Emissions Modeling and Implentation into CORSIM

MISSING IMAGE

Material Information

Title:
Emissions Modeling and Implentation into CORSIM
Physical Description:
1 online resource (85 p.)
Language:
english
Creator:
Hulsberg, Jack W
Publisher:
University of Florida
Place of Publication:
Gainesville, Fla.
Publication Date:

Thesis/Dissertation Information

Degree:
Master's ( M.E.)
Degree Grantor:
University of Florida
Degree Disciplines:
Civil Engineering, Civil and Coastal Engineering
Committee Chair:
WASHBURN,SCOTT STUART
Committee Co-Chair:
ELEFTERIADOU,AGELIKI

Subjects

Subjects / Keywords:
corsim -- emissions -- fuel
Civil and Coastal Engineering -- Dissertations, Academic -- UF
Genre:
Civil Engineering thesis, M.E.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract:
Transportation accounts for much of the world’s energy use and emissions. Transportation professionals are becoming more and more concerned about air quality, and there needs to be an accurate way to predict emissions on a micro-scale basis. This project used the traffic micro-simulation program, CORSIM, and it changed the way that CORSIM predicts emissions.   CORSIM currently uses lookup tables for fuel use and emissions. The program looks up the speed, acceleration, and road grade for every second and assigns an emissions value to each vehicle for each second. The data for these values come from research conducted in the 1980’s. Also, CORSIM does not account for cold starts, which can contribute to up to 40% of trip-based emissions. In this project, the current emissions and fuel use estimation method was replaced by the vehicle specific power (VSP) method. In this method, the VSP, which is a measure of engine load, was calculated for each second for every vehicle in the simulation. VSP was separated into 14 different modes, and CORSIM assigned emissions and fuel use rates based on which VSP mode the vehicle was in. The CORSIM code was also expanded to account for cold starts.  The VSP method was verified using code written in C#, and the cold start method was easily verifiable by hand calculations.  Users of CORSIM will now be able to view emissions that are coming from data that is much more up to date, and addition of the cold start method will prevent CORSIM from underestimating emissions on networks with many cold starts.
General Note:
In the series University of Florida Digital Collections.
General Note:
Includes vita.
Bibliography:
Includes bibliographical references.
Source of Description:
Description based on online resource; title from PDF title page.
Source of Description:
This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility:
by Jack W Hulsberg.
Thesis:
Thesis (M.E.)--University of Florida, 2013.
Local:
Adviser: WASHBURN,SCOTT STUART.
Local:
Co-adviser: ELEFTERIADOU,AGELIKI.

Record Information

Source Institution:
UFRGP
Rights Management:
Applicable rights reserved.
Classification:
lcc - LD1780 2013
System ID:
UFE0046393:00001


This item is only available as the following downloads:


Full Text

PAGE 1

1 EMISSIONS MODELIN G AND IMPLENTATION INTO CORSIM By JACK HULSBERG A THESIS PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF ENGINEERING UNIVERSITY OF FLORIDA 2013

PAGE 2

2 2013 Jack Hulsberg

PAGE 3

3 To my p arents

PAGE 4

4 ACKNOWLEDGMENTS I thank Dr. Scott Washburn (my supervisory committee chair) and my other committee m ember s Dr. David Hale and Dr. Lily El efteriadou for their mentoring and support. I would like to thank my fiance, Sarah for putting a smile on my face every day. I would finally like to thank my parents for their lov e and support.

PAGE 5

5 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ 4 LI ST OF TABLES ................................ ................................ ................................ ........... 7 LIST OF FIGURES ................................ ................................ ................................ ........ 8 ABSTRACT ................................ ................................ ................................ .................. 10 CHAPTER 1 INTRODUCT ION ................................ ................................ ................................ ... 12 Background ................................ ................................ ................................ ........... 12 Problem Statement ................................ ................................ ................................ 13 Research Objectives an d Tasks ................................ ................................ ............ 13 Document Organization ................................ ................................ ......................... 14 2 LITERATURE REVIEW ................................ ................................ ......................... 15 Revie w of Current CORSIM Emissions and Fuel Use Modeling ............................ 15 Overview of Emissions Measurement Techniques ................................ ................. 17 Examination of Cold Start s ................................ ................................ .................... 17 Comparing Emissions for Different Driving Modes ................................ ................. 20 Experiment ................................ ................................ ................................ ...... 20 Results ................................ ................................ ................................ ............ 22 Conclusions ................................ ................................ ................................ .... 25 VSP Approach to Classifying Fuel Use and Emissions ................................ .......... 27 Experimental Section ................................ ................................ ...................... 27 Results ................................ ................................ ................................ ............ 29 3 RESEARCH APPROACH ................................ ................................ ...................... 38 Proposed Approach ................................ ................................ ............................... 38 Task 1 Fuel and Emissions Model Extensions ................................ .................... 38 Task 2 Data Collection and Reduction ................................ ................................ 40 Instrumentation ................................ ................................ ............................... 40 Gainesville Data Collection ................................ ................................ ............. 41 Orlando Data Co llection ................................ ................................ .................. 43 Synchronization ................................ ................................ ............................... 44 Screening ................................ ................................ ................................ ........ 44 Processing ................................ ................................ ................................ ...... 44 Task 3 VSP Model Implementation and Testing in CORSIM .............................. 45 Task 4 Cold Start Data Analysis and Implementation into CORSIM ................... 46

PAGE 6

6 4 RESULTS AND ANALYSIS ................................ ................................ ................... 50 Testing of I 4 Route in CORSIM ................................ ................................ ............ 50 Inputs ................................ ................................ ................................ .............. 50 Comparing Emissions Totals for Separate Runs ................................ ............. 51 Comparing Percent Time Spent in Each VSP Mode ................................ ........ 52 Testing of Newberry Road Route in CORSIM ................................ ........................ 54 5 SUMMARY AND RECOMMENDATIONS ................................ .............................. 65 APPENDIX: FIGURES F ROM DATA COLLECTION ................................ .................... 69 LIST OF REFERENCES ................................ ................................ .............................. 83 BIOGRAPHICAL SKETCH ................................ ................................ ........................... 85

PAGE 7

7 LIST OF TABLES Table pag e 3 1 CORSIM NG Predicted Emissions (mg) ................................ ............................ 47 3 2 CORSIM 6.3 Predictions with VSP Implementation (mg/mile) ........................... 47 3 3 Gainesville Route Driving Times ................................ ................................ ........ 47 3 4 Driving Times for Portion of Gainesville Route Used for Testing ....................... 48 3 5 Orlando Route Driving Times ................................ ................................ ............ 48 3 6 VSP Modal Fuel Use and Emissions Rates ................................ ....................... 48 3 7 Average Fuel Use and Emission Rates for 15 Vehicles ................................ ..... 49 3 8 Empirical Cold Start Excess: Average for 30 Vehicles ................................ ....... 49 4 1 Empirical Data (I 4 segment) ................................ ................................ ............. 57 4 2 VSP Model Hand Calculations (I 4 segment) ................................ ..................... 57 4 3 CORSIM Predictions with VSP Implementation (I 4 Segment) .......................... 58 4 4 Simulation Travel Times (I 4 Segment) ................................ ............................. 58 4 5 Average Difference from Field Vehicle for All Modes (I 4 segment) ................... 58 4 6 Average VSP Mode (I 4 Segment) ................................ ................................ .... 59 4 7 Simulation Travel Times (Newberry) ................................ ................................ 59 4 8 Average Difference From Field Vehicle Average for All Modes (Newberry) ....... 59 4 9 Average VSP Mode (Newberry) ................................ ................................ ........ 60

