiniti
22.5
2
1252
1
153611
10/22/2019
2:47:16 PM
4-Infiniti
22.5
-2
158
5
154304
10/22/2019
3:26:12 PM
4-Infiniti
22.5
0
107
7
154439
10/22/2019
3:29:37 PM
4-Infiniti
22.5
1
289
-1
154316
10/22/2019
3:57:31PM
4-Infiniti
22.5
0
10
1
154545
10/22/2019
4:01:33 PM
4-Infiniti
22.5
0
384
6
154268
10/22/2019
4:05:30 PM
A-5
-------
Convoy
Position -
Reference
Vehicle ID
Nominal
Vehicle
Speed
(mph)
Exhaust
HC
(ppmC6)
Exhaust
CO
(ppm)
Exhaust
NO
(ppm)
Exhaust
CO2
(ppm)
Transit
Date
Transit
Time
4-Infiniti
22.5
0
29
7
154526
10/22/2019
4:09:31PM
4-Infiniti
22.5
-3
167
115
154385
10/23/2019
8:44:21 AM
4-Infiniti
22.5
0
109
4
154450
10/23/2019
8:48:22 AM
4-Infiniti
22.5
-3
187
7
154415
10/23/2019
8:52:25 AM
4-Infiniti
22.5
-2
203
4
154391
10/23/2019
8:56:22 AM
4-Infiniti
22.5
0
189
6
154414
10/23/2019
9:00:28 AM
4-Infiniti
22.5
-4
228
3
154372
10/23/2019
9:31:05 AM
4-Infiniti
22.5
-6
352
9
154239
10/23/2019
9:35:01 AM
4-Infiniti
22.5
-3
628
1
154075
10/23/2019
9:38:58 AM
4-Infiniti
22.5
-21
367
11
154251
10/23/2019
9:42:57 AM
4-Infiniti
22.5
-3
102
13
154475
10/23/2019
9:47:03 AM
4-Infiniti
22.5
-1
605
3
154110
10/23/2019
10:11:01 AM
4-Infiniti
22.5
-6
161
8
154418
10/23/2019
10:15:11 AM
4-Infiniti
22.5
-2
281
12
154344
10/23/2019
10:18:58 AM
4-Infiniti
22.5
0
478
21
154197
10/23/2019
10:22:59 AM
4-Infiniti
22.5
-2
130
3
154457
10/23/2019
10:26:56 AM
4-Infiniti
22.5
0
235
6
154360
10/23/2019
10:30:54 AM
4-Infiniti
22.5
-7
330
24
154267
10/23/2019
11:26:57 AM
4-Infiniti
22.5
-1
275
6
154347
10/23/2019
11:31:08 AM
4-Infiniti
22.5
-8
116
13
154438
10/23/2019
11:47:06 AM
4-Infiniti
22.5
-20
117
6
154446
10/23/2019
11:50:58 AM
4-Infiniti
22.5
9
318
15
154303
10/23/2019
11:56:59 AM
4-Infiniti
22.5
0
264
-13
154342
10/23/2019
12:21:47 PM
4-Infiniti
22.5
-2
288
17
154326
10/23/2019
12:27:02 PM
4-Infiniti
22.5
-102
236
6
154322
10/23/2019
12:31:02 PM
4-Infiniti
22.5
-13
103
13
154460
10/23/2019
12:43:03 PM
4-Infiniti
22.5
-2
61
1
154426
10/23/2019
12:50:58 PM
4-Infiniti
22.5
0
326
4
154292
10/23/2019
12:59:01PM
4-Infiniti
22.5
-16
256
-9
154298
10/23/2019
1:07:00 PM
4-Infiniti
22.5
-15
263
1
154264
10/23/2019
1:14:58 PM
4-Infiniti
22.5
-2
643
-4
154002
10/23/2019
1:23:05 PM
4-Infiniti
22.5
-10
126
-4
154443
10/23/2019
1:32:24 PM
4-Infiniti
22.5
0
230
1
154355
10/23/2019
1:36:59 PM
4-Infiniti
22.5
-32
363
0
154276
10/23/2019
1:41:02 PM
4-Infiniti
22.5
-5
115
37
154010
10/23/2019
1:45:05 PM
4-Infiniti
22.5
-3
369
2
154265
10/23/2019
1:49:03 PM
4-Infiniti
22.5
-1
147
-10
154408
10/23/2019
2:16:58 PM
4-Infiniti
22.5
-15
184
6
154398
10/23/2019
2:21:01PM
4-Infiniti
22.5
-1
161
8
154400
10/23/2019
2:25:00 PM
4-Infiniti
22.5
0
283
-11
154326
10/23/2019
2:29:05 PM
4-Infiniti
22.5
-2
276
4
154352
10/23/2019
2:34:58 PM
4-Infiniti
22.5
8
289
7
154311
10/23/2019
2:59:04 PM
4-Infiniti
22.5
-10
411
-2
154180
10/23/2019
3:03:09 PM
4-Infiniti
22.5
-9
359
11
154262
10/23/2019
3:07:21PM
A-6
-------
Convoy
Position -
Reference
Vehicle ID
Nominal
Vehicle
Speed
(mph)
Exhaust
HC
(ppmC6)
Exhaust
CO
(ppm)
Exhaust
NO
(ppm)
Exhaust
CO2
(ppm)
Transit
Date
Transit
Time
4-Infiniti
22.5
-7
746
13
153794
10/23/2019
3:13:24 PM
4-Infiniti
22.5
0
15
9
154468
10/23/2019
3:17:25 PM
4-Infiniti
22.5
-11
134
5
154448
10/23/2019
3:21:24 PM
4-Infiniti
22.5
-2
-16
-1
154536
10/24/2019
12:28:51PM
4-Infiniti
22.5
-7
268
11
154334
10/24/2019
12:36:51PM
4-Infiniti
22.5
-2
370
2
154227
10/24/2019
12:44:46 PM
4-Infiniti
22.5
-1
177
36
154395
10/24/2019
1:12:53 PM
4-Infiniti
22.5
0
315
5
154312
10/24/2019
1:20:50 PM
4-Infiniti
22.5
-6
334
16
154277
10/24/2019
1:28:46 PM
4-Infiniti
22.5
3
488
5
154179
10/24/2019
1:48:51PM
4-Infiniti
22.5
0
342
16
154294
10/24/2019
1:52:48 PM
4-Infiniti
22.5
-5
230
15
154370
10/24/2019
2:12:47 PM
4-Infiniti
22.5
0
61
1
154512
10/24/2019
2:16:46 PM
4-Infiniti
22.5
-1
105
7
154458
10/24/2019
2:28:56 PM
4-Infiniti
22.5
0
168
8
154429
10/24/2019
2:32:53 PM
4-Infiniti
22.5
-3
-91
-13
154469
10/24/2019
2:36:58 PM
4-Infiniti
22.5
-1
184
15
154401
10/24/2019
2:52:59 PM
4-Infiniti
22.5
-4
124
8
154456
10/24/2019
2:56:50 PM
4-Infiniti
22.5
-1
119
25
154451
10/24/2019
3:00:48 PM
4-Infiniti
22.5
-7
238
19
154365
10/24/2019
3:05:18 PM
4-Infiniti
22.5
-4
174
14
154417
10/24/2019
3:09:21PM
4-Infiniti
45
-6
109
4
154474
10/20/2019
11:59:14 AM
4-Infiniti
45
5
323
6
154311
10/20/2019
12:02:48 PM
4-Infiniti
45
6
37
-20
154414
10/20/2019
12:06:29 PM
4-Infiniti
45
8
699
33
154025
10/20/2019
12:10:08 PM
4-Infiniti
45
4
231
3
154378
10/20/2019
12:13:51PM
4-Infiniti
45
-26
68
-6
154463
10/20/2019
12:17:29 PM
4-Infiniti
45
1
566
14
154074
10/20/2019
12:41:21PM
4-Infiniti
45
-2
368
9
154221
10/20/2019
12:44:58 PM
4-Infiniti
45
5
173
9
154364
10/20/2019
12:48:39 PM
4-Infiniti
45
4
5098
7
150728
10/20/2019
12:52:19 PM
4-Infiniti
45
-11
79
5
154407
10/20/2019
12:55:59 PM
4-Infiniti
45
1
949
3
153837
10/20/2019
1:32:39 PM
4-Infiniti
45
29
351
-13
154087
10/20/2019
1:39:56 PM
4-Infiniti
45
-3
886
3
153877
10/20/2019
2:02:03 PM
4-Infiniti
45
35
319
15
154286
10/20/2019
2:11:12 PM
4-Infiniti
45
0
8
7
154533
10/20/2019
2:31:26 PM
4-Infiniti
45
0
364
5
154283
10/20/2019
2:38:43 PM
4-Infiniti
45
-20
217
-1
154386
10/20/2019
2:49:42 PM
4-Infiniti
45
-9
77
-6
154472
10/20/2019
2:57:02 PM
4-Infiniti
45
-7
134
33
154384
10/20/2019
3:09:50 PM
4-Infiniti
45
-23
112
10
154445
10/20/2019
3:13:30 PM
4-Infiniti
45
6
100
5
154475
10/20/2019
3:17:10 PM
4-Infiniti
45
1
211
4
154384
10/20/2019
3:28:19 PM
A-7
-------
Convoy
Position -
Reference
Vehicle ID
Nominal
Vehicle
Speed
(mph)
Exhaust
HC
(ppmC6)
Exhaust
CO
(ppm)
Exhaust
NO
(ppm)
Exhaust
CO2
(ppm)
Transit
Date
Transit
Time
4-Infiniti
45
0
482
4
154171
10/20/2019
3:31:51PM
4-Infiniti
45
0
979
11
153755
10/20/2019
3:57:32 PM
4-Infiniti
45
-1
726
20
153959
10/20/2019
4:12:11PM
4-Infiniti
45
-15
795
8
153927
10/20/2019
4:15:51PM
4-Infiniti
45
-14
31
9
154518
10/21/2019
9:40
18 AM
4-Infiniti
45
-15
237
18
154371
10/21/2019
9:45
52 AM
4-Infiniti
45
-41
138
18
154424
10/21/2019
9:48
16 AM
4-Infiniti
45
-10
267
3
154336
10/21/2019
9:52
13 AM
4-Infiniti
45
-5
213
4
154356
10/21/2019
9:56
12 AM
4-Infiniti
45
-12
153
10
154412
10/21/2019
10:00
10 AM
4-Infiniti
45
423
433
43
153957
10/21/2019
10:24
12 AM
4-Infiniti
45
0
515
6
154159
10/21/2019
10:48
15 AM
4-Infiniti
45
-5
324
28
154294
10/21/2019
10:52
11 AM
4-Infiniti
45
1
2134
5
152968
10/21/2019
11:04
10 AM
4-Infiniti
45
-3
872
5
153893
10/21/2019
11:28
13 AM
4-Infiniti
45
98
276
26
154256
10/21/2019
11:32
13 AM
4-Infiniti
45
1
396
3
154161
10/21/2019
11:36
11 AM
4-Infiniti
45
-45
75
25
154466
10/21/2019
11:40
12 AM
4-Infiniti
45
-189
91
43
154430
10/21/2019
11:44
11 AM
4-Infiniti
45
-3
207
4
154392
10/21/2019
12:00
13 PM
4-Infiniti
45
2
1006
1
153808
10/21/2019
12:04
12 PM
4-Infiniti
45
0
549
8
154132
10/21/2019
12:16
15 PM
4-Infiniti
45
-19
92
4
154473
10/21/2019
12:20
16 PM
4-Infiniti
45
-2
449
16
154219
10/21/2019
12:24
14 PM
4-Infiniti
45
-1
240
6
154374
10/21/2019
12:40
16 PM
4-Infiniti
45
-17
1286
-1
153579
10/21/2019
12:46
12 PM
4-Infiniti
45
-3
109
6
154405
10/21/2019
1:06
15 PM
4-Infiniti
45
-5
780
1
153973
10/21/2019
1:14
16 PM
4-Infiniti
45
12
55
26
151221
10/21/2019
1:24
12 PM
4-Infiniti
45
-300
911
-8
153848
10/21/2019
1:32
14 PM
4-Infiniti
45
37
175
4
154366
10/21/2019
1:42
14 PM
4-Infiniti
45
77
38009
2
126574
10/21/2019
1:52
12 PM
4-Infiniti
45
175
47348
6
119640
10/21/2019
2:00
12 PM
4-Infiniti
45
0
120
9
154406
10/22/2019
10:52:04 AM
4-Infiniti
45
-1
210
11
154391
10/22/2019
10:56:05 AM
4-Infiniti
45
-3
279
19
154339
10/22/2019
11:00:05 AM
4-Infiniti
45
3
583
10
154073
10/22/2019
11:07:59 AM
4-Infiniti
45
0
228
11
154366
10/22/2019
11:18:01 AM
4-Infiniti
45
-14
212
1
154367
10/22/2019
11:48:03 AM
4-Infiniti
45
-16
279
-2
154311
10/22/2019
11:56:02 AM