PAGE 8

8 LIST OF FIGURES Figure page 2 1 CO emission rate and coolant temperature over time for a cold start ................ 33 2 2 Average emission rate for each mode over all trips in dataset ........................... 34 2 3 Time traces of vehicle speed, emission rates, and fuel consumption for a 1999 Ford Taurus driven on Chap el Hill Road on August 29, 2000 ................... 35 2 4 ECDF of CO emissions rates in acceleration mode based on 141 trips ............. 36 2 5 Average modal emission rates for 141 trips by a 1999 Ford Taurus .................. 36 2 6 ECDF for average emissions rates for two different drivers operating 1999 Ford Taurus on Chapel Hill Road ................................ ................................ ...... 37 2 7 Definition of VSP modes and average emissions and fuel use rates for a 2005 Chevrolet Cavalier 2.2L ................................ ................................ ............ 37 4 1 Run 1 VSP mode frequencies (I 4 segment) ................................ ..................... 60 4 2 Run 2 VSP mode frequencies (I 4 segment) ................................ ..................... 61 4 3 Run 3 VSP mode frequencies (I 4 segment) ................................ ..................... 62 4 4 VSP mode frequencies, all field runs (Newberry Road segment) ...................... 63 4 5 VSP mode frequencies, simulated vehicl es and average of field vehicles (Newberry Road segment) ................................ ................................ ................ 64 A 1 Gas zeroing tubes and weather station ................................ ............................. 69 A 2 PEMS device connection running from side window to tailpipe ......................... 69 A 3 Side view of instrumented Honda Pilot ................................ .............................. 70 A 4 PEMS connection to tailpipe ................................ ................................ .............. 70 A 5 OBD connection ................................ ................................ ................................ 71 A 6 GPS devices ................................ ................................ ................................ ..... 7 1 A 7 PEMS computer and laptop connected to OBD ................................ ................. 72 A 8 Gainesville routes ................................ ................................ ............................. 73 A 9 University Avenue near starting point ................................ ................................ 73

PAGE 9

9 A 10 Newberry Road near Oaks Mall ................................ ................................ ......... 74 A 11 Ramp to get on I 75 N ................................ ................................ ....................... 74 A 12 US 441 ................................ ................................ ................................ .............. 75 A 13 Right turn onto NW 34 th Blvd ................................ ................................ ............. 75 A 14 NW 34 th Blvd ................................ ................................ ................................ ..... 76 A 15 NW 39 th Ave ................................ ................................ ................................ ...... 76 A 16 Right turn onto NE 15 th St ................................ ................................ .................. 77 A 17 NE 15 th St ................................ ................................ ................................ .......... 77 A 18 NE 16 th Ave ................................ ................................ ................................ ....... 78 A 19 Left onto NW 13 th St ................................ ................................ .......................... 78 A 20 NW 13 th St ................................ ................................ ................................ ......... 79 A 21 Right onto University Ave ................................ ................................ .................. 79 A 22 Orlando route ................................ ................................ ................................ ... 80 A 23 On ramp for I 4 E at S Orange Blossom Trail ................................ .................... 80 A 24 Merging onto I 4 E from S Orange Blossom Trail ................................ .............. 81 A 25 I 4 E ................................ ................................ ................................ .................. 81 A 26 Ramp to get on I 4 W from Maitland Blvd ................................ .......................... 82 A 27 I 4 W ................................ ................................ ................................ ................. 82

PAGE 10

10 Abst ract of Thesis Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Master of Engineering EMISSIONS MODELIN G AND IMPLENTATION INTO CORSIM By Jack Hul sberg December 2013 Chair: Scott Washburn Major: Civil Engineering Transportation p rofessionals are becoming more and more concerned about air quality, and there needs to be an a ccurate way to predict emissions on a micro scale basis. This project used the traffic micro simulation program, CORSIM, and it changed the way that CORSIM predicts emissions. CORSIM currently uses lookup tables for fuel use and emissions. The program loo ks up the speed, acceleration, and road grade for every second and assigns an emissions value to each vehicle for each second. The data for these values come from Also, CORSIM does not account for cold starts which can co ntribute to up to 40% of trip based emissions In this project, the current emissions and fuel use estimation method was replaced by the vehicle specific power (VSP) method. In this method, the VSP, which is a measure of engine load, was calculated for eac h second for every vehic le in the simulation. VSP was separated into 14 different modes, and CORSIM assign ed emissions and fuel use rates based on which VSP mode the vehicle was in. The CORSIM code was also expanded to account for cold starts.

PAGE 11

11 The VSP met hod was verified using code written in C#, and the cold start method was easily v erifiable by hand calculations. Users of CORSIM will now be able to view emissions that are coming from data that is much more up to date, and addition of the cold start meth od will prevent CORSIM from underestimating emissions on networks with many cold starts.

PAGE 12

12 CHAPTER 1 INTRODUCTION Background Transportation accounts for 28% of all U.S. energy use (EIA, 2006). Highway transportation accounts for 32% of national annual emissi ons of nitrogen oxides (NO x ), 50% of CO, and 22% of volatile organic compounds (VOC) (EPA, 2012). In the latest draft Strategic Plan, the USDOT recognized the environmental impacts of the transportation system. Two of its five goals dealt with livability a nd environmental sustainability (USDOT 2012). Since so much of energy use and emissions comes from transportation, there needs to be an accurate way to predict the fuel use and emissions for vehicles along certain routes. The Energy Use and Emissions (EU &E) estimation method that will be used in this project is based on Vehicle Specific Power (VSP), the same concept used in the EPA Motor Vehicle Emissions Simulator (MOVES) (EPA 2004). VSP is a vehicle activity measure of engine load. VSP is separated int o a discrete number of modes, which are correlated to fuel use and emission rates per unit time (Frey et al, 2002). R esearch ers from North Carolina State University, most notably Dr. Christopher Frey, have developed VSP calibrated fuel and emission models for light duty vehicles, transit vehicles, and heavy trucks. While VSP estimates emissions for running and idling modes, it does not account for the effect of cold starts, which can contribute to up to 40% of trip based emissions. To estimate fuel and emi ssions in a traffic micro simulator, one needs estimations of instantaneous speed, acceleration, road grade, veh icle class, and vehicle starts. If all of this information is available, the VSP calibrated models can be used to

PAGE 13

13 predict emissions and fuel use In traffic micro simulators, like CORSIM, the speed, acceleration, road grade, and vehicle class are all known. Therefore, if the VSP model could be implemented into CORSIM, it could be used to predict emissions. The cold start emissions would also have to be taken into account in CORSIM, as the current version does not consider the effect of cold starts. Problem Statement With the increasing concern for air quality, there is a need for accurate measurements or estimates of micro scale vehicle Energy Use and Emissions. The traffic simulation program used for this project will be CORSIM. CORSIM is currently maintained by the Mc Trans Center at the University of Florida. The current data for the fuel use and emissions in CORSIM is based on research conducte d in the mid up tables indexed by vehicle type and instantaneous speed and acceleration. Also, the effects of cold starts are not currently modeled in CORSIM. The CORSIM code needs to be updated to account for new engines and new technologies that are available to measure EU&E. This will help assess the environmental effectivenes s of traffic and ITS implementations more accurately. Research Objectives and Tasks If the Energy Use and Emis sions can be modeled in a widely used traffic simulation model, transportation analysts will have the ability to see the environmental impact of different traffic management strategies. The author of this paper worked with Mc Trans to modify the CORSIM cod e corresponding to the fuel use and emissions outputs. The look up tables for fuel use and emissions were replaced by the VSP approach, with a value for fuel use and emissions rates corresponding to each mode of

PAGE 14

14 VSP. Mc Trans will distribute the revised CO RSIM to its large network of users in the U.S. and worldwide. The main steps for implementing the new VSP model into CORSIM we re to: (1) replace the current set of look up tables with the VSP model and the corresponding mode to emissions and fuel consumpti on values; (2) consider the effect of cold starts, (3) accumulate emissions statistics on a link, OD, and route basis. The tasks that were undertaken to complete this worked are summarized: Task 1 : Fuel and emissions model extensions The capabilities of CORSIM w ere extended to better estimate fuel use and emissions rate s Task 2: Data c ollection and reduction Data was collected by author, with help from the team at North Carolina State University The fi eld studies obtain ed a wide range of speeds, acce lerations, and road grades. This task involved preparation for field data collection, field measurements, quality assurance and control, data analysis, and reporting. Task 3 : VSP model i mplementation and t esting in CORSIM This task include d replacing the current look up tables with the VSP model from the data analysis verifying implementation against hand calculations, and comparing the CORSIM results to empirical results and identifying pa rameters that need calibration. Task 4: Cold start data analysis and implementation into CORSIM This task use d data collected by the team at NCSU for cold starts. The data include d a wide range of vehicles Document Organization In the following report, C hapter 2 contains a summary of literature published by other age ncies that is related to emissions and fuel use measurements. C hapter 3 describes the research approach, including fuel and emissions model extensions in CORSIM, data collection and reduction, the field data VSP model implementation into CORSIM, and cold s tart data analysis and implementation into CORSIM. Chapter 4 contains results and analysis of testing in CORSIM. Finally, C hapter 5 summarizes the study, announces the conclusions reached, and recommends topics for future research.

PAGE 15

15 CHAPTER 2 LITERATURE RE VIEW Review of Current CORSIM Emissions and Fuel Use Modeling In the current version of CORSIM, HC, CO, and NO x emissions only depend on indices for passenger cars (low pe rformance and high performance) and 5 performance indices for trucks. Speed indexes range from 0 to 110 ft/s and acceleration from 10 to 10 ft/s 2 The program looks up the speed and acceleration for every second and assigns an emissions value in mg/s for each pollutant and each second. Road grade has no effect on the emissions values. This implies that a vehicle traveling at a consta nt speed up a hill will emit the same amount of pollutants as a vehicle traveling on a flat surface at the same constant speed. This approach needs to be altered. More effort from the engine is required for traveling the same speed up a hill, which implies more emissions. The lack of an effect of road grade on emissions is one of the reasons the VSP model will be implemented. The VSP model takes road grade into account, and finding the VSP is a good indicator of how much effort the engine is exerting. Fuel consumption rates are modeled similarly to emissions rates in the current CORSIM but the calculation differs for the two different simulators in CORSIM. CORSIM is made up of NETSIM, which represents the traffic on urban streets, and FRESIM, which represen ts the traffic on freeways and urban highways. In NETSIM, t he fuel consumption value depends only on speed, acceleration, and performance index. The program looks up the speed and acceleration for every second and assigns a fuel consumption value in 0.0001 gallons per second for each second. In FRESIM, when there is zero road grade, the program calculates fuel use in the same way. When there

PAGE 16

16 is a grade in FRESIM however, the program makes an adjustment to the acceleration that is looked up. The following e quation comes from the current code in FRESIM used to model fuel use: JACC = JACC + ZACEM(JTYP FEVTBL+JSP) RGRADE 100 (2 1) The program in effect is creating a new dummy acceleration. That is, the program considers the vehicle to have a higher acceleration when traveling on an uphill grade (or lower on a downhill grade) than the vehicle actually does. The JACC that appears after the = sign is the actual vehicle acceleration. The ZACEM refers to a grade correction factor for fuel consump tion. The variables inside the parenthesis are used for referencing the table that contains these grade correction factors. This table has a certain grade correction factor for every speed from 0 to 110 ft/sec and every performance index from 1 to 7. The R GRADE is the road grade, expressed as a decimal. For example, if a vehicle (performance index 1) were traveling on a 3% uphill grade at a constant 70 ft/sec, the actual acceleration would be zero. The ZACEM variable would be equal to 0.305, for speed 70 f t/sec and performance index 1. The RGRADE would be equal to 0.03, and the equation to determine the dummy acceleration would be: JACC = 0 + 0.305 0.03 100 = 0.915. (2 2 ) The program then treats th is dummy acceleration as the real acceleration, goe s into the lookup table for fuel consumption, and assigns a value for fuel consumption for every second.

PAGE 17

17 Overview of Emissions Measur ement Techniques Emissions from vehicles are currently measured by a few different methods. The most common methods for lig ht duty gasoline vehicles include chassis dynamometer tests, remote sensing, and portable emissions measurement systems (PEMS). Chassis dynamometers can produce either average or second by second data on emissions for a specified standardized driving cycle (EPA 1993). In remote sensing, a sensor captures instantaneous ratios of pollutants in the vehicle exhaust as the vehicle passes through a specific location (Bishop et al ., 1998). PEMS can be installed on a vehicle and used to collect micro scale data on any route driven by the vehicle (Frey et al ., 2003). The advantage of PEMS is that it represents actual conditions along any portion of any route, but the measur ement methods of a dynamometer a r e typically more accurate and precise. A PEMS device has been available to the research team for many years and was used for the data collection for this project. Examination of Cold Starts Frey, et al. examined the effects of cold starts on vehicle emissions. Cold start emissions are significantly higher than hot (Frey, et al., 2002) In this study, the engine and catalyst temperatures were not measured. The coolant temperature was measured and evaluated as a substitute for the detecting cold starts. Figure 2 1 shows the relationship between CO and coolant temperature for a particular vehicle start. The emissions start out very high for the first 130 seconds. During that period, the coolant temperature stays lower than 80 F. After about 130 seconds, the coolant temperature and CO emissions stabilize. It can be concluded that

PAGE 18

18 this particular trip has a cold start of about 130 seconds. The coolant temperature was only a surrogate for the engi ne temperature, though, so the authors decided not to classify cold starts based on coolant temperature. They pursued an approach that used the second by second time series of emissions data to classify cold starts. (Frey, et al., 2002) To measure emissio ns, the research team used a PEMS. The authors used statistical techniques based on non linear regression to estimate the duration of the cold start b y determining the time at which the emissions stabilized. The authors called this time t c They wrote a pr ogram in SAS that used non linear regression to estimate t c They used the upper bound of the 95 % confidence interval as the assumed t c in order to reduce the chances that a vehicle would be classified in the hot stabilized operation phase when it was act ually in the cold start phase. (Frey, et al., 2002) For the estimation of the duration of cold starts, HC and CO were given more emphasis than NO emissions because HC and CO seem to be more affected by cold starts than NO does. For cases in which differen t values of t c were found for different pollutants, the highest value was taken to make sure that the cold start data was not recorded as hot stabilized emissions. The authors examined 34 trips that were deemed to experience a cold start, with the duration of the cold starts ranging from 70 to 391 seconds. They wrote a program that determines the driving mode for each second of data and estimates the value of emissions for each mode. One of the modes was cold start mode. All of the other modes were only def ined for the time after the cold start ended. Idle mode refers to zero speed and zero acceleration. Acceleration mode refers to speeds greater than zero accompanied b y acceleration of at least 2 mi/h/s The

PAGE 19

19 acceleration mode also include s acceleration of at least 1 mi/h/s for 3 seconds or more, so as not to exclude more conservative accelerations. Deceleration is defined the same way that acceleration is, only for negative values of acceleration. Anything that does not fall into one of these four modes is considered to be in the cruising mode. The program was also able to calculate total trip emissions. Figure 2 2 shows the average emission rates for each driving mode. The averages were taken for all vehicles and all trips. The cold start mode includes all the activity that took place during the cold start. Some vehicles were driven during the cold start. (Frey, et al., 2002) The cold start emission rate is the highest for HC emissions. For CO emissions, the cold start mode has the highest mean rate, but it is not statistically significantly different from the acceleration rate. The confidence intervals for the two modes also overlap for NO emissions. The amount of time spent in each mode was also examined. After averaging all of the trips, it was found that cold starts account for about five percent of the trip time, but for about 10 % to 15% of NO, CO, and CO 2 emissions and more than 20% of HC emissions. (Frey, et al., 2002) sh ould play a role in making average predictions of the duration of a cold start, perhaps exploration of this hypothesis for future work. The soak times could be used to predic t cold start emissions, with the distribution of the soak times depending on the time of day. (Frey, et al., 2002)

PAGE 20

20 Comparing Emissions for Different Driving Modes Experiment H. Christopher Frey, et al. developed a study design procedure to measure emissio ns using PEMS for vehicles fueled by gasoline and E85 (a blend of 85% ethanol world hot stabilized operation on them to quantify certain aspects of intra vehicle variability in hot stabilized emissions. The study aimed to establish modal emissions rates for idle, acceleration, cruise, and deceleration. The authors also present a statistical method for comparing emissions of different drivers. The authors emphasize the importance of on board data because it accounts for the variability encountered in real life driving cycles. The authors mention that the typical approach for estimating emissions used in models such as MOBIL E5b, MOBILE6, EMFAC7, Mobile Emission Assessment System for Urban and Regional Evaluation, and the Comprehensive Modal Emissions Model is dynamometer testing. The emissions data for these models are calculated from average emissions totals per mile over st andardized driving cycles. Idle emissions are not actually measured in these models; they are estimated based on grams per mile totals from driving cycles with low average speeds. Frey, et al. hypothesize that emissions rates during idling are much lower t han for other modes. They base this hypothesis on the fact that laboratory data from PEMS show that emissions are episodic in nature. Therefore, the average emissions for a trip are greatly influenced by short term events, and the driving cycle based model s cannot capture emissions for the micro scale the way that PEMS measurements can. (Frey, et al, 2003)