4-Infiniti
45
0
463
3
154186
10/22/2019
12:04:01 PM
4-Infiniti
45
465
389
52
153955
10/22/2019
12:34:01PM
4-Infiniti
45
-8
28
4
154477
10/22/2019
12:38:04 PM
4-Infiniti
45
-33
284
-2
154339
10/22/2019
1:47:11PM
A-8
-------
Convoy
Position -
Reference
Vehicle ID
Nominal
Vehicle
Speed
(mph)
Exhaust
HC
(ppmC6)
Exhaust
CO
(ppm)
Exhaust
NO
(ppm)
Exhaust
CO2
(ppm)
Transit
Date
Transit
Time
4-Infiniti
45
-7
835
7
153909
10/22/2019
1:54:28 PM
4-Infiniti
45
-5
935
2
153824
10/22/2019
2:10:05 PM
4-Infiniti
45
-30
351
23
154203
10/22/2019
2:18:03 PM
4-Infiniti
45
-12
575
8
154120
10/22/2019
2:26:05 PM
4-Infiniti
45
5
436
29
154051
10/22/2019
2:34:03 PM
4-Infiniti
45
-6
890
16
153895
10/22/2019
2:38:03 PM
4-Infiniti
45
0
227
8
154350
10/22/2019
2:42:02 PM
4-Infiniti
45
0
184
13
154394
10/22/2019
2:58:03 PM
4-Infiniti
45
-10
636
20
154006
10/22/2019
3:02:03 PM
4-Infiniti
45
-9
279
7
154334
10/22/2019
3:53:07 PM
4-Infiniti
45
79
34380
7
129227
10/22/2019
4:17:10 PM
4-Infiniti
45
0
383
1
154227
10/23/2019
9:04:05 AM
4-Infiniti
45
-4
207
0
154360
10/23/2019
9:08:00 AM
4-Infiniti
45
-4
342
29
154239
10/23/2019
9:12:02 AM
4-Infiniti
45
1
683
14
154036
10/23/2019
9:18:02 AM
4-Infiniti
45
0
306
11
154281
10/23/2019
9:22:03 AM
4-Infiniti
45
2
186
17
154388
10/23/2019
9:26:07 AM
4-Infiniti
45
-37
143
-5
154348
10/23/2019
9:50:03 AM
4-Infiniti
45
288
151
5
154267
10/23/2019
9:54:04 AM
4-Infiniti
45
485
383
8
153972
10/23/2019
9:58:03 AM
4-Infiniti
45
-1
121
-7
154447
10/23/2019
10:02:04 AM
4-Infiniti
45
0
158
2
154426
10/23/2019
10:34:03 AM
4-Infiniti
45
-5
364
21
154265
10/23/2019
10:38:03 AM
4-Infiniti
45
-2
241
12
154363
10/23/2019
10:42:02 AM
4-Infiniti
45
-6
184
16
154402
10/23/2019
10:50:01 AM
4-Infiniti
45
-18
291
16
154313
10/23/2019
10:53:59 AM
4-Infiniti
45
-3
581
7
153996
10/23/2019
11:18:05 AM
4-Infiniti
45
1
835
29
153729
10/23/2019
11:22:03 AM
4-Infiniti
45
0
278
8
154350
10/23/2019
11:38:03 AM
4-Infiniti
45
-8
294
12
154323
10/23/2019
11:42:01 AM
4-Infiniti
45
0
2488
8
152682
10/23/2019
12:00:02 PM
4-Infiniti
45
-25
752
8
153998
10/23/2019
12:04:05 PM
4-Infiniti
45
50
398
114
154141
10/23/2019
12:34:00 PM
4-Infiniti
45
17
134
-6
154403
10/23/2019
12:38:04 PM
4-Infiniti
45
-5
174
28
154249
10/23/2019
12:46:03 PM
4-Infiniti
45
-3
1412
13
153403
10/23/2019
12:54:03 PM
4-Infiniti
45
-13
160
20
154077
10/23/2019
1:02:02 PM
4-Infiniti
45
-10
236
1
154319
10/23/2019
1:10:02 PM
4-Infiniti
45
-15
545
-11
153359
10/23/2019
1:18:04 PM
4-Infiniti
45
-1
335
2
154303
10/23/2019
1:52:02 PM
4-Infiniti
45
-83
268
-1
154260
10/23/2019
1:56:02 PM
4-Infiniti
45
1047
366
32
153602
10/23/2019
2:00:00 PM
4-Infiniti
45
-50
272
-5
154256
10/23/2019
2:04:00 PM
4-Infiniti
45
-2
55
-3
154496
10/23/2019
2:08:00 PM
A-9
-------
Convoy
Position -
Reference
Vehicle ID
Nominal
Vehicle
Speed
(mph)
Exhaust
HC
(ppmC6)
Exhaust
CO
(ppm)
Exhaust
NO
(ppm)
Exhaust
CO2
(ppm)
Transit
Date
Transit
Time
4-Infiniti
45
0
194
6
154368
10/23/2019
2:11:59 PM
4-Infiniti
45
-71
120
1
154448
10/23/2019
2:37:59 PM
4-Infiniti
45
-28
56
12
154419
10/23/2019
2:42:02 PM
4-Infiniti
45
5
215
-15
154383
10/23/2019
2:45:58 PM
4-Infiniti
45
-164
178
23
154360
10/23/2019
2:50:00 PM
4-Infiniti
45
-7
208
16
154378
10/23/2019
2:53:59 PM
4-Infiniti
45
1
29
16
154323
10/23/2019
3:24:26 PM
4-Infiniti
45
2
706
7
154023
10/23/2019
3:28:48 PM
4-Infiniti
45
1
940
20
153387
10/23/2019
3:32:34 PM
4-Infiniti
45
5
563
1
154058
10/23/2019
3:36:27 PM
4-Infiniti
45
-21
186
5
154366
10/23/2019
3:44:47 PM
4-Infiniti
45
-10
282
16
154276
10/24/2019
12:21:21PM
4-Infiniti
45
-7
343
49
154000
10/24/2019
12:39:51PM
4-Infiniti
45
13
1419
19
153494
10/24/2019
1:07:57 PM
4-Infiniti
45
-1
1308
32
153467
10/24/2019
1:15:54 PM
4-Infiniti
45
-5
410
12
154240
10/24/2019
1:23:55 PM
4-Infiniti
45
0
315
10
154290
10/24/2019
1:31:52 PM
4-Infiniti
45
-157
91
19
154399
10/24/2019
1:35:51PM
4-Infiniti
45
-83
17
16
154484
10/24/2019
1:39:56 PM
4-Infiniti
45
2
1782
23
153236
10/24/2019
1:43:54 PM
4-Infiniti
45
2
1924
23
153108
10/24/2019
1:55:53 PM
4-Infiniti
45
-21
574
-9
154085
10/24/2019
1:59:53 PM
4-Infiniti
45
-39
187
-4
154404
10/24/2019
2:03:53 PM
4-Infiniti
45
-22
173
12
154414
10/24/2019
2:19:55 PM
4-Infiniti
45
-16
1000
28
153785
10/24/2019
3:16:55 PM
4-Infiniti
45
0
571
-5
154122
10/24/2019
3:20:16 PM
4-Infiniti
45
-2
191
14
154254
10/24/2019
3:24:11PM
4-Infiniti
45
-52
949
83
153772
10/24/2019
3:28:12 PM
4-Infiniti
45
1
297
226
154236
10/24/2019
3:32:13 PM
A-10
-------
Appendix B:
Westminster Dataset and Analysis Program Locations
B-0
-------
Tables B-l and B-2 provide the locations of the datasets and analysis programs that were used to
archive and analyze the Westminster data.
B-l
-------
Table B-1. Vehicle Description Data for the Westminster Sample2
Directory
Dataset
SAS program
Inputs
Outputs
V
P/Colorado_OBD_IM240/
RegistrationData/2019 RegData/
col_reg_2019 Julyupdate. sas7bdat
u
P/CDPHE/Regis2019
copy2022_COreg_find
mk mod yr.sas
V @ Line
41
T @ Line 157
T
P/CDPHE/Regis2019
co reg 2019 wts.sas7bdat
A
P/EDARinDenver-OCT2019/Analysis/
Westminster OCT2019Results 200124Reprocess-
200219 modified.CSV
B
P/EDARinDenver-OCT2019/Analysis/
YIN output.CSV
2 P/EDARinDenver-OCT2019/Analysis_MLout/SAS Program Flow.xlsx
B-2
-------
Table B-2. Analysis Programs for the Westminster Sample
P:/EDARinDenver-OCT2019/
Dataset
SAS program
Inputs
Outputs
c
AnalysisMLout/
OCT19_metadata. S AS
A a Line 50
B a Line 304
D (ai Line 510
D
AnalysisMLout/
spreadsheet_vin_211115. sas7bdat
Y
Analysis_MLout/211108/Data_MLout
Results_Westminster_Series5##-211108.zip
H
Analy sis_MLout/211108/Anal_MLout
OCT 19_read_matlab. sas
Y Line 74
I a Line 725
I
Analy sis_MLout/211108/Anal_MLout
OCT19 scansubset-211108.sas7bdat
E
Analy sis_MLout/211108/Anal_MLout
OCT 19_bumper. sas
D a Line 52
I a Line 167
F Line 1583
G Line 1585
F
Analy sis_MLout/211108/Anal_MLout
scansum_autofronttrailer_211108.sas7bdat
G
Analy sis_MLout/211108/Anal_MLout
BumperLocations-211108_v2.CSV
AA
Analysis_MLout/211122/Data_MLout
Results_Westminster_Series5##-211122.zip
Z
Analy sis_MLout/211122/Anal_MLout
OCT 19_read_matlab. sas
AA 'a Line 82
J Line 734
J
Analy sis_MLout/211122/Anal_MLout
OCT19_scan_211122_standard.sas7bdat
K
Analy sis_MLout/211122/Anal_MLout
0CT19_interpshapeC02_5.sas
D a Line 175
F (cii Line 303
J Line 132
L (@, Line 1705
M@ Line 1729
L
Analy sis_MLout/211122/Anal_MLout
BumperLocations_2 111 22_draft_v3 .CSV
M
Analy sis_MLout/211122/Anal_MLout
C02shapes_by_udst.sas7bdat
Q
Analy sis_MLout/220817/Data_MLout
OCT 19_Unit_Date_Series_Transit_"randate" *. C S V
N
Analy sis_MLout/220817/Anal_MLout
OCT 19_read_matlab. sas
Q a Line 117
F a Line 512
M a Line 576
D 'a Line 625
O a Line 661
P (®, Line 727: std
(741: bootstrap
755: lsmean)
0
Analy sis_MLout/220817/Anal_MLout
OCT 19_pixel_readout_"rundate". sas7bdat
P
Analy sis_MLout/220817/Anal_MLout
OCT 19_scan_readout_"randate". sas7bdat
w
Analy sisMLout/22080 l/Anal_MLout
temp:
OCT 19_make_flags. sas
0 a Line 57
X@ Line 812
AB a Line 807
X
Analy sisMLout/22080 l/Anal_MLout
temp: udstplusflags.sas7bdat
AB
Analy sisMLout/22080 l/Anal_MLout
temp: pixelplusflags.sas7bdat
R
Analy sis_MLout/220817/Anal_MLout
OCT19_scan_postprocess.sas
P a Line 96
D a Line 141
T (a\ Line 296
F a Line 341
M a Line 365
X a Line TBD
S (ai Line 539
Analy sis_MLout/220817/Anal_MLout
OCT 19_scanmetaflag. sas7bdat
B-3
-------
B-4
-------
Appendix C:
Plots for Signal Adjustment Demonstration
c-o
-------
Figure C-1. Offset Adjustment Example for no CO and NO Emissions: EV-1, Low EvapHC from DOOR, Low Speed
CD
.Q
E
c
CD
o
CO
0 100 200
Scan Position
7 20191020 000506 car 000108
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
N02
1
,1 ,M
. ¦ i'iVM
•mv.w Mj
TiiW
V.'iViVi'ii I Imi'i 1
¦' 1 ,v ' IJ
iii .