PAGE 21

21 This study used the OEM comprised of a five gas analyzer, an engine diagnostic scanner, and an onboard comp uter. The five gas analyzer measures the volume percentage of CO, CO2, HC, NO, and oxygen (O 2 ) in the vehicle exhaust. Simultaneously, the engine scanner is connected to the Onboard Diagnostics (OBD) link of the vehicle, from which engine and vehicle data OEM 2100 collects the following parameters: manifold absolute pressure, vehicle speed, engine speed (r evolutions/min ), intake air temperature, coolant temperature, intake mass airflow, p ercent of wide open throttle, and open/closed loop flag. The main goal of this particular study was to develop an understanding of the n of the study used a small amount of vehicles, drivers, and routes for selected times of the day. There were two primary drivers, who each drove the two primary vehicles on two primary corridors. Most of the data was collected with the two primary drivers driving the same two vehicles on the same two routes to characterize intra vehicle variability and to compare emissions between drivers. There was also a smaller amount of data collected with secondary vehicles and an additional corridor. These secondary vehicles were driven by the two primary drivers, as well as a few other drivers. The purpose of the secondary vehicles and drivers was to assess the strength of the data analysis methodology when applied to different ors are primary arterials with heavy traffic flow during peak travel times. The road grades on these corridors are modest, typically ranging well within + or

PAGE 22

22 Results Figure 2 3 shows an example of the speed, emissions, and fuel consumption plotted against time. For this run, the travel time on the corridor was about 14 minutes, the speed ranged f rom 0 to 45 mi/ h, and t he average speed was about 10 mi/ h. The spikes in the graph show that the peak emissions rates occur over a very small period of time. The largest emissions rates coincide with the acceleration from 0 to 40 mi/ h, as the vehicle clears an intersection. Most of the spikes i n emissions rates coincide with accelerations. (Frey, et al., 2003) The CO, HC, and NO emissions rates are very low for the first 10 minutes, while the vehicle is in stop and go t raffic and does not exceed 20 mi/ h. The rates are much higher during the hig h speed portion of the trip, in which there is also a lot of variation in speed. The HC and CO rates are very similar to each other. The peaks occur at about the same times, especially for the first 10 minutes. (Frey, et al., 2003) The CO 2 and fuel consum emissions of CO and HC are low compared with the CO 2 emissions, it is estimated that more than 99.8% of the carbon in the fuel is emitted as CO 2 Therefore, CO 2 emissions are a good surrogate for fue 2 and fuel consumption occur at similar times to those of HC, CO, and NO emissions, during acceleration and higher speeds. The results shown in Figure 2 3 are similar to those of many other trips in the study. The graphs show that short term events contribute a significant amount to the road emissions should be aimed at understanding and mitigating these short

PAGE 23

23 Frey, et a l. used previous work to define four driving modes. Idle acceleration, deceleration and cruising modes are defined the same way they were in the previous study (Frey, et al., 2001, Frey et al., 2003) A Visual Basic program was created, and it was able to calculate the driving mode for every second of data, determine the average value of emissions for each mode, and calculate the total emissions for a trip. Frey, et al. use one particular trip as an example. This trip was made 141 times, using a 1999 For d Taurus. Figure 2 4 shows an empirical cumulative distribution function of the average CO emissions for acceleration mode for each of the 141 trips. The average CO emission rate for acceleration mode varies from 2 mg/sec to 400 mg/sec across the 141 trips The average for all of the trips is 44 mg/sec. The 95% confidence interval is 33 55 mg/sec. (Frey, et al., 2003) Several statistical methods were used in previous research to explain the variability between runs for emission rates. Significant factors t hat affect the rates were traffic flow, average speed, ambient temperature, and relative humidity. Road grades were not shown to be a statistically significant factor, likely because the grades were not steep or long enough to have a significant effect. (F rey, et al., 2001, Frey, et al., 2003) Figure 2 5 shows the average modal rates for the different pollutants for the 141 trips by the 1999 Ford Taurus. The 95% confidence intervals are also shown. Based on t tatistically significantly different from 2003) The mean rates for all four pollutants decrease in order from acceleration to cruise to deceleration to idle.

PAGE 24

24 Similar results were found when tests were conducted on nine other vehicles. The average emission rate corresponding to acceleration mode was the highest for every vehicle and every pollutant. Cruise mode had the second highest rate for all cases. Deceleration and idle were the third highest and lowest rates, respectively, in almost all cases. The authors conducted t tests to determine if the modal rates were statistically different from each other. Out of 264 possible combinations, 247 of them (94%) were statistic ally significantly different. Therefore, the definitions of the driving modes are useful in portraying the variability of different types of driving. (Frey, et al., 2003) The authors summarized that emissions differ significantly when comparing different vehicles, but similar vehicles produced similar emissions rates, typically. Therefore, in the implementation of emissions rates into micro simulation, it is critical to capture the different rates for different types of vehicles. To understand the impact of each driving mode, the authors summarize the data based on trip times and total emissions. On average, cruising accounted for over 40% of the total time spent driving. The rest of the time was split almost equally between the other three modes. Idling t yp ically only contributes to 5% or less of the total fuel consumption and emissions, whereas acceleration contributes about 35 40%, even though less than 20% of the time is typically spent accelerating. Deceleration contributes less than 10% of the total e missions, even though about 15% of the time is typically spent decelerating. The main conclusion that the authors draw from this summary is that on signalized arterials, cruise and acceleration modes contribute the most to total emissions. (Frey, et al., 2 003)

PAGE 25

25 To compare two drivers on the same route, the authors present the empirical cumulative distribution functions of average emission rates for NO, HC, and CO, shown in Figure 2 6 The data is based on 31 trips made by one driver and 41 trips made by ano ther, both on the same rou te with the same car. Figure 2 6 also shows the 95% confidence intervals for the means. There is overlap in the confidence intervals for all three pollutants. None of the means were statistically significantly different at a 0.05 significance level, based on t tests. The authors found similar results for CO 2 even though it is not shown in Figure 2 6 (Frey, et al., 2003) The limitation with this comparison is that it only compares two drivers. While these drivers were found to be very similar, it is still possible that a comparison of two other drivers would produce statistically different results. Conclusions The different driving modes defined in this study (cruise, acceleration, deceleration, and idle) yielded statistically diff erent emission rates, and tendencies for these rates were similar for the ten vehicles that were tested. The average HC and CO 2 emission rates for acceleration were about 5 times greater than that those for idle. For NO and CO, acceleration emission rates were 10 or more times greater than idle rates. priori modal definitions employed in this work are useful in characterizing at least a portion of the intra vehicle var Another conclusion drawn in this study was that vehicle emissions and fuel use occurred over short periods of time, most ly corresponding with accelerations or high

PAGE 26

26 speeds. Therefore, actions aimed at reducing on road emissions should involve reducing periods of high acceleration. A key finding in this study was the low impact of idling on vehicle emissions. When using area wide driving cycle based models like the MOBILE models, the emissions are based on average emissions per mile over standardized driving cycles using a dynamometer. Idle emissions are not actually measured in this method; they are extrapolated. It is likel y that the estimations for idling using the dynamometer method have been too high, and the contribution of idling to emissions has been overemphasized. (Frey, et al., 2003) Using real world on board data shows that idling actually contributes a very small amount to emissions when compared with the other three modes. There are a few limitations of using on board data. One of these is that using the measurement of evaporat limitation is that NDIR, which was used in the on board instrument, does not measure total HC. The measurements obtained with the NDIR had to be multiplied by a factor of 1.5 2 to get a more acc urate total of HC emissions. Also, vehicles made before 1990 do not have an OBD port, so additional instrumentation would be required to synchronize the on board instrument with the vehicle. (Frey, et al., 2003) This study demonstrates how on board emissi ons measurements can be used to develop useful insights regarding the episodic nature of vehicle emissions, on road emissions hotspots, intra vehicle variability in emissions, and inter vehicle variability in ecommend using on board emissions

PAGE 27

27 measurements to improve emissions estimations for micro scale and macro scale purposes VSP Approach to Classifying Fuel Use and Emissions y gasoline vehicles by routes, time of day, road grade, and vehicle with a focus on the vehicle emissions and fuel use. PEMS have made the measurement of vehicle activity and emissions under real world conditions a possibility. (Frey, et al., 2003) PEMS are starting to replace laboratory based chassis dynamometer measurements because of conducted fo r a transient speed profile and can be replicated, but may not be adequately representative of real 2008) Even though PEMS measurements are made under real world conditions, they still vary from o ne run to another, even with the same vehicle, driver and route. The reason for the variation is the variations in traffic and ambient conditions. One of the sources of v et al., 2008) Experimental Section Experimental design Frey, et al. considered several factors that would affect the emissions and fuel use in the experiment. These factors in for three vehicles, with about 65 hours of data for each vehicle. The vehicles included a