. 11
ifji
m
;¦
-------
Figure C-2. Offset Adjustment Example for no CO and NO Emissions: EV-1, High EvapHC from DOOR, Low Speed
C-2
-------
Figure C-3. Offset Adjustment Example with CO and NO Emissions: EV-2, Low EvapHC from DOOR, Low Speed
C-3
-------
Figure C-4. Offset Adjustment Example with CO and NO Emissions: EV-2, Low EvapHC from DOOR, Low Speed
C-4
-------
Figure C-5. Offset Adjustment Example with CO and NO Emissions: EV-2, High EvapHC from DOOR, Low Speed
C-5
-------
Figure C-6. Offset Adjustment Example indicating CO Channel Failure: 2013 Ford F-150 (Fleet Vehicle)
a>
TO
O
\
r' j
r • \
I" 'I
¦ /
1
0 5
Bins x-io"4
-2
0 2
Bins x 1Q"4
0 1
Bins x10"4
C-6
-------
Figure C-7. Outlier Removal Example: EV-2, Low EvapHC from HOOD, Low Speed
C-7
-------
Figure C-8. Outlier Removal Example: EV-2, High EvapHC from DOOR, Low Speed
C-8
-------
Figure C-9. HC Multi-Tonal Cancellation Example: EV-2, High EvapHC from DOOR
7 20191023 000512 car 001168
HC before MTC
50
100 150
Scan Position
200
250
20
15
-jq~4 HC Scan Value, Line 100
10
4>
jg
>
-5
10
20
30
40
50
Before MTC
After MTC; Mode! - 2-EV2 6400mg Door 22mph Y
x 1 n 3 HC Scan Value, Line 200
2.5 r 1 1 1
HC after MTC
50 100 150 200 250
Scan Position
20 30
Scan Number
C-9
-------
Figure C-10. HC Multi-Tonal Cancellation Example: EV-2, High EvapHC from TANK
CD
JO:
E
TO
O
03
ja
E
8
CO
7 20191024 000514 car 001635
HC before MTC
5
10
15
20
25
30
35
40
45
50
¦is? Lra
n" i
50
1 !
I
100 150
Scan Position
HC after MTC
100 150
Scan Position
200
200
x 10
HC Scan Value, Line 100
250
Before MTC
After MTC, Model - 2-EV2 64Q0mg Tank 22mph Y
rto
HC Scan Value, Line 200
250
20 30
Scan Number
C-10
-------
Figure C-11. HC Multi-Tonal Cancellation Example: EV-2, High EvapHC from HOOD
X2
E
TO
o
CO
<13
n
e
TO
O
CO
7 20191020 000505 car 001813
HC before MTC
HC Scan Value, Line 100
5
10
15
20
25
30
35
40
45
¦"i—7
'
Kill
w
111
|; M
Kl
"T
Ik '
! ' !i
,
i
'
r
u_
0)
_3
CD
>
50
100 150
Scan Position
HC after MTC
200
250
-0,5
Before MTC
After MTC; Model - 2-EV2 6400mg Hood 22mph Y
, ,'h %mwm
Wmmm*
i i 1 • i iii • M mil i it i
• • ¦ "¦ T >, I 'I ¦
-'±&
10 4 HC Scan Value, Line 200
'R! JJJ
<« ¦ r f r iiti
50
100 150
Scan Position
200
250
20 30
Scan Number
C-11
-------
Figure C-12. HC Multi-Tonal Cancellation Example: EV-2, Low EvapHC from DOOR
7 20191024 000514 car 001202
1
0.5
0
0)
ro
>
-0.5
-1
* Before MTC
After MTC; Model - 2-EV2 200rrig Door 22mph Y
HC before MTC
100 150
Scan Position
x10"3 HC Scan Value, Line 100
o
£2
E
03
O
CO
10
20
30
40
50
60
70
V " '
¦ ii • If
50
HC after MTC
"! ¦
i \
_L
t ".'.'i
¦ I'
m
100 150
Scan Position
200
250
x10
HC Scan Value, Line 200
10 20 30 40 50 60 70 80
Scan Number
C-12
-------
Figure C-13. HC Multi-Tonal Cancellation Example: EV-2, Low EvapHC from TANK
7 20191023 000512 car 002206
HC before MTC
10
20
&_
CD
_Q
30
E
Z5
z
40
tz
03
o
c/5
50
60
70
0
10
20
CD
.Q
30
E
3
2
40
c
ra
o
CO
50
60
70
50
50
100 150
Scan Position
HC after MTC
100 150
Scan Position
200
200
x10
HC Scan Value, Line 100
250
0 10 20 30 40 50 60 70 80
Before MTC
After MTC; Model - 2-EV2 2G0mg Tank 22mph Y
x10
HC Scan Value, Line 200
250
0 10 20 30 40 50 60 70 80
Scan Number
C-13
-------
Figure C-14. HC Multi-Tonal Cancellation Example: EV-2, Low EvapHC from HOOD
7 20191024 000514 car 001053
HC before MTC
x10
HG Scan Value, Line 100
CD
ja
E
CO
o
C/5
o
£2
E
03
O
CO
100 150
Scan Position
HC after MTC
50
100 150
Scan Position
200
250
0 10 20 30 40 50 60 70 80
Before MTC
After MTC; Model - 2-EV2 200mg Hood 22mph Y
x10
HC Scan Value, Line 200
250
0 10 20 30 40 50 60 70 80
Scan Number
C-14
-------
Figure C-15. HC Multi-Tonal Cancellation Example: EV-1, High EvapHC from DOOR
a>
ja
E
8
CO
7 20191023 000512 car 001167
HC before MTC
5
10
15
£_
CD
20
_Q
E
3
z
25
c
TO
O
30
35
40
45
Scan Position
HC after MTC
x 10
HC Scan Value, Line 100
250
Before MTC
After MTC, Model - 1-EV1 6400mg Door22mph Y
x10
HC Scan Value, Line 200
50
100 150
Scan Position
200 250
-0,5
10 20 30 40
Scan Number
C-15
-------
Figure C-16. HC Multi-Tonal Cancellation Example: EV-1, High EvapHC from TANK
CD
JO:
E
TO
O
03
ja
E
8
CO
7 20191022 000509 car 001451
HC before MTC
x 10
HC Scan Value, Line 100
5
10
15
20
25
30
35
40
45
—r
' li
'i I
" r
! i
'
, ; i,; ,
i it i
1:
i i
, i
. i
i i ' J li
11 i
i •! 1 ,
I,
l'
I 'li ill i
I11 I "'I
r
il
ii'i.-fi
i:rj
50
100 150
Scan Position
HC after MTC
200
250
Before MTC
After MTC, Model - 1-EV1 64Q0mg Tank 22mph Y
x10
HC Scan Value, Line 200
50
100 150
Scan Position
200
250
20 30
Scan Number
C-16
-------
Figure C-17. HC Multi-Tonal Cancellation Example: EV-1, High EvapHC from HOOD
CD
JO:
E
TO
O
03
ja
E
8
CO
7 20191022 000509 car 002092
HC before MTC
5
10
15
20
25
30
35
40
45
50
50
M
I
,
I
i j-
_
100 150
Scan Position
HC after MTC
200
100 150
Scan Position
200
x 10
HC Scan Value, Line 100
_JL
250
Before MTC
After MTC; Model -1 EV" 6400mg Hood 22mph Y
x10
HC Scan Value, Line 200
250
20 30
Scan Number
C-17
-------
Figure C-18. HC Multi-Tonal Cancellation Example: EV-1, Low EvapHC from DOOR
7 20191020 000506 car 000108
X2
E
TO
o
CO
<13
n
e
TO
O
CO
10
20
30
40
50
60
70
HC before MTC
1 q-3 HC Scan Value, Line 100
tin
11 lil
„!•
li'm'j. i
• i'v
n r
'
I
14' r1 ii,
: i
I
¦ i-1
1
i i
! I
Mi
1 \>
IV? J
i ' i| !
V. !|
I
\m:
i
'!
i i
I I I
jUt
50
100 150
Scan Position
HC after MTC
200
250
-0,5
10 20 30 40 50 60
Before MTC
¦After MTC; Model-1-EV1 200mg Door22mphY
10 4 HC Scan Value, Line 200
50 100 150 200 250
Scan Position
0 10 20 30 40 50 60
Scan Number
70 80
70 80
C-18
-------
Figure C-19. HC Multi-Tonal Cancellation Example: EV-1, Low EvapHC from TANK
7 20191021 000507 car 002354
HC before MTC
10
20
&_
CD
_Q
30
E
Z5
z
40
tz
03
o
c/5
50
60
70
10
20
%—
CD
.Q
30
E
3
2
40
c
ra
o
CO
50
60
70
50
'
1 1
_J L
100 150
Scan Position
HC after MTC
x10"
HC Scan Value, Line 100
200
0)
rj
-2
-4
-6
-10
-Vi
¦ ¦
250
0 10 20 30 40 50 60 70 80
Before MTC
After MTC; Model - 1-EV1 2G0mg Tank 22mph Y
x10
HC Scan Value, Line 200
50 100 150 200 250
Scan Position
0 10 20 30 40 50 60 70 80
Scan Number
C-19
-------
Figure C-20. HC Multi-Tonal Cancellation Example: EV-1, Low EvapHC from HOOD
7 20191020 000505 car 002864
HC before MTC
1.5
10
20
| 30 0.5
E 0)
40 |
>
« 50
60
-0.5
70
100 150
Scan Position
Before MTC
• After MTC, Model - 1-FV1 200mg Hood 22mph Y
HC before MTC x103 HC Scan Value, Line 100
HC after MTC
50 100 150 200 250
Scan Position
x1q 4 HC Scan Value, Line 200
15 1 1 1 1 1 1
10 20 30 40 50
Scan Number
C-20
-------
Figure C-21. Before (top) and After (bottom) Adaptive Notch Filtering. Example: EV-2, High EvapHC from DOOR
C-21
-------
Figure C-22. Before (top) and After (bottom) Adaptive Notch Filtering. Example: EV-2, High EvapHC from TANK
C-22
-------
Figure C-23. Before (top) and After (bottom) Adaptive Notch Filtering. Example: EV-2, High EvapHC from HOOD
C-23
-------
Figure C-24. Before (top) and After (bottom) Adaptive Notch Filtering. Example: EV-2, Low EvapHC from DOOR
C-24
-------
Figure C-25. Before (top) and After (bottom) Adaptive Notch Filtering. Example: EV-2, Low EvapHC from TANK
0 100 200
Scan Position
7 20191023 000512 car 002206
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
C-25
-------
Figure C-26. Before (top) and After (bottom) Adaptive Notch Filtering. Example: EV-2, Low EvapHC from HOOD
0 100 200
Scan Position
7 20191024 000514 car 001053
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
C-26
-------
Figure C-27. Before (top) and After (bottom) Adaptive Notch Filtering. Example: EV-1, High EvapHC from DOOR
C-27
-------
Figure C-28. Before (top) and After (bottom) Adaptive Notch Filtering. Example: EV-1, High EvapHC from TANK
5
10
15
20
25
30
35
40
45
HC
Hi
I
0 100 200
Scan Position
7 20191022 000509 car 001451
C02
0 100 200
Scan Position
100 200
Scan Position
NO
100 200
Scan Position
N02
0 100 200
Scan Position
NO
N02
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
¦
iilnuD
M OP? |i ;f
n.../1'fji ii.ipi Hi1.1 ilj
is
IijIii,IMifK
i i i
0 100 200
Scan Position
5
10
15
20
25
30
35
40
45
Jift
ii in
1 • Ha
Iffi
1,1 ! /ri'Ui.M
. 1 ii 1 !n ii
1 ;|||
i I
1 liW
1 I IIIII
.ii
0 100 200
Scan Position
C-28
-------
Figure C-29. Before (top) and After (bottom) Adaptive Notch Filtering. Example: EV-1, High EvapHC from HOOD
0 100 200
Scan Position
7 20191022 000509 car 002092
C02
' n
0 100 200
Scan Position
100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
C-29
-------
Figure C-30. Before (top) and After (bottom) Adaptive Notch Filtering. Example: EV-1, Low EvapHC from DOOR
7 20191020 000506 car 000108
C02
0 100 200
Scan Position
100 200
Scan Position
0 100 200
Scan Position
CO
NO
10
¦ ' '¦ ¦'1 ''i1,'/
10
-
20
i, i I I ii mi
i "Ii1 r "'I ' vM
20
: v.\';! J'"-1!
30
. ¦' '' ;V,
u H:i i '' *1!., ¦'» r'w
• , ' ... ' 1 i. M.'4
30
40
40
50
¦' III
50
60
60
70
I1 ¦ ' '''ift
is :
70
0 100 200
Scan Position
N02
":;'i t'ff
BHiWM
j\\\>
:®r
i ; I ¦
0 100 200
Scan Position
0 100 200
Scan Position
:
i! 'Mil',,mi
www
if!
if
iW
!}>J 1 "i
,!,«l 1 \W
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
C-30
-------
Figure C-31. Before (top) and After (bottom) Adaptive Notch Filtering. Example: EV-1, Low EvapHC from TANK
7 20191021 000507 car 002354
HC C02 CO NO N02
100 200 0 100 200 0 100 200 0 100 200 0 100 200
Scan Position Scan Position Scan Position Scan Position Scan Position
cil
-------
Figure C-32. Before (top) and After (bottom) Adaptive Notch Filtering. Example: EV-1, Low EvapHC from HOOD
7 20191020 000505 car 002864
HC C02 CO NO N02
0 100 200 0 100 200 0 100 200 0 100 200 0 100 200
Scan Position Scan Position Scan Position Scan Position Scan Position
C-32
-------
Figure C-33. ZigZag Interpolation Example: EV-2, High EvapHC from DOOR
7 20191023 000512 car 001168
C-33
-------
Figure C-34. ZigZag Interpolation Example: EV-2, High EvapHC from TANK
7 20191024 000514 car 001635
C-34
-------
Figure C-35. ZigZag Interpolation Example: EV-2, High EvapHC from HOOD
7 20191020 000505 car 001813
C-35
-------
Figure C-36. ZigZag Interpolation Example: EV-2, Low EvapHC from DOOR
7 20191024 000514 car 001202
HC before interp HC after interp C02 before interp C02 after interp
C-36
-------
Figure C-37. ZigZag interpolation Example: EV-2, Low EvapHC from TANK
7 20191023 000512 car 002206
HC before interp
j—
10
20
30
40
50
60
70
i I
ii II i:I L
HC after interp C02 before interp C02 after interp
100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
C-37
-------
Figure C-38. ZigZag Interpolation Example: EV-2, Low EvapHC from HOOD
7 20191024 000514 car 001053
HC before interp
HC after interp
C02 before interp
C02 after interp
CD
.Q
E
C
CT5
O
CO
I
I Ml I
,, 11111 lit
. 3I.li. I IB
10
mi I I
wUdkiMKUn
J, iJl;Jiiil|llJJ i I', f i'IiI J
10
20
30
40
50
60
70
100 200 0 100 200 0 100 200 0 100 200
Scan Position Scan Position Scan Position Scan Position
C-38
-------
Figure C-39. ZigZag Interpolation Example: EV-1, High EvapHCfrom DOOR
7 20191023 000512 car 001167
C-39
-------
Figure C-40. ZigZag Interpolation Example: EV-1, High EvapHC from TANK
7 20191022 000509 car 001451
C-40
-------
Figure C-41. ZigZag Interpolation Example: EV-1, High EvapHCfrom HOOD
7 20191022 000509 car 002092
HC before interp
HC after interp
C02 before interp
C02 after interp
10
15
20
25
30
35
40
45
~\ r
-
J L
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
100 200
Scan Position
C-41
-------
Figure C-42. ZigZag Interpolation Example: EV-1, Low EvapHC from DOOR
7 20191020 000506 car 000108
C-42
-------
Figure C-43. ZigZag Interpolation Example: EV-1, Low EvapHC from TANK
7 20191021 000507 car 002354
HC before interp HC after interp C02 before interp C02 after interp
C-43
-------
Figure C-44. ZigZag Interpolation Example: EV-1, Low EvapHC from HOOD
7 20191020 000505 car 002864
C-44
-------
Appendix D:
Examples of Blind Source Separation
using Standard ICA
D-0
-------
Figure D-1. BSS Example: EV-1, no EvapHC, Low Speed, no Exh
EvapPlume
ExhPlume
Noisel
Noise2
Noise3
0 100 200
Scan Position
0 100 200
Scan Position
i
i
ii
i i i.i i
jo I II I 111 | U II.
I ' i nil
,il
mi:.
1 lll.ll I'll
l. ii
E
0 100 200
Scan Position
D-1
5
10
15
20
25
30
35
I
Hi II [i I'WifflJU
I I ill I . I ll II
lip ' y .ii. in'J
1IK1 111 I'lli
ill il l ! ®I Ml
in
I inlfli HIBFflHI
IP ifJ
f
0 100 200
Scan Position
51
li raiMU Wi i ii
10 \
Mil i Hi ' 1IMI
¦!v 1'1l11 i1' ¦1 'Ii 11111!
11 mil ii|l || ill 11 ill ii I
15 I
Illllii
L'U 111 luiil DII1111'
20 I
111 I 1H 111 11 Hi i
1 II I llh f i'
25 I
III fir IJf Hi ill111
V 111!1 lllilill 11
H1II! lliji llii i\
30 |
l
35 |
1,1' \
0 100 200
Scan Position
-------
Figure D-2. BSS Example: EV-1, no EvapHC, Low Speed, C02 Exh
CD
n
E
3
CD
O
CO
0 100 200
Scan Position
7 20191020 000505 car 001287
1-EV1 Omg n/a 22mph Y
NO
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
N02
,—
11 > If,' l| ! !J!J
mill I II;
1
IJ I nil
i11
IJ
nil
ii' I
fir i'iiiii 11
ill" ' I
)
I I
i h.J. I ,n ii
hi ir ii iir
lJij. i
. 11 ii i ii
1
11 j i
ii i
0 100 200
Scan Position
EvapPlume
ExhPlume
Noisel
Noise2
Noise3
CD
.Q
E
3
c
CT3
O
CO
0 100 200
Scan Position
o 100 200
Scan Position
I
I I il l -
5
II 1 III III 1
¦ 1 : , ''!¦
1 IN fill ¦ :||
10
¦ i 1 '¦
15
i i 1
i
" i'i
20
ii 11
25
|.
30
i
i j iii
35
100 200
Scan Position
D-2
i > l 1 • i JI l IN I I I I .. II ,11, I .11
,li: i m ; ,: i| I I; III
ill
I Ml I J 1
J l,M . Wi I
Mil
1. I.I
Hlffll
H1
iliii
P'ill
fill
in I
i
ilH IJ l|F I ill.ti I If III, 'i!
¦II("flji ui ill
i
! II
Scan Position
5
10
15
20
25
30
35
.
Scan Position
-------
Figure D-3. BSS Example: EV-1, High EvapHC from DOOR, Low Speed, C02 Exh
EvapPlume
ExhPlume
Noisel
Noise2
NoiseS
0 100 200
Scan Position
0 100 200
Scan Position
1
I U LI
kill
ft is
ii Jin
ii iiirnHH
a I'll [Mini" T'
*, villi.
IIL
iili
II M il 111,1 I I
1 I. "
I ililM
I'll||| III 111 .
I il!|
§
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
D-3
-------
Figure D-4. BSS Example: EV-1, Medium EvapHC from DOOR, Low Speed, C02 Exh
7 20191023 000512 car 002585
C02 CO
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
1-EV1 800mg Door22mph Y
NO
25
30
35
40
45
50
55
60
65
IJill illil
''(In ii ir'Mli
1 |! ii'JlJM
! i
I !i IJilMIM 1U
I I -
!
1 I I
0 100 200
Scan Position
N02
25
30
35
40
45
50
55
60
65
i
BIH ' r,
H jr I l i,ii i ) l i. :, M |
:: lim R |i§ ll'l IUI
u1 'Mali iffifMil wMral'i j 1
I ilnMpi i 11 ¦ i i" u, ii ii y J
i' l
i 1 pin
¦iWiiilil
jMWj "l!i1 h ; I1'
i 1. i|j_
y uffrari1,'
ii
iffrjiiiiBL
V ' »|l 'h|
i'liil
0 100 200
Scan Position
EvapPlume
ExhPlume
Noisel
Noise2
Noise3
' ''' 1 , 1 in,
1111 i'Vi I mill 1 I
¦ fiM
' i" '' I ' ll i',
¦ ,1 ,, 1 VV
,, ,, 1 1 11 1 I 1
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
D-4
25
30
35
40
45
50
55
60
65
l!!i: 11, II
If! i'IJI
ill i I k ill
i i
i
ll'!!'i'liil1
jitl
m
ItJliUl1
iiiiii 'Hi; rd'yiifi
i
0 100 200
Scan Position
25
30
35
40
45
50
55
60
65
iM
Mm
I jd i J,
i
i III i i
p mi j
/ If*
III
0 100 200
Scan Position
-------
Figure D-5. BSS Example: EV-1, Low EvapHC from DOOR, Low Speed, C02 Exh
7 20191022 000509 car 003372
0 100 200
Scan Position
100 200
Scan Position
CO
i J ,¦ i ii
H i
if J
0 100 200
Scan Position
1-EV1 200mg Door22mph Y
NO
N02
0 100 200
Scan Position
0 100 200
Scan Position
EvapPlume
ExhPlume
Noisel
Noise2
NoiseS
n
E
3
c
m
o
CO
25
30
35
40
45
50
55
60
65
0 100 200
Scan Position
100 200
Scan Position
iiil ii if h ill I
ii1" {ftl'llfli
ili rt , i
1,0111 il
. u
J IlL i " ffl if
II :i! i', i. ill 111 I Hill i: I ill
lllliflfi11
lul.in InMlii, I frli , 'lftl.I
11 I
H¦ i II i' h I i' in l iij l 'I ii-
Hi! an I 1
Itil'll
ii In 11 ' iiiii
liii i ll'
ii.
i
0 100 200
Scan Position
D-5
0 100 200
Scan Position
0 100 200
Scan Position
-------
Figure D-6. BSS Example: EV-1, Medium EvapHC from DOOR, Highspeed, C02 Exh
EvapPlume
ExhPlume
Noisel
Noise2
Noise3
10
CD
i 15
Z
§ 20
W
25
30
I
I l'l I I I
X!