PAGE 28

28 sedan (Chevrolet Cavalier with a 2.2 L engine), a minivan (Dodge Caravan with a 3.3 L engine) and a large SUV (Chevrolet Tahoe with a 5.3 L engine). After the data was collected for the primary vehicles, it was supplemented with more data from seven secondary vehicles, for which only a few hours of data was collected. Routes were selected to include a wide range of road grades and facility classes, and times of day were selected to include a wide range of traffic conditions. The experiment involved three different drivers. Instrumentation For this experi reported on a second by second basis. Compared to dynamometer measurements, this PEMS has good precision, and has been used in a 2008) Field data collection. The PEMS was calibrated every day of data collection. Because this experiment was not considering cold starts, the vehicle was warmed up for 15 minutes prior to measurements being taken. Du ring the experimental drives, a passenger was in charge of the instrumentation and data logging. Data post processing. Data post processing involved analyzing the PEMS data and searching for errors, combining the PEMS and road grade data into a single da tabase, and combining the results from multiple runs into a single database. A standardized vehicle specific power (VSP) based modal average rate of fuel use and emissions was derived to compare the emissions and fuel us e by vehicle, route, driver, and time of day. VSP accounts for power demand, rolling resistance, road grade, and aerodynamic drag, and can be

PAGE 29

29 et al., 2008) The authors provi de a generic VSP equation, using several coefficients, for a typical light duty vehicle: VSP = 0. 278 v [0.305 a + 9.81(sin( a tan( r/ 100))) + 0.132] + 0.0000065 v 3 (2 3 ) where, VSP = vehicle specific power (kw/ton). v = speed (km/h) a = accelera tion (km/h/s) r = road grade (%) Vehicle specific power is separated into 14 different modes. The average emissions and fuel use rates are assigned to each mode, and then an average rate for the entire driving cycle can be derived based on how much time the vehicle spent in each VSP mode. The resulting rate is the standardized rate. Comparisons can be made between the observed data and the standardized data to see how well the standardized VSP model predicts emissions and fuel use. Figure 2 1 shows the 14 VSP modes and the corresponding emission and fuel use rates for one of the vehicles tested. Results Speed, acceleration, and road grade turned out to be the factors that had the biggest effect on intra vehicle variability in emissions and fuel use. The ef fects of temperature and humidity had a negligible effect. Comparing routes. The preferred route can be defined as the route that produces the lowest emissions total for a given origin destination pair. The authors present the total emissions and average emission rates for each route using both the standardized VSP prediction and the empirical results. (Frey, et al., 2008) The empirical

PAGE 30

3 0 and standardized estimates have the same preferred route for 64% of all cases. This suggests that the standardized method for predicting emissions was mostly consistent with the observed results. One particular route yielded the same preferred route over 80% of the time for both the empirical and standardized methods. In a few cases, the preferred route varied depending on t he pollutant, vehicle, driver, and time of day. One way that the standardized and empirical data differs drastically is in their variability. There is more variability in the empirical data because of the changes in traffic and ambient conditions. This le ads to less percent difference in total emissions for percentage difference in total emissions when comparing the highest to the lowest emitting route for a given O/D pair range d from 14 % to 41% for the empirical data, and 13 % Comparing drivers. Driver behavior was shown to have an impact on s when stratified by route, travel direction, vehicle, and time of day, range from 4 % to 5% for CO2, 9 % to 11% for HC, 16 % to 18% for NOx, and 102 % differences are due to different drivers having different desired speeds and accelerations. Comparing times of the day. The emissions and fuel use data were compared for peak and off peak periods, using the same vehicle, route, and driver. The empirical data showed that fuel use and emissions were higher for the off pea k periods than the peak periods. On average the fuel use increased during off peak periods by 7% and 9% for rates and totals, respectively. Emissions rates and totals increased by 32% and

PAGE 31

31 33%. The higher values occurred during the off peak periods partiall y because the speeds were higher. The average speeds during the peak periods were 10% lower than those for the non peak periods. (Frey, et al., 2008) The results observed in this comparison of peak periods to off peak periods will not always hold true. T here is a certain level of congestion for every route that will lead to higher emissions totals for the route. The emissions rates are generally going to be lower for congested periods because of reduced speeds, but if the travel time increases enough, the emissions and fuel use totals will be higher for congested periods than free flow conditions. Comparing road grades. The authors mention one of their previous studies conducted in 2006, in which they studied ve hicles cruising from 35 to 45 mi/ h on level terrain and on positive and negative road grades. (Zhang, et al., 2006) In the 2006 study, the average NO emission rate for positive grades of 5% or more was 4 times higher than it was for level or downhill terrain, and HC, CO, and fuel use were 40 % to 10 0% higher for the uphill segments. It is difficult to compare empirical emissions rates based on road grades for studies more complex than the 2006 one because of variations in road grade over short segments on the road. Therefore, the authors used the sta ndardized approach, with each mode of VSP corresponding to a rate of fuel use or emissions. The VSP modal rates were used to calculate total emissions and fuel use for a particular route. A few different cases were studied using the modal rates from the 2 005 Chevrolet Cavalier. In one case, the authors used the road grade for the specific route on a segment level. This involved short segments, typically 30 570 m. In another case,

PAGE 32

32 the authors only used the average road grade of the entire trip. Both of thes e cases were compared to another case, in which the road grade of the entire route was set to zero When positive road grades were ignored on the segment level, the standardized model un derestimated the fuel use by 16% to 22%, and when negative road grades were ignored, they were overestimated by 22% to 24%. (Frey, et al., 2008) These comparisons are greatly influenced by that fact that the VSP equation is only considering speed and acceleration when road grades are ignored. A vehicle traveling a certain sp eed with a certain acceleration on level terrain is going to have a different VSP value than one with the exact same speed and acceleration on an uphill segment, but the comparison in this study does not take that into account. On the route level, the dif way trip, the average change in fuel use and emissions totals when grades are considered versus when they are not ranges from the authors considering only one segment for the calculation instead of multiple segments. The more segments they considered, the more the variation would have increased between grades being considered an d ignored.

PAGE 33

33 Figure 2 1. CO emission rate and coolant temperature over time for a cold start (Frey, et al., 2002)

PAGE 34

34 Figure 2 2. Average emission rate for each mode over all trips in dataset (Frey, et al., 2002)

PAGE 35

35 Figure 2 3. Time traces of vehicle spe ed, emission rates, and fuel consumption for a 1999 Ford Taurus driven on Chapel Hill Road on August 29, 2000 (Frey, et al., 2003)

PAGE 36

36 Figure 2 4. ECDF of CO emissions rates in acceleration mode based on 141 trips (Frey, et al., 2003) Figure 2 5. Avera ge modal emission rates for 141 trips by a 1999 Ford Taurus (Frey, et al., 2003)

PAGE 37

37 Figure 2 6. ECDF for average emissions rates for two different drivers operating 1999 Ford Taurus on Chapel Hill Road (Frey, et al., 2003) Figure 2 7 Definition of V SP modes and average emissions and fuel use rates for a 2005 Chevrolet C avalier 2.2L a (Frey, et al., 2008)

PAGE 38

38 CHAPTER 3 RESEARCH APPROACH Proposed Approach For this project, the VSP approach was implemented. The modes of VSP were the same ones tha t were defined in the 2008 study This approach is preferred over the driving mode approach because it gives a more specific idea of how much effort the engine is using. Also, the effect of road grades needs to be considered, and the driving mode method does not take road grades into consideration The lack of an effect of road grade in CORSIM on emissions and fuel use is one of the reasons that the capabilities of CORSIM to estimate EU&E must be extended. The VSP model takes road grade into account, and the VSP i s a good indicator of how much effort the engine is exerting. Task 1 Fuel and Emissions Model Extensions The capabilities of CORSIM have been extended to estima te fuel use and emissions rates based on VSP. The author of this paper worked with Mc Trans to get the VSP model in place. The code had to be modified to get the program to ignore the current emissions and fuel use calculations and assign rates based on the VSP mode. Initially, the author and Mc Trans tried implementing the modal rates of the 200 5 Chevrolet Cavalier from the study conducted by Frey, et al. in 2008. These rates can be seen in Figure 2 7. The rates had to be implemented into the code in units specified by the CORSIM data dictionary. According to the data dictionary, emissions rates must be implemented in mg/s, and fuel use must be implemented in gal/s 100,000. ( Mc Trans 2011) It was found that the program does not handle decimals for fuel use and emissions rates inputs. Therefore, the values of emissions rates were all multiplied