In
i i
0 100 200
Scan Position
100 200
Scan Position
5 ¦
iff
10
15
20
ill Ml
I|N ''l' i ''
ujifiliji ill i
("ill1
Hi, 111) II!
I J J
25 ¦
30
I
0 100 200
Scan Position
10
15
20
25
30
II I
I I
I
0 100 200
Scan Position
0 100 200
Scan Position
D-6
-------
Figure D-7. BSS Example: EV-1, High EvapHC from TANK, Low Speed, C02 Exh
7 20191024 000514 car 001634
HC
5
10
fc 15
_Q
E
i 20
c
cs
o
w 25
-
30
35
' '
—1—
1 ||H
I1!,'
Hill [III:
Of
III 11
ill ill, ll
111 11
111 M l
1
I 1 111
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
1-EV1 6400mg Tank 22mph Y
NO
IliilMliH"1!!
5
1!
ii 11 «¦¦[ Fir
llwMin
i ii.
i..ii
0 100 200
Scan Position
N02
0 100 200
Scan Position
EvapPlume
ExhPlume
Noisel
Noise2
NoiseS
5
10
*5 15
is
£
z 20
c
CD
o
w 25
30
35
1 1J
'II 1
J
'
I I III
'II
IB
0 100 200
Scan Position
0 100 200
Scan Position
Jill II
¦
III Iff 1 III
ii-
Mil
J i RnilU
HIE MM
nwii
gwiiinii
II
1 fl I
IWBIil
F W J' 11!
11
i ll ,n
him -
i:,1 i.. 'ii (i
: till it
: 1
IlllliMf
,ii.l. J J1 illJH.! :,t
i
0 100 200 0 100 200
Scan Position Scan Position
0 100 200
Scan Position
D-7
-------
Figure D-8. BSS Example: EV-1, Medium EvapHC from TANK, Low Speed, C02 Exh
EvapPlume
ExhPlume
Noisel
Noise2
Noise3
25
30
35
® 40
z 45
I 50
55
60
65
0 100 200
Scan Position
100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
D-8
-------
Figure D-9. BSS Example: EV-1, Low EvapHC from TANK, Low Speed, C02 Exh
7 20191020 000506 car 000244
1-EV1 200mg Tank 22mph Y
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
EvapPlume
ExhPlume
Noisel
Noise2
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
25
30
35
40
45
50
55
60
65
"
i
WIJ
I I
111 yfl™
lj ii iill I i |||i Ib
i Ii I illi .in .. ....
hi lilt ill n Jul i l;« ijuJi .
II
!
I
if II i i II : . Vill i
0 100 200
Scan Position
25
30
35
40
45
50
55
60
65
NoiseS
Impi im—ittt
- | [I Kill ill II III ll Pi
Ii I I,' 1 I1! \j!
Ill I
I I nil J
I.U
W UL\
I
I I.L1
III i ll lii|l|
0 100 200
Scan Position
D-9
-------
Figure D-10. BSS Example: EV-1, Medium EvapHC from TANK, High Speed, C02 Exh
7 20191020 000505 car 001347
C02 CO
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
1-EV1 800mg Tank 45mph Y
NO
HUH
5
N02
ffllll
11
I
iWiil TMrillll
III mill ii i
Nlilil
ij pj
i
I
I" II
ml I fill ¦.III [llll
I' ;.,
I fit III! I i''""1*'
HMHL«
,111®
0 100 200
Scan Position
10
15
20
25
30
||IfS|
1
... . Ill II: I II 111 ill J
,| |. I f, Sl f
||J ll[ ll|||| 111
II III I t IJ !1
ll
I I 1 11
III
II III I
1
111 J 'III1
_
0 100 200
Scan Position
EvapPlume
ExhPlume
Noisel
Noise2
Noise3
!D
n
E
3
c
CO
G
C/3
0 100 200
Scan Position
0 100 200
Scan Position
r 1
i. I ii.i
-
¦ifIn! i 1J:lli ii1
i rlijl'i i|!%
1,| M II 111 L
fill II 11 L
H ll J . U I 1)1 H
1
ll| I I I M| I I II |
Jill 1 1 [ll1',
F' l I Mill! Ii I | . I H
0 100 200
Scan Position
0 100 200
Scan Position
h ill i n i l l II , 11
I i . i],i i .Ii 1 Jli 10II
ii i i nil 11ii |ii ll I
l
0 100 200
Scan Position
D-10
-------
Figure D-11. BSS Example: EV-1, High EvapHC from HOOD, Low Speed, C02 Exh
D-ll
-------
Figure D-12. BSS Example: EV-1, Medium EvapHC from HOOD, Low Speed, C02 Exh
7 20191021 000507 car 002554
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
1-EV1 800mg Hood 22mph Y
NO
25
¦ 30
||
35
• 40
45
Mi
o
LO
r
Hill
55
60
65
.
0 100 200
Scan Position
N02
1
Ml
mfci ll lr
,i
SWi
0 100 200
Scan Position
EvapPlume
ExhPlume
Noisel
Noise2
0 100 200
Scan Position
0 100 200
Scan Position
itl Ml!
Ji
,vi: '
I |
(HMIH
1 i ¦;) '1
W
'''I'liili j) 1 i.ll'i'iij I, i:(): )]111 I'!
II fc
I
ii-'i ft iI'Mi;.
Pliill
0 100 200
Scan Position
25
30
35
40
45
50
55
80
65
111 liM
I II II:-
i I ifi.! iVrli i {
""¦til
I i ilJfrlmiiMi ti'
ill i"),1'
1 .
M Ml ' ' 'I' "1 I I 'l I'ill
'' 1V' I'.lull' ' ) 'll'l
v 111
1
i
illii]!l|' i ih|j liiifiv
i a
III |l|l I i
¦
fill
0 100 200
Scan Position
0 100 200
Scan Position
D-12
-------
Figure D-13. BSS Example: EV-1, Low EvapHC from HOOD, Low Speed, C02 Exh
EvapPlume
ExhPlume
Noisel
Noise2
Noise3
OJ
E
25
30
35
40
45
50
55
60
65
0 100 200
Scan Position
100 200
Scan Position
I -II I I fl
itfwWiW
It1® lliSi
11
. i !
,'V
11 I
MMBi—
*wAwm
0 100 200
Scan Position
25
30
35
40
45
50
55
60
65
1 in
I I'll I
i Ifl
i -
I'll V, 1, ... IV 1. I II
ii 'ii iilfl'1,1
111 III 1:11 II III! f||] f
i11 fiW'iW iluii'MVIPU
i . i i-
Ul 1L III J II1. Mi ll
II II
nil l i i 111 lll
iiiii ( ifI III i il i mill Mil r ^
UJI . I IIII,J,ft1111 10 H i J
i
0 100 200
Scan Position
25
30
35
40
45
50
55
60
65
7m
lililH
ilili
till' "HII ii 4 I'm I
II "ill1 1 ! I" I
li l il 1'"\ 111 mi I i !
| i ill] | i I Mi Jl. : ill. I
liiWIifll,
lliH
. Ml pI'll
III
uflnitinmiMiiii
f lMI l l'l
1.1 nimu h.(j >.
0 100 200
Scan Position
D-13
-------
Figure D-14. BSS Example: EV-1, Medium EvapHC from HOOD, High Speed, C02 Exh
D-14
-------
Figure D-15. BSS Example: EV-2, no EvapHC, Low Speed, no Exh
7 20191021 000507 car 001619
C02 CO
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
2-EV2 Omg n/a 22mph N
NO
!
I
II
I
I I
|l ' I nil
0 100 200
Scan Position
N02
T71~"
0 100 200
Scan Position
EvapPlume
ExhPlume
Noisel
Noise2
Noise3
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
D-15
; S
1
h
IIPIB11
miima
Jfl II! ] I'll
Mil liil »1
i i
I ii .
ulllM !i I 1
jjjii)
BL
If"
iilj [ji
III
./Hi
I -
I
0 100 200
Scan Position
5
10
15
20
25
30
35
I :
¦ ilttliltti
ill ' ill i ¦ i;il III i
fli||ll
Mlllill JIM. I i flllP ¦
I,! §;.y ijlliliiii
iinHj i ii ii
11
0 100 200
Scan Position
-------
Figure D-16. BSS Example: EV-2, no EvapHC, Low Speed, HC/C02/C0/N0 Exh
D-16
-------
Figure D-17. BSS Example: EV-2, High EvapHC from DOOR, Low Speed, HC/C02/C0/N0 Exh
D-17
-------
Figure D-18. BSS Example: EV-2, Medium EvapHC from DOOR, Low Speed, HC/C02/C0/N0 Exh
HC
as
¦Q
E
3
05
o
C/3
25
30
35
40
45
50
55
60
65
"
II
¦
I «
hi# (mho
¦
0 100 200
Scan Position
7 20191024 000514 car 000795
C02
0 100 200
Scan Position
2-EV2 800mg Door 22mph Y
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
EvapPiume
ExhPlume
25
30
35
¦ —
i ' <|
i -ilr
0 100 200
Scan Position
100 200
Scan Position
Noisel
i ft
l
.ill i
.,111,1 J
111 ¦ l i' 'V
IIP
111®
¦ '
HHiHi
fuSlr
' ' ¦' H
, ii. i
0 100 200
Scan Position
I J|l|l[ In
i i iii
Noise2
' „
'jfli Fplil'lllJlVl'i
(ill In
i 1 . 1 1
li ' ' 1
ill
0 100 200
Scan Position
25
30
35
40
45
50
55
60
65
!Sloise3
1
:
IliflraeiJwPMl
¦
0 100 200
Scan Position
D-18
-------
Figure D-19. BSS Example: EV-2, Low EvapHC from DOOR, Low Speed, HC/C02/C0/N0 Exh
D-19
-------
Figure D-20. BSS Example: EV-2, Medium EvapHC from DOOR, High Speed, HC/C02/C0/N0 Exh
0 100 200
Scan Position
7 20191021 000507 car 001455
C02
10
15
20
25
30
CO
2-EV2 800mg Door45mph Y
NO
N02
0 100 200
Scan Position
0 100 200
Scan Position
5
10
15
20
25
30
0 100 200
Scan Position
30
0 100 200
Scan Position
D-20
-------
Figure D-21. BSS Example: EV-2, High EvapHC from TANK, Low Speed, HC/C02/C0/N0 Exh
7 20191024 000514 car 001635
-------
Figure D-22. BSS Example: EV-2, Medium EvapHC from TANK, Low Speed, HC/C02/C0/N0 Exh
HC
7 20191024 000514 car 000917
C02
CO
2-EV2 800mg Tank 22mph Y
NO
25
¦ I
25
25
25
25
30
30
30
30
30
35
35
35
35
35
40
40
40
40
40
45
1
"
45
45
45
45
50
50
50
50
50
;
r'WI
55
55
p* ,
55
V 1
55
-------
Figure D-23. BSS Example: EV-2, Low EvapHC from TANK, Low Speed, HC/C02/C0/N0 Exh
D-23
-------
Figure D-24. BSS Example: EV-2, Medium EvapHC from TANK, High Speed, HC/C02/C0/N0 Exh
7 20191021 000507 car 001234
HC
5
10
CD
| 15
§: 20
o
W
25
30
-i 1—
0 100 200
Scan Position
2-EV2 800mg Tank 45mph Y
0 100 200
Scan Position
0 100 200
Scan Position
100 200
Scan Position
0 100 200
Scan Position
EvapPlume
ExhPlume
Noisel
Noise2
Noise3
!D
n
E
3
c
CO
o
C/3
10
15
20
25
30
I.
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
D-24
-------
Figure D-25. BSS Example: EV-2, High EvapHC from HOOD, Low Speed, HC/C02/C0/N0 Exh
7 20191023 000512 car 001416
2-EV2 6400mg Hood 22mph Y
NO
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
N02
:.!i: mm
.. '.I "J.
,ii iiiii ijf.hi111
BBMMfflllPlilHPlB
. I 1 i|!H| i^nJjIjj
II Jill, r
II. II
I ill I I
,1 I,
'."I
ii
0 100 200
Scan Position
EvapPlume
ExhPlume
Noisel
Noise2
NoiseS
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
ill i.
in in in iii.i ii
llllll ii
11Ifl f
ill 1
ill i hill J ipll f
911 ji.'lJ.irtijlt
iini mil!I«li:11 l,| ii,
ml
0 100 200
Scan Position
D-25
-------
Figure D-26. BSS Example: EV-2, Medium EvapHC from HOOD, Low Speed, HC/C02/C0/N0 Exh
0 100 200
Scan Position
7 20191022 000509 car 002751
C02
y
100 200
Scan Position
100 200
Scan Position
2-EV2 800mg Hood 22mph Y
100 200
Scan Position
0 100 200
Scan Position
EvapPlume
ExhPlume
Noisel
Noise2
0 100 200
Scan Position
100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
Noise3
0 100 200
Scan Position
D-26
-------
Figure D-27. BSS Example: EV-2, Low EvapHC from HOOD, Low Speed, HC/C02/C0/N0 Exh
HC
EvapPlume
0 100 200
Scan Position
ExhPlume
Noisel
Noise2
NoiseS
100 200
Scan Position
'In
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
D-27
-------
Figure D-28. BSS Example: EV-2, Medium EvapHC from HOOD, High Speed, HC/C02/C0/N0 Exh
IMoise3
5
10
15
20
25
30
0 100 200
Scan Position
EvapPlume
5
10
0
n
E 15
3
z
c
o 20
M
25
30
0 100 200
Scan Position
ExhPlume
0 100 200
Scan Position
Noisel
0 100 200
Scan Position
Noise2
0 100 200
Scan Position
D-28
-------
Figure D-29. BSS Example: Subaru, High EvapHC from DOOR, Low Speed, natural Exh
7 20191020 000505 car 001495
-------
Figure D-30. BSS Example: Subaru, Medium EvapHC from DOOR, Low Speed, natural Exh
7 20191020 000505 car 002573
3-Subaru 800mg Door22mph Y
0
.Q
£
co
o
W
0 100 200
Scan Position
100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
EvapPlume
ExhPlume
Noisel
Noise2
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
!\loise3
0 100 200
Scan Position
D-30
-------
Figure D-31. BSS Example: Subaru, Low EvapHC from DOOR, Low Speed, natural Exh
HC
tu
.Q
£
co
o
W
25
30
35
40
45
50
55
60
65
0 100 200
Scan Position
7 20191021 000507 car 002191
C02
100 200
Scan Position
0 100 200
Scan Position
3-Subaru 200mg Door 22mph Y
NO
JY ;
i; (
."H' I'"! l|iii:!j|l|ll|,|'!i
| . |i ||
II , I! I
Jf®*
II II I II II I Jj
Kfkm if Mi InfJ i tj I u I
iiiliij
i I II
'
-------
Figure D-32. BSS Example: Subaru, Medium EvapHC from DOOR, High Speed, natural Exh
D-32
-------
Figure D-33. BSS Example: Subaru, High EvapHC from TANK, Low Speed, natural Exh
-------
Figure D-34. BSS Example: Subaru, Medium EvapHC from TANK, Low Speed, natural Exh
D-34
-------
Figure D-35. BSS Example: Subaru, Low EvapHC from TANK, Low Speed, natural Exh
£D
£1
E
cn
o
CO
25
30
35
40
45
50
55
60
65
HC
0 100 200
Scan Position
7 20191021 000507 car 002356
C02
0 100 200
Scan Position
0 100 200
Scan Position
3-Subaru 200mg Tank 22mph Y
NO
25
30
I
I 111 1
1 i Tii -
null!.
I lilimi i
1 11 In
35
1 1 III 1
VTi "¦
m il
40
1
'ill I1 1
li.ml
|tii ii' 1 "
I nil
45
1 MM
it iiii i1 r
50
m i
1 111 hi i i-
liiiihn
55
i
60
"'fill
i
65
BMIji
i
0 100 200
Scan Position
N02
0 100 200
Scan Position
EvapPlume
ExhPlume
Noisel
Noise2
Noise3
a>
E
25
30
35
40
45
50
55
60
65
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
- »I mi
ii i hi i|i
n ii i i
"li *
, f! 11 si
\s
I Mil 11,1,1; 1.^1,11:
I i | il i
i
I II i ill
to,,
r
i i, ,1
0 100 200
Scan Position
0 100 200
Scan Position
D-35
-------
Figure D-36. BSS Example: Subaru, Medium EvapHC from TANK, High Speed, natural Exh
D-36
-------
Figure D-37. BSS Example: Subaru, High EvapHC from HOOD, Low Speed, natural Exh
720191020 000505 car001814
3-Subaru 6400mg Hood 22mph Y
CD
-Q
E
33
O
CO
0 100 200
Scan Position
100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
EvapPlume
ExhPlume
Noisel
Noise2
Noise3
0J
E
0 100 200
Scan Position
100 200
Scan Position
0 100 200
Scan Position
_L
1
III ¦
II 'I,111 ,11"
! Ill
0 100 200 0 100 200
Scan Position Scan Position
D-37
-------
Figure D-38. BSS Example: Subaru, Medium EvapHC from HOOD, Low Speed, natural Exh
7 20191020 000505 car 002430
3-Subaru 800mg Hood 22mph Y
0 100 200
Scan Position
100 200
Scan Position
100 200
Scan Position
100 200
Scan Position
0 100 200
Scan Position
EvapPlume
ExhPlume
Noisel
Noise2
NoiseS
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
D-38
111
|i ill Wil' i u ' ||ll;pt1 ii|. 'j iljjili ji'lt
25
i i i i in-
30
1 i1 iff ! m
iii
i
35
I'JllMiIi
40
Iff;, ft 'Sill
45
'1'
50
i '¦
111 ¦
n 111 ,,11.
55
i
¦ 1 | 'm ll'jLl) j'M; ll
60
''I
65
100 200
Scan Position
0 100 200
Scan Position
-------
Figure D-39. BSS Example: Subaru, Low EvapHC from HOOD, Low Speed, natural Exh
7 20191020 000505 car 002866
3-Subaru 200mg Hood 22mph Y
0 100 200
Scan Position
100 200
Scan Position
0 100 200
Scan Position
JtiMfaif §
|_ I Ml
/11. Jil jfl Wii lit# pi, 11 Hp,
i
0 100 200
Scan Position
N02
iij,['ft aril fflt
i y
i,
II I-
HI
-
0 100 200
Scan Position
EvapPlume
ExhPlume
Noisel
Noise2
Noise3
0 100 200
Scan Position
100 200
Scan Position
25
/Ml
Vljj jjll : 1 l ' llf
25
30
iff'! "
i i'JiJii i i |j_
ii ii.r
30
35
i
,i ii ,li
' 1 ' l" ' l"]
ii iiimi L
35
40
"
'J ll'MV "|!
M
40
45
¦
III"1 ' I'Liii'Jh
in i il i 1 I! i i
45
50
11
„1 'J ,
1 i i ii i H
, "SiiJi
50
55
1 1 ' 1
ii i 'i ii
il ! f\\
i, 1 'I"1!1,
55
60
ii ii
il i' in i
, I1
[J
60
65
ill
III
65
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
D-39
-------
Figure D-40. BSS Example: Subaru, Medium EvapHC from HOOD, High Speed, natural Exh
720191021 000507 car001816 3-Subaru 800mg Hood 45mph Y
HC C02 CO NO N02
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
EvapPlume ExhPlume Noisel Noise2 Noise3
0 100 200 0 100 200 0 100 200 0 100 200 0 100 200
Scan Position Scan Position Scan Position Scan Position Scan Position
D-40
-------
Figure D-41. BSS Example: F-150, High EvapHC from DOOR, Low Speed, natural Exh
D-41
-------
Figure D-42. BSS Example: F-150, Medium EvapHC from DOOR, Low Speed, natural Exh
HC
tu
.Q
£
co
o
W
30
40
50
60
70
if,
[kill 1
f
0 100 200
Scan Position
7 20191023 000512 car 002587
C02
100 200
Scan Position
30
40
50
60
70
CO
11, j. m11—r
11
I jl'lu
' 'I
.'"li
v; f
I ill li I
m1
ii t iii
KM ||
0 100 200
Scan Position
3-F150 800mg Door 22mph Y
NO
N02
0 100 200
Scan Position
0 100 200
Scan Position
EvapPlume
ExhPlume
0
_Q
E
3
C
03
O
CO
30
40
50
60
70
[ii iii j
iiJiti i * I
y i.i
iiiiiii
JY
0 100 200
Scan Position
100 200
Scan Position
30
40
50
60
70
Noisel
iii in 1 i—r
| j: |.;t;h h i m ini» JIM
1 i,j ! 1 !
ii i ii lira
i 1 i,il 1 in ii
i in, I
I I
III' I
111 .
11
1'
i III
"l
1'
1 III"
jjl
:
-------
Figure D-43. BSS Example: F-150, Low EvapHC from DOOR, Low Speed, natural Exh
03
£1
E
3
33
O
CO
0 100 200
Scan Position
7 20191023 000512 car 001956
3-F150 200mg Door 22mph Y
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
EvapPlume
ExhPlume
Noisel
Noise2
Noise3
0J
E
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
D-43
30
40
50
60
I pi i
I ¦ LI 111 111
II I
I ¦ I I I It
!(ftj
ill I
I III |l I 4
ilVl l i/ iVJ m!
i n i n
i i ¦ iii i
slli'J
70
0 100 200
Scan Position
H
30
40
50
I II
, i IV ' lit i
II
mli
60 ¦
70
0 100 200
Scan Position
-------
Figure D-44. BSS Example: F-150, Medium EvapHC from DOOR, Highspeed, natural Exh
D-44
-------
Figure D-45. BSS Example: F-150, High EvapHC from TANK, Low Speed, natural Exh
D-45
-------
Figure D-46. BSS Example: F-150, Medium EvapHC from TANK, Low Speed, natural Exh
0 100 200
Scan Position
7 20191023 000512 car 002719
C02
0 100 200
Scan Position
CO
30
40
50
60
1
70
_
111 1311,1
1,1 W,;
ffi
30
IL r
i' 1 JH
"IV1,!' ii
i i 11
1
$!fr«'.,
40
50
1,1
-i i ' ii inn, , i
iilli,
| ! f" f;| J
illli
60
.