PAGE 39

39 the rates (in mg/s) were less than 1 and rounded down to 0 because decimal place s were not allowed. Testing of the VSP im plementation into CORSIM involved comparison with CORSIM NG. CORSIM NG is a program writt en in C# and based on the .NET F ramework by Dr. Scott Washb urn and a team of students at the University of Florida The NG in the name stand NG is a n effort to create a new version of CORSIM based on state of the art software architecture. The car following and lane changing algorithms replicate the functionality from the current version of CORSIM, with the exception that CORSIM NG uses the Modified Pitt ( Cohen, 2002 ) car following model as opposed to the Pitt model. CORSIM NG a lso contains an emissions class that the author helped to write, which calculates emissions with the VSP approach. It uses the same modal rates that were implemented into CORSIM for testing (from the 2008 st udy by Frey, et al.). In CORSIM NG a simple network is in place in order to run tests. The network is a one mile long freeway link. The network was replicated in CORSIM wi th the same entry volume (2400 veh/h ) number of lanes (3), and free flow speed (50 mi/h ). Outputs of the emissions in CORSIM can be seen on the output printout within the program or in the CSV file. Emissions outputs we re displayed in mg/mile for every l ink in the network after the label modification In CORSIM NG, the output is viewed in a CSV file. There is no aggregation of the emissions into mg/mile or mg/vehicle. The emissions o utputs that are shown in CORSIM NG are the rates that are assigned every

PAGE 40

40 tenth of a second for every vehicle. Therefore, in order to find the mg/mile which in this case is the total mg of emissions per vehicle, since the link is 1 mile long, the mg/s values have to be summed for every second that the vehicle is on the link. Th ey must also be divided by 10 because the rates are for every tenth of a second. To verify that the VSP model was implemented correctly, the CORSIM emissions resu lts were compared to the CORSIM NG results. The results fro m both programs, shown in Table 3 1 and Table 3 2, came out to be very similar for emissions, and it was concluded that the VSP model had been implemented correctly Task 2 Data Collection and Reduction The most critical factors that impact EU&E are speed and acceleration. Road grade ca n also have an impact, as seen in the literature review. Therefore, the field study obtain ed a representative range of these factors. Measurements taken by the University of Florida Transportation Research Center (UF TRC) instrumented vehic le (a 2004 Honda Pilot) in Gainesville, FL were used for data collection. Instrumentation Preparation for field data collection include d verification of the status of the PEMS, verification that all the parts and equipment were available, and laboratory calibration of th e PEMS. Taking f ield measurements consist ed of the installation of the instrumentation into the Honda Pilot data collection, and decommissioning. The instruments include d the Axion portable emission measurement system (PEMS) manufactured by MRVGlobal, Pro Scan on board diagnostics (OBD) data logger, and Garmin 76CSx tracking global positioning system (GPS). The Axion is made up of two five gas analyzers. It takes 45 minutes or less to install in a vehicle. The analyzers measure exhaust concentration of CO, CO 2 hydrocarbons (HC), NO, and O 2 CO, CO 2

PAGE 41

41 and HC are measured using non dispersive infrared (NDIR). O 2 and NO are measured using electrochemical cells. The system draws less than 6 amps at 12 volts and does not significantly affect engine load. The Axio n is calibrated using a reference gas containing known concentrations of CO, CO 2 HC (as C 3 H 8 ), and NO. The instrument holds a calibration within 5 percent for three months, but is calibrated more frequently. proximately every 10 minutes. In order to zero, measurements are made on ambient air, which is taken as a refer ence gas. The Axion includes a weather station that records ambient temperature, humidity, and pressure. A ProScan OBD scan tool, software, and dedicated laptop that logs data will be connec ted from the OBD II interface. Vehicle and engine activity data including Engine revolutions per minute (RPM), manifold absolute pressure (MAP), intake air temperature (IAT), mass of air flow (MAF), mass of fue l flow (MFF), vehicle speed (VS), engine coolant temperature (ECT), catalyst temperature, ambient temperature and pressure, and others will b e read and recorded Garmin 76CSx tracking GPS receivers, which record latitude, longitude, and elevation, will be used to record the driving routes and to quantify road grade. P hotos of the instrumentation can be seen in Figures A 1 through A 7 of the appendix Gainesville Data Collection The route driven for data collection in Gainesville is about 39 miles long, with 2 4 miles of freeway and 15 miles of arterial. The author of this paper was the driver for all data collection in the Honda Pilot. The starting point is the intersection of University Avenue and Gale Lemerand Drive. The route continues on westbound Universit y

PAGE 42

42 Avenue, which becomes Newberry Road. At the starting point, the road has 2 lanes in the westbound direction, and a lane is added after about 3 miles at the intersection of Newberry Road and NW 8 th Avenue. The road continues with 3 lanes to the intersecti on with I 75. There are several traffic signals on this segment of the route, and the arterial experiences high congestion during peak periods. There is a shopping mall on the south side of Newberry Road about 0.3 miles east of I 75, which contributes to a lot of the congestion. The University Avenue/Newberry Road segment of the route is about 4.2 miles. The route continues no rthbound on I 75. This is a six lane divided interstate highway. The driver experienced free flow conditions for all four runs on t his segment of the route. The length of the interstate segment is about 11.8 miles. The next step in the route is to exit eastbound onto M L King Boulevard/US 441. This segment continues for another 11.8 miles. It is a four lane divided highway. There ar e five traffic signals on this segment. The route continues southbound on NW 34 th Boulevard, a two lane road. This se gment of the route is about 2.1 miles long. A left turn is made on NW 39 th Avenue, which continues eastbound for about 4 miles. This is a four lane road with a median and several traffic signals. A right turn is made on NE 15 th Street, which continues southbound for about 1.5 miles. This is a two lane undivided road. Another right turn is then made on NE 16 th Avenue, which runs westbound f or about 2 miles to NW 13 th Street. NE 16 th Avenue is another two lane undivided road.

PAGE 43

43 A left turn is made on NW 13 th Street. This segment of the route runs southbound for about 1 mile. It is a 4 lane road with a left turn lane separating the two directi ons for virtually the entire mile. A right turn is made on University Avenue, which brings the route back to the starting point after about 0.5 miles. The route was driven four times for data collection. The starting times and travel times of each trip are shown in Table 3 3 Only a small portion of the route was used for testing with simulation. The portion of the route that was used for simulation testing was a portion along Newberry Road, directly east of I 75 The driving times for this route can be seen in Table 3 4. The Gainesville route and screenshots from the in vehicle camera s can be seen in Figures A 8 to A 21 of the appendix The driving occurred during Spring break for UF students ; thus, there was less traffic congestion than usual. Orlando D ata Collection Data were also collected on I 4 in Orlando on Monday, March 4 th A portion of the freeway was already modeled in CORSIM in a previous study at the University of Florida, and that particular portion was chosen as the data collection site. Thi s portion spanned from S Orange Blossom Trail to Maitland Boulevard. The segment is about 8.4 miles long. There are parts of the segment with 3 lanes and other parts with 4 lanes. The route was driven four times in the eastbound direction and four times in the westbound direction. The eastbound segment was the direction that was already coded in CORSIM. Starting times and travel times for the ea stbound direction are shown in T able 3 2. The first two runs experienced nearly free flow conditions with more co ngestion on runs three and four The biggest source of congestion was vehicles getting on and off at toll road 408, the Spessard L. Holland East West Expressway.

PAGE 44

44 Starting times and travel times for the eastbound direction runs can be seen in T able 3 5 The Orlando route and screenshots from the in vehicle cameras can be seen in Figures A 22 to A 27. Synchronization Because exhaust gas concentrations, engine and GPS data, and recorded videos we re measured fro m different instruments, there wa s a need to sync hronize them to represent simultaneous emissions and vehicle activities. The synchronization was based on trends in time between different sets of data using particular parameters as indicators. The OB D data and the PEMS data were synchronized using engine RPM and CO concentrations as indicat ors. The GPS and the OBD data were synchronized using speed as the indicator. Screening The data was checked for errors and problems. To do this, NCSU has developed a series of Mac ros in LabView. The data w ere processed using these macros to identify a ny problems in the data. If possible, the problems were corrected. If n ot, the data with errors were omitted from the final dataset. The typical data errors or possible problems include invalid data, unusual engine RPM, unu sual intake air temperature, both analyzers zeroing at the same time analyzer freezing, negative emission values, inter analyzer discrepancy, and air leakage. Processing The emissions rates of CO, HC, and NO as well as fuel use rates, were processed by another series of Macros in Labview developed by NCSU. To quantify hot stabilized emis sions and fuel use, the VSP were separated into 14 modes. Average fuel use and emissions rates were estimated for each VSP mode. Uncertainty in the rates

PAGE 45

45 was quantified a t 95% confidence. These rates were implemented into the CORSIM code in the next task. The final modal rates for the 2004 Honda Pilot can be seen in Table 3 6 After the research and analysis for this project had been completed, the team at NCSU sent the U F team data for more vehicles. The data included the VSP modal emissions and fuel use rates for 15 passenger vehicles. The average modal rates for all 15 vehicles can be seen in Table 3 7, and these are the rates that will be used in the CORSIM file that i s distributed to the public. Every few years, the average modal rates will be updated based on data collected for vehicles that represent more current vehicles. Task 3 VSP Model Implementation and Testing in CORSIM This task consist ed of the following s teps: (1) replace the current look up tables with the VSP model and mode to emissions/fuel consumption relationships from the data reduc t ion ; ( 2 ) develop a method to view emissions and fuel estimates from specific routes; ( 3 ) verify implementation against 4 ) compare results from CORSIM to a sample of field data results and identify any parame ters that need to be calibrated. In order to implement the modal rates into CORSIM, the author and took the same steps as in tas k 1. This time, however, the CO emissions values were too large to be multiplied by 1000. The program did not have room for that many digits. Therefore, the CO emissions rates were left in mg/s, and the label for CO emissions was changed back to grams/mile from milligrams/mile. In order to view emissions and fuel use statistics on a link, route, and OD basis, the user can use the Microsoft Excel comma separated values file that CORSIM

PAGE 46

46 automatically outputs. The user can view the emissions outputs for one s pecific link. The emissions output is in mg/mi, so if the user wants total milligrams, than the output in mg/mi must be multiplied by the length of the link. If the user wishes to view the emissions for a specific r oute, the links correspond ing to that rou te can be s ummed. For emissions analysis in this project, summing the link emissions is the approach that the researcher used for finding emissions totals for a certain route. Task 4 Cold Start Data Analysis and Implementation into CORSIM For this task, a method w as established within the source code to add the extra emissions and fuel use caused by cold starts to the CORSIM output Because hot stabilized emissions are becoming lower through better management of the engine and catalytic converter, the re lative importance of the cold start is likely to increase. In a recent study by co PI Frey, it was found that cold starts make up 30% of total trip emissions for NO x 46% for HC, and 12% for CO based on measurement of 7 vehicles These estimates differ gre atly from those made by the MOVES model. The MOVES model may not be providing accu rate results for newer vehicles. (Frey, et al., 2002) There is a need to characterize cold starts based on real world data and to develop a model that better quantifies the e ffects of cold starts. For data collection on cold starts, the vehicles w ere left alone for a 12 hour overnight soak period. The team at NCSU made cold start measurements for 16 passenger cars and 14 passenger trucks For each vehicle used for cold start measurements, one cold start was measured after the 12 hour soak period, followed by hot stabilized driving on the prescribed routes. The EU&E were compared for the cold start period to the hot stabilized period to estimate the effects of a cold start.

PAGE 47

47 To quantify t he effect of cold starts, the fuel use, emissions rates, engine speed (revolutions/min), and catalyst temperatures were plotted against time. The cold start phase was defined as the start time until th e parameters we re stable. T he average fuel us e and emissions excesses from a cold start were documented and can be see n in Table 3 8 The CORSIM code was modified to include the excess emissions and fuel use for cold starts. The percentage of vehicles cold starting was set to 10%. The average excess es for fuel and emissions were calculated for the 30 field vehicles. For 10% of vehicles that start after the initialization period, the vehicles were given an instantaneous fuel or emissions value equal to the average of the excesses of the 30 vehicles. T esting confirmed that 10% of vehicles were undergoing a cold start in the simulation. Table 3 1. CORSIM NG Predicted Emissions (mg) Vehicle NOx HC CO Vehicle 1 16.3 12.7 278 Vehicle 2 16.7 12.7 292 Vehicle 3 16.3 12.1 278 Table 3 2. CORSIM 6.3 P redictions with VSP Implementation (mg/mile) Vehicle Type NOx HC CO 1 16.3 12.4 278 Table 3 3 Gainesville Route Driving T imes Date Starting time Travel time Monday, March 4 th 11:05 AM 61 minutes Tuesday, March 5 th 1:30 PM 60 minutes Tuesday, M arch 5 th 2:45 PM 56 minutes Tuesday, March 5 th 3:45 PM 62 minutes

PAGE 48

48 Table 3 4. Driving Times for Portion of Gainesville Route Used for Testing Run Time to complete portion of route Run 1 34 seconds Run 2 40 seconds Run 3 45 seconds Run 4 48 second s Table 3 5 Orlando Route Driving Times Starting time Travel time 2:35 PM 8 minutes 23 seconds 3:01 PM 8 minutes 9 seconds 3:28 PM 9 minutes 10 seconds 3:52 PM 15 minutes 15 seconds Table 3 6 VSP Modal Fuel Use and Emissions Rates VSP mod e VSP Range Sample Size Fuel NO as NO 2 HC CO CO 2 (kW/ton) (g/s) (mg/s) (mg/s) (mg/s) (g/s) 1 Below 2 4,521 0.50 0.27 0.22 2.92 1.59 2 2 0 1,439 0.67 0.39 0.24 3.22 2.12 3 0 1 5,886 0.50 0.46 0.13 0.70 1.59 4 1 4 2,581 1.11 0.56 0.32 6.01 3.5 1 5 4 7 2,311 1.64 0.80 0.50 10.85 5.16 6 7 10 2,367 2.05 0.99 0.61 19.55 6.45 7 10 13 2,855 2.44 1.12 0.78 22.30 7.71 8 13 16 2,730 2.83 1.47 1.02 36.37 8.89 9 16 19 3,507 3.17 1.84 1.31 75.81 9.92 10 19 23 3,899 3.48 2.57 1.53 200.30 10 .71 11 23 28 1,942 3.96 6.16 1.64 468.74 11.79 12 28 33 1,008 4.37 7.09 1.89 815.02 12.57 13 33 39 560 4.52 8.47 1.99 854.33 12.96 14 Over 39 280 5.50 11.20 2.49 1960.19 14.33

PAGE 49

49 Table 3 7. Average Fuel Use and Emission Rates for 15 Vehicles VSP VSP Range Sample Fuel NO as NO2 HC CO CO2 Mode (kW/ton) Size (g/s) (mg/s) (mg/s) (mg/s) (g/s) 1 Below 2 31,850 0.44 0.11 0.45 1.96 1.49 2 2 0 9,885 0.58 0.09 0.45 2.14 1.89 3 0 1 44,197 0.37 0.03 0.24 0.87 1.18 4 1 4 15,563 0.9 1 0.14 0.68 3.83 2.97 5 4 7 15,102 1.25 0.18 0.89 4.93 4.07 6 7 10 14,706 1.58 0.22 1.08 5.67 5.08 7 10 13 12,836 1.87 0.29 1.28 6.49 6.01 8 13 16 10,847 2.16 0.38 1.46 7.18 6.88 9 16 19 7,932 2.42 0.50 1.61 9.02 7.66 10 19 23 6,829 2.72 0.64 1.86 12.98 8.58 11 23 28 4,704 3.01 0.87 2.04 15.66 9.42 12 28 33 2,481 3.33 1.19 2.29 22.73 10.40 13 33 39 1,514 3.80 2.20 2.61 44.24 11.78 14 Over 39 951 4.51 3.68 3.24 118.06 13.95 Table 3 8 Empirical Cold Start Excess: Average for 30 Vehicles Fuel ( g) CO ( g) HC (g) NOx (g) Sample Size 30 30 30 30 Mean 81.0 9.1 0.70 0.19

PAGE 50

50 CHAPTER 4 RESULTS AND ANALYSIS Testing of I 4 Route in CORSIM Inputs The CORSIM network file for the eastbound segment of the I 4 route in Orlando was already created as part of two separate FDOT projects. This file served as the starting point for data analysis on this project. To obtain traffic volume data that would replicate field conditions the author used t he UF TRC STEWARD traffic data archive. ( Statewide Transportation Engineering Warehouse for Archived Regional Data 2013) In this archive, transportation engineers can look at a specific day and time and get traffic volumes speeds, and occupancies on I 4, as well as many other roads, as recorded by detectors. For the first run, the data from the STEWARD archive di d not make sense. The volume increased significantly from one detector to the next, when there was only an off ramp between the two detectors. Fo r the other three runs, the data from the STEWARD archive seemed logical. The traffic volumes were entered into the CORSIM input file. Simulation travel times were calculated by summing all of the link travel times from the output files. The goal of init ial testing was to match the travel times in the simulation within 10% of the actual travel times. For runs 2 and 3, this was the case. For run 4, which was by far the slowest and most congested run in the field, the travel time did not match up very well. The field travel time for run 4 was 15 minutes 45 seconds. To get the simulation travel time to match more closely with this travel time, the author tried implementing the volumes that were used in this network for the previous project. ( Elefteriadou et al ., 2012) These volumes are generally a good representation of the recurring traffic