70
0 100 200
Scan Position
3-F150 800mg Tank 22mph Y
NO
30
40
50
60
70
¦ Fin—r
,1 f:f
'
„ ,ir
, ,i i
iMliJI
i ill i in'
ii,
111.
0 100 200
Scan Position
30
40
50
60
70
N02
0 100 200
Scan Position
EvapPlume
ExhPlume
Noisel
Noise2
Noise3
0 100 200 0 100 200
Scan Position Scan Position
~i
30
40
ill
¦ «
i|;T i, ' , I",, !
Li1 "" 1
i'i i , i til
50
60
70
,
I.'] |
il i! i'i
f r II'IJ! to
, III I i II
I lli i .i; ii i i l
, i,Ii ii' I win!,! 1
_i i_
0 100 200
Scan Position
D-46
30
40
50
60
70 -
-1—1 i i—i mi in jui
3!
i i i i (i
!:| 1 11 , I
iiiiii (~}ili
M!1
, iii'
) I ,11 ,l|;
III,.' 11,1
II 1 I
0 100 200
Scan Position
30
40
50
60
70 ¦
0 100 200
Scan Position
-------
Figure D-47. BSS Example: F-150, Low EvapHC from TANK, Low Speed, natural Exh
7 20191023 000512 car 002207
0 100 200
Scan Position
100 200
Scan Position
0 100 200
Scan Position
3-F150 200mg Tank 22mph Y
NO N02
30 L
40
50
60
70 ¦
I II I I
H ,VY
'' nil'. II
i , in/ nil
i In li i ig.i
in ii ii
0 100 200
Scan Position
0 100 200
Scan Position
EvapPlume
ExhPlume
Noisel
Noise2
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
Noise3
0 100 200
Scan Position
D-47
-------
Figure D-48. BSS Example: F-150, Medium EvapHC from TANK, Highspeed, natural Exh
EvapPlume
ExhPlume
Noisel
Noise2
Noise3
0 100 200
Scan Position
5
5
10
10
15
15
20
20
25
25
30
V
30
35
35
0 100 200
Scan Position
iii i n
, m ,
! I III
I III! . r, . 1-Iimt „
I I ,lf .. 11 ! If Hi
I'll ' I 'Ii 'h
i .I
! ' ! Ill
l II I
I !
Ill | J . ,1 j L J I 111
I.I I'll 111 11 III-
p ii ii Tj mi i
i i ii i iii in
1 i hi/1 i
r 11
i. mi
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
D-48
-------
Figure D-49. BSS Example: F-150, High EvapHC from HOOD, Low Speed, natural Exh
7 20191023 000513 car 000272
C02
3-F150 6400mg Hood 22mph Y
0 100 200
Scan Position
100 200
Scan Position
CO
0 100 200
Scan Position
0 100 200
Scan Position
N02
L
[i
i i
i i i'I
iihi.J f
i mil
1
(III
IV1
il
0 100 200
Scan Position
EvapPlume
ExhPlume
0 100 200
Scan Position
100 200
Scan Position
Noisel
0 100 200
Scan Position
Noise2
0 100 200
Scan Position
NoiseS
i i' I
1(1 I, :
I j I 1.1 I'
it ' I -fruit
il! I I
0 100 200
Scan Position
D-49
-------
Figure D-50. BSS Example: F-150, Medium EvapHC from HOOD, Low Speed, natural Exh
7 20191023 000512 car 002437
03
£1
E
3
33
O
CO
0 100 200
Scan Position
3-F150 800mg Hood 22mph Y
NO N02
100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
EvapPlume
ExhPlume
Noisel
Noise2
Noise3
0J
E
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
D-50
30
40
50
60
70
I ¦ jh II ¦ I fi Miff ill j |I |
i'l ! I ]i- e|u
111F
I.',' ! Aiftijj
uili til I IjiMilftPI
Ei D HI
f fi M
i fl
i ; II.
I ,,11 I
!| III
ill
i , If' II jy
II11 nl; liiill'i
¦ i
0 100 200
Scan Position
0 100 200
Scan Position
-------
Figure D-51. BSS Example: F-150, Low EvapHC from HOOD, Low Speed, natural Exh
7 20191023 000512 car 001835
tu
.Q
£
co
o
W
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
3-F150 200mg Hood 22mph Y
NO
N02
n r—r
i'i .it 11
'I i
1 1!
30
' 1,1 ii'Vl
ill I
lii; ww
40-1
!W' n
11 ' I I III
ttl I' I ill! I I I III
I I II
50
60
70
: I
i inil
'' i| i
i mil i
0 100 200
Scan Position
30
40
50
60
70
1
i.J! iiif'l
-
1
1 1 -
',,J j'lltI'iVI
1 1 !|,I In
1 tf, ™
;1Ji
'> 1 ht' 'I1
v 1 ''" ¦ i1114 ji'11
• ir1' 'u!
!j (I'' Si'j'il'li
i s
1 1
11
1
1
1 1 II I II
0 100 200
Scan Position
EvapPlume
ExhPlume
Noisel
n
E
3
c
m
o
CO
0 100 200
Scan Position
100 200
Scan Position
30
i
11
40
50
60 ¦
70
if 111
it V V
1
ii
iljipliiffi,1'
i
t
1H
pi
h
0 100 200
Scan Position
D-51
Noise2
NoiseS
0 100 200
Scan Position
30
40
50
60
70
' ~71
i H J i,I ' 'I1!11',1 1 11 ,'WJ
,1 ' 1 .Nil ll
1 ii V %i 'i." iii:i
I I I !l Ijl I
/1 1 'I ;v
I-1
i 'i i'i1M
0 100 200
Scan Position
-------
Figure D-52. BSS Example: F-150, Medium EvapHC from HOOD, Highspeed, natural Exh
D-52
-------
Appendix E:
Examples of Blind Source Separation
with Correlation Constraints (BSScov)
E-0
-------
Figure E-1. BSS ICA Separation (p=0) of Example:
7 20191020 000505 car 001188
EV-1, High EvapHC from TANK
1-EV1 6400mg Tank 22mph Y
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0)
XI
E
EvapPlume
5
10
15
20
n i[j|f i^Tj
I I I I
I'll ' II , Ill 11
I'll I
cu
o
w 25
30
35
1 1 1 V
„ VI ill
I il l,, riii "
'
(ir J
i i
iilill
I [ill
ii
0 100 200
Scan Position
rho = 0
ExhPlume
Noisel
0 100 200
Scan Position
0 100 200
Scan Position
Noise2
TFr,TTiii ....
!
' tfcrf Ik
1, 1 11 I
i i,
mi ,u / • I
iJilo,
,1
1 I I !Jl J
III I
1 I" I"1
,M la#l
l'l llllllll
Iwfi'l I
Hi m
j i , ,
0 100 200
Scan Position
Noise3
5
ill' Ii ii
10
I
" |l||)
"
15
20
l!
I ' I
25
1
30
-
'ill! Ill
¦1
35
-
i.
0 100 200
Scan Position
E-1
-------
Figure E-2. BSScov Separation (p=0.1) of Example: EV-1, High EvapHC from TANK
7 20191020 000505 car 001188 1-EV1 6400mg Tank 22mph Y
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
EvapPlume
rho = 0.1
ExhPlume
Moisel
Noise2
IMoiseS
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
E-2
-------
Figure E-3. Evaluation of Plume Outputs while Varying p for Example: EV-1, High EvapHC from TANK
7 20191020 000505 car 001188 1-EV1 6400mg Tank 22mph Y
HC
5
10
'5 15
E
I 20
e
re
o
w 25
30
35
l I'll'I J
i I
III
IWl'l s: !i
, ,11 „
I
i H
0 100 200
Scan Position
EvapPlume
5
10
15
20
25
30
35
ii I III III
ir. m
.Mil
"lilt n,
II I
Pill
I I
I li .
Ill
0 100 200
Scan Position
5
10
15
20
25
30
35
EvapPlume, p=0.05 EvapPlume, /j=0,10 EvapPlume, p=0,15
5
II
T
I '
I
I 11 i III
illii [ 11 If. Hi
ill
fil
i'1!" i
fllli'iii1
0 100 200 0 100 200 0 100 200
Scan Position Scan Position Scan Position
E-3
-------
Figure E-4. BSS ICA Separation (p=0) of Example: EV-2, High EvapHC from DOOR
7 20191023 000512 car 001168
CD
.Q
E
CD
O
CO
2-EV2 6400mg Door 22mph Y
NO
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
N02
11
i
Hill
Hi
0 100 200
Scan Position
CD
rs
E
c
co
o
CO
EvapPlume
rho = 0
ExhPlume
0 100 200
Scan Position
0 100 200
Scan Position
Noisel
jaw
i i
Hi
0 100 200
Scan Position
E-4
Noise2
0 100 200
Scan Position
Noise3
1—
I
l i
0 100 200
Scan Position
-------
Figure E-5. BSScov Separation (p=0.1) of Example: EV-2, High EvapHC from DOOR
7 20191023 000512 car 001168 2-EV2 6400mg Door 22rnph Y
E
c
03
o
CO
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
5D
,Q
E
C
cc
o
CO
EvapPlume
rho = 0.1
ExhPlume
Noisel
Noise2
100 200
Scan Position
100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
NoiseS
0 100 200
Scan Position
E-5
-------
Figure E-6. Evaluation of Plume Outputs while Varying p for Example: EV-2, High EvapHC from DOOR
7 20191023 000512 car 001168 2-EV2 6400mg Door 22mph Y
EvapPlume
EvapPlume, p=0.05 EvapPlume, /j=0,10 EvapPlurne, p=0,15
il
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
Kill
llll l
1—[
'I ,1'J
I
III
0 100 200
Scan Position
a)
X2
E
ro
o
W
ExhPlume
ExhPlume, /;=0.05 ExhPlume, p=
5
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0.10 ExhPlume, p=0.15
0 100 200
Scan Position
0 100 200
Scan Position
E-6
-------
Figure E-7. BSS ICA Separation (p=0) of Example: EV-2, High EvapHC from TANK
7 20191023 000512 car 000926 2-EV2 6400mg Tank 22mph Y
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0)
n
E
3
EvapPlume
5
10
15
20
03
CO 25
30
35
,i
I HI
0 100 200
Scan Position
rho = 0
ExhPlume
0 100 200
Scan Position
Noisel
Noise2
NoiseS
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
E-7
-------
Figure E-8. BSScov Separation (p=0.15) of Example: EV-2, High EvapHC from TANK
7 20191023 000512 car 000926 2-EV2 6400mg Tank 22mph Y
.Q
E
e
a:
o
CO
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
EvapPlume
rho = 0.15
ExhPlume
0 100 200
Scan Position
0 100 200
Scan Position
Noisel
0 100 200
Scan Position
Noise2
0 100 200
Scan Position
Noise3
0 100 200
Scan Position
E-8
-------
Figure E-9. Evaluation of Plume Outputs while Varying p for Example: EV-2, High EvapHC from TANK
7 20191023 000512 car 000926 2-EV2 6400mg Tank 22mph Y
EvapPlume EvapPlume, p=0.05 EvapPlume, p=0.10 EvapPlume, p=0,15
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
E-9
-------
Figure E-10. BSS ICA Separation (p=0) of Example: EV-2, High EvapHC from HOOD
2-EV2 6400mg Hood 22mph Y
7 20191023 000512 car 001416
C02
CO
NO
N02
5
10
15
20
25
30
35
M,
5
10
15
20
25
30
E~9
0 100 200
Scan Position
EvapPlume
0 100 200
Scan Position
rho = 0
ExhPlume
0 100 200
Scan Position
Noisel
0 100 200
Scan Position
Noise2
0 100 200
Scan Position
Noise3
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
E-10
-------
Figure E-11. BSScov Separation (p=0.1) of Example: EV-2, High EvapHC from HOOD
7 20191023 000512 car 001416 2-EV2 6400mg Hood 22mph Y
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
EvapPlume
0 100 200
Scan Position
rho = 0.1
ExhPlume
0 100 200
Scan Position
Noisel
0 100 200
Scan Position
Noise2
5
.
"
I II i ii !!' 1 1
5
MIIM .
I III
10
•
10
II lllli
15
¦
A! II
15
20
¦
20
II I llll Ill I
25
i i
25
' "i "i
I! I'lil I
1
30
IN
30
1. 1
35
¦
35
I I
0 100 200
Scan Position
Noise3
5
10
15
20
25
30
35
1
1
ML
ill!
"i r nil ilifii nii
' I
I
mil
J|])(i, ||[ I;. U I l U
Ml IBP l1 (i in
J it
III 1M
mi
ispi
i
ill I l]ll I " 1
I IPiillJip
flwlBi I
llMi,
i all
#11 ilfi)
ifiiy
1
' ;U
UK
L
-
0 100 200
Scan Position
E-11
-------
Figure E-12. Evaluation of Plume Outputs while Varying p for Example: EV-2, High EvapHC from HOOD
7 20191023 000512 car 001416 2-EV2 6400mg Hood 22mph Y
EvapPlume EvapPiume, ^=0.05 EvapPlume, jy=0.10 EvapPlume, ^=0.15
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
a>
E
3
C
CD
O
to
C02
0 100 200
Scan Position
ExhPlume ExhPlume, p=0.05 ExhPlume, /;=0.10 ExhPlume, p=0.15
i i i i i i i i " i
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
0 100 200
Scan Position
E-12
-------
Appendix F:
Comparing Westminster and MOVES Release Rates
by Class and Age Group
F-0
-------
Figure F-1. Comparison of Release Rates for O-to-3-Year-Old LDTs in 2019
1000.0
o>
¦5 100.0
ro
a»
a>
a:
x
o
o
O
10.0
1.0
0.1
1000
10000
• *•
* * *
. .. y.
¦ \ ¦ ''."¦"¦.AO
• i-l jS
.*• . '** ¦: * *
> ' . 'v•.' .v
• i «• *•
. * » .». • .*• ' • .
a> 1000
a>
*¦>
ro
ac
a>
(/>
to
CL
100
o
u
10
3
a>
+¦»
ra
a:
4>
(/>
ro
_a»
* * • ¦
... . ¦. .
- \J • ¦ • i ' .
• <¦ *•.*>•'••„ ¦ .
-• •- :xvv..^-,y. .. •
10000
100000
*¦ "¦ •;
1**0 i*
. •• • '
. ¦ . . "* .v V-
• ' • •• '¦••• .- y.; '?"«y.."'.':
/* 1
** * • .* * * *
, *' ¦
100000
10000
C02 Release Rate (g/hr)
Source: ••• Westminster Vehicle RSD Transits MOVES Average for Calendar 2019
/projl/EDARinDenver-OCT2D 19/Analysis_MLout/220113/Anal_MLout/OCT19_VSPbiris_4_CP.sas 02MAR23 13:09
F-1
-------
Figure F-2. Comparison of Release Rates for 4-to-5-Year-Old LDTs in 2019
_ 1000.0i =
ro
a>
¦£ 100.0
(0
CL
<0
V)
n5
aj 10.0
a>
ac
x
O
o
O
v
i/>
ro
0)
CL
O
U
. v
•. .ir
. . ¦: «¦ yjv. ~ •
V- ¦' .v.
0.1 ^
1000
10000
100000
a> 1000
a>
a:
100
10
u>
a>
m
a
a»
ro
_0>
a>
a:
u
x
LLI
1
1000
1000.0
100.0
. \ •• a m ;
• * * * • / r . l." .
• *
./ ¦ '• • •
. . * * ••
. • / • . * V- .
. ,/ . . ¦ ...t
. •. ¦ v • >v'
. : -*. :') -
•' "¦ ¦
v¦' ' .
10000
100000
10.0
1.0
v • ;v.-
¦ ;V .
0.1
1000
100000
10000
C02 Release Rate (g/hr)
Source: • • • Westminster Vehicle RSD Transits MOVES Average for Calendar 2019
/projl/EDARinDenver-OCT2D 19/Analysis_MLout/220113/Anal_MLout/OCT19_VSPbiris_4_CP.sas 02MAR23 13:09
F-2
-------
Figure F-3. Comparison of Release Rates for 6-to-7-Year-Old LDTs in 2019
1000.0
3
(/>
ns
aj io.o
a>
ac
x
O
o
O
1.0
TO
ro
CL
re
aj
•*-*
rrs
a:
a>
i/>
ro
a»
a:
u
x
LLI
1000
1000.0
100.0
10.0
1.0
0.1
1000
10000
100000
\
¦ ' * v y l
• .••¦¦¦ •- ¦ •
.. " A :0.. .
. y.. • * ••
I-?}'¦
• * . I#*-* S #* ~.
• C .""i
. • X_v< t.*\*
• * . ' '' ' ¦
10000
100000
't: • ;
• r
••, \ \
«." ' * ' *.
100000
10000
C02 Release Rate (g/hr)
Source: ••• Westminster Vehicle RSD Transits MOVES Average for Calendar 2019
/projl/EDARinDenver-OCT2D 19/Analysis_MLout/220113/Anal_MLoul/OCT19_VSPbiris_4_CP.sas 02MAR23 13:09
F-3
-------
Figure F-4.
Comparison of Release Rates for 8-to-9-Year-Old LDTs in 2019
5
v
-M
a.
0)
I/)
ns
ju
V
a:
x
o
z
0
o
z
1000.0
100.0
10.0
1.0
• • • •
• " . • " . .
v- =. • ..
... V\ -
A.; ¦ - ¦ '
¦* ' * .. V i .•
» . • ¦ • .
¦ • f ,
re
a:
v
v>
ro
ra
a:
"•••; - :r • •
• • •
.«/ ; •.
¦ .'¦¦¦'• • jK'T' ¦¦ '"'¦ ¦
' r ¦¦ *1 / .Nf t .
100000
10000
C02 Release Rate (g/hr)
Source: ••• Westminster Vehicle RSD Transits MOVES Average for Calendar 2019
/projl/EDARinDenver-OCT2D 19/Analysis_MLout/220113/Anal_MLout/OCT19_VSPbiris_4_CP.sas 02MAR23 13:09
F-4
-------
Figure F-5. Comparison of Release Rates for 10-to-14-Year-Old LDTs in 2019
^ 1000.0
.C
01
v
+¦<
re
cn
v
vt
ro
v
¦ * »
100
10
1
1000
1000.0
3
ns
:. ^ ¦ :
10000
100000
.j ' •
; •*.' .. ¦ t •: \ * •'
¦ ** • *
t ; / ' •
•\£ * .* ,.;-A •*'-f'*_** •
' \z ¦ '•
100000
10000
C02 Release Rate (g/hr)
Source: ••• Westminster Vehicle RSD Transits MOVES Average for Calendar 2019
/projl/EDARinDenver-OCT2D 19/Analysis_MLout/220113/Anal_MLout/OCT19_VSPbiris_4_CP.sas 02MAR23 13:09
F-5
-------
Figure F-6. Comparison of Release Rates for 15-to-19-Year-Old LDTs in 2019
_ 1000.0
£
5)
re
a:
" '•
.. / ;• •. •.\ -4?-!>C ¦ V- !.v"' • •
> •'• *.• . ••
; ,
i" ¦' •
10000
100000
. /
-i# •. •
100000
10000
C02 Release Rate (g/hr)
Source: ••• Westminster Vehicle RSD Transits MOVES Average for Calendar 2019
/projl/EDARinDenver-OCT2D 19/Analysis_MLout/220113/Anal_MLoul/OCT19_VSPbiris_4_CP.sas 02MAR23 13:09
F-6
-------
Figure F-7. Comparison of Release Rates for 20-to-99-Year-Old LDTs in 2019
.c
3
re
a:
(/)
re
_aj
-------
Figure F-8. Comparison of Release Rates for O-to-3-Year-Old LDVs in 2019
1000.0
.£
o) 1000
V
-M
a:
-------
Figure F-9.
Comparison of Release Rates for 4-to-5-Year-Old LDVs in 2019
^ 1000.0
£
01
100.0
+¦>
re
a:
re
£
-------
Figure F-10. Comparison of Release Rates for 6-to-7-Year-Old LDVs in 2019
^ 1000.0
;C
3
0)
¦S 100.0-
Q1
- - ¦¦¦'• '• , '¦ ' ' '
' ' '• *,
? • • .
¦ ¦4 : ¦ .
.*•'.¦•.¦'•.~-¦•7 '
. > ' •• •
.£
"oi 1000
(V
-M
a:
v
v>
ro
ra
a:
-------
Figure F-11. Comparison of Release Rates for 8-to-9-Year-Old LDVs in 2019
^ 1000.0
.n
O)
re
c*
v
(/)
re
-------
Figure F-12. Comparison of Release Rates for 10-to-14-Year-Old LDVs in 2019
1000.0
£
ai
oj
¦£ 100.0
a:
¦ t ,s .
V;-"-"' •. ¦'
100000
5
•*-*
nj
cm
v
i/i
re
_aj
v
a:
u
x
.n
x
LU
1000
1000.0
100.0
10.0
y. '('j.v ,^y v' ¦ .* *
•/ ••• . ,?• i1*¦¦."OA"'-.'¦ ¦
. . ; ; r*-*—¦—/»• . —i' s\ .'«/¦«—¦—
I • • ' " j ¦ ^...v '
I •• ¦*¦ ? * *
/ ¦ ' s.-'s. ¦ »/" •
10000 100000
1.0
0.1
1000
.r
I
'¦••'a > •
/
"A\ ^l/C . ?V y.. .. . •
"!a «.. ;.v)v .¦
; .*: ' • ' .
Source:
10000 100000
C02 Release Rate (g/hr)
1 Westminster Vehicle RSD Transits MOVES Average for Calendar 2019
/projl/EDARinDenver-OCT2D 19/Analysis_MLout/220113/Anal_MLoul/OCT19_VSPbins_4_CP.sas 02MAR23 13:00
F-12
-------
Figure F-13. Comparison of Release Rates for 15-to-19-Year-Old LDVs in 2019
C02 Release Rate (g/hr)
Source: ••• Westminster Vehicle RSD Transits MOVES Average for Calendar 2019
/projl IE DARinDenver-OCT2D 19/Analysis_M Lout/220113/Anal_MLout/OCT19_VSPbins_4_CP.sas 02MAR23 13:09
F-13
-------
Figure F-14. Comparison of Release Rates for 20-to-99-Year-Old LDVs in 2019
^ 1000.01
3
v
+¦»
ft
en
re
0)
Q)
en
X
O
o
O
100.0
.C
3
01
*¦>
ro
a.
a)
i/i
re
_aj
0)
a.
o
u
3
a>
+¦>
re
a:
a>
V)
re
_aj
a>
ft
U
X
LLI
10000
C02 Release Rate (g/hr)
100000
Source: ••• Westminster Vehicle RSD Transits MOVES Average for Calendar 2019
/proj1/EDARinDenver-OCT2D19/Analysis_MLout/220113/Anal_MLout/OCT19_VSPbins_4_CP.sas 02MAR23 13:09
F-14
------- |