PAGE 51

51 congestion on I 4 during the afternoon peak period. The simulation travel time matched up much more closely with the field travel time when these volumes were used Comp aring Emissions Totals for Separate Runs The initial plan was to compare the fuel use and emissions values from the field to those from the simulation. Also, hand calculations that implemented the VSP model from the field data were used for comparison. A n emissions and fuel use comparison of the field data, the simulation data with the VSP model implementation, and the hand calculations for the four eastbound I 4 runs can be seen in Table s 4 1 through 4 3 Simulation travel times can be seen in Table 4 4. In T able s 4 3 and 4 4 the general congestion run refers to the run in which the traffic volumes from the previous project were used. Looking at the empirical data, the emissions values are very unpredictable For example, even though the travel time inc reased from run three to run four the NO emissions were highest for the first run, followed by the third run and fourth run, respectively. After discussing these results with the team at NCSU, it was concluded that summing second by second empirical emiss ions outputs is only us eful if summed for a period of three hours or more. When summing empirical e missions for periods less than three hours, the data tends to be very noisy. The VSP modal model that was obtained from the field data represents an aggregat e of a large portion of data (over 10 hours). It apparently does not work very well to use the average emission rates from each mode from the large portion of data to try to predict emissions outputs over a much smaller portion of time (around 10 minutes). The predicted emissions should still be valid for shorter periods of time, but they are not likely to match closely with the PEMS output. For example, there are several seconds in which the PEMS data says

PAGE 52

52 that there are zero emissions coming out of the ve hicle. This is not accurate; the vehicle is always emitting. This is why the averages of the PEMS output are taken over a long period of time, rather than looking at the PEMS output for specific seconds. Comparing Percent Time Spent in Each VSP Mode After it was concluded that comparing the emissions total s from short runs was not productive the author took a different approach to data analysis. It was decided that instead of trying to compare emissions totals for empirical versus simulated data, the perc entage of time in each VSP mode would be compared for empirical versus simulated data. With this approach, the validity of using the VSP based emissions calculations in CORSIM can be examined. Histograms showing the percent of time spent in each mode can b e seen in Figures 4 1 through 4 3. Analysis for this approach was only done for the uncongested runs (1 3). The hardest part of this analysis was finding a simulated vehicle that started at the entry node and stayed on t he freeway network without exiting b efore the actual field exit. Because the simulation duration was set to the same amount of time as the actual travel time, vehicles that entered more than 30 seconds into the simulation were not considered for this analysis. The reason for this is because the simulation ended while those vehicles were still not close to traveling the entire network. After eliminating these vehicles, there was only one vehicle for the simulated run 2 that started at the entry node and stayed on the freeway the entire time. F or run 3, there were several vehicles that fell into the category, and four of them were analyzed. Looking at the histograms, there are differences between the field vehicle and the simulated vehicle for perce nt time spent in each VSP mode. This can acco unt for some of the error in comparing simulated data to empirical data. The differences were

PAGE 53

53 looked into on a mode by mode basis for the simulated vehicles versus the field vehicles. For each simulated vehicle, the differences in the total seconds in each mode from those of the field vehicles were computed. The average difference for the 14 modes was then computed for each simulated vehicle. For example, in run 2, the simulated vehicle spent 65 seconds in mode 1, 10 seconds in mode 2, and 23 seconds in mod e 3. The field vehicle for run 2 spent 63 seconds in mode 1, 14 seconds in mode 2, and 8 seconds in mode 3. The differences for modes 1, 2, and 3, respectively were 2 seconds, 4 seconds, and 15 seconds, respectively. The differences for modes 4 through 14 were calculated in the same way, and all differences were averaged out. Table 4 5 shows the average differences from the field vehicle for each mode for the simulated vehicle on run 2 and the four simulated vehicles on run 3. Looking at Table 4 5, it woul d appear that the driver of the field vehicle most closely matched up with driver type 10 (most aggressive) in CORSIM and least closely with driver type 1 (most conservative) This would lead one to think that the lower the driver type in CORSIM the highe r the average differenc e would be, but that is not the case, as the vehicle that has driver type 5 has less difference than the vehicle that has driver type 7. The average VSP mode, which can be seen in Table 4 6 is actually similar for most of the runs. For run 2, the average mode of the field data only differs from the simulation data by 0.01. For run 3, the average mode of the field data differs from the average mode of the 4 simulated runs by 0.31. While there are differences in the percentages of time spent in each VSP mode for simulated versus field data, the average modes come out to be quite comparable.

PAGE 54

54 Testing of Newberry Road Route in CORSIM The portion of the Gainesville route that was used for testing was a portion along Newberry Road, directly east of I 75. A CORSIM network had already been set up from a previous study (Washburn and Kondyli, 2006) that included the 3 traffic signals immediately to the east of I 75. Traffic volumes from the previous study were also used because the team at UF di d not have access to detectors, as they did for the I 4 segment. The author did, however, obtain the traffic signal timings from the City of Gainesville Traffic Management office. These actuated timings were input into the CORSIM file, and the author again took the approach of comparing the amount of time spent in each VSP mode for the simulated versus the field data. Simulation travel times can be seen in Table 4 7. In the field, the driver did not have to stop at any of the three signalized intersections that were part of the CORSIM network for any of the four runs. Also, the driver did not stop at the two signalized intersections immediately upstream of these intersections for any of the four runs. This is unusual for this stretch of arterial, but it can be explained by the reduced congestion because of data being collected during spring break for UF students. The four field runs along the stretch modeled in CORSIM were examined for the amount of time in each VSP mode. A histogram showing the percentages for each run can be seen in Figure 4 4. Because the driver did not experience any stops at the signalized intersections in the field, simulated vehicles that did not experience stops were also examined. Four simulated vehicles, all of different driver typ es, were analyzed for VSP mode frequencies. To see how well the simulated vehicles match up with field driving, they

PAGE 55

55 were compared to the average of the field vehicles in terms of percent time in each mode. This comparison can be seen in Figure 4 5. From Figures 4 4 and 4 5, it is apparent that the simulated vehicles are not replicating the field vehicle very well in terms of VSP modes. In four trips, the field vehicle did not experience a VSP mode over nine for any second. All four of the simulated vehicl es experienced VSP modes of 12 or greater. This would suggest that simulated vehicles undergo higher accelerations than the driver of the field vehicle d oes, at least in segments with lower velocities, like on Newberry Road. For the four field runs, the average field travel time was about 42 seconds. T he average amounts of time in each VSP mode were computed for the field runs Then the difference from that average in each mode for each of the four simulated vehicles was computed, similar to the way it wa s computed for the analysis of the I 4 segment. For example, the average amount of time spent in mode 1 for the four field runs was 16 seconds. Simulated vehicle 1 spent 9 seconds in mode 1. The difference is 7 seconds, and the differences of the other mod es were computed in the same way and averaged for simulated vehicle 1. This average came out to be 2.77 seconds. The average differences from the average time in each VSP mode for the field vehicle can be seen in Table 4 8 Table 4 8 would suggest that th e driver of the field vehicle is closest to driver type 8, of the four driver types examined. This is surprising, considering that the field vehicle did not undergo the high VSP modes, as the simulated vehicles did. Because the field vehicle seemed to have lower VSP modes than the simulated vehicles did, one might think that a conservative driver type might match up most closely but this is not

PAGE 56

56 the case. This is a limitation of this mathematical approach on such a short segment. One big difference in any o f the 14 modes can lead to a large average difference. In Table 4 9, the average VSP mode can be seen for all four of the field runs and for each simulated vehicle. This is another suggestion that the simulated vehicles along this stretch of arterial are producing VSP modes that are way too high. The field vehicle averaged a VSP mode of 3.2 over four runs, while all four of the simulated vehicles average over 5. In this short arterial section, the VSP based emissions mod el in CORSIM is not likely to predic t emissions very accurately because it is not predicting the VSP very accurately at all. In order to get the VSP modes to match more closely, one could tinker with the acceleration settings in CORSIM. It is possible that the default settings are producing unrealistically high accelerations for segments with low speeds, like the Newberry Road segment. If the accelerations were lowered, it is possible that the VSP modes would resemble field driving more closely.

PAGE 57

57 Table 4 1. Empirical D ata (I 4 segment ) Tra vel Time NO (mg) HC (mg) CO (mg) fuel (g) run 1 8 min 23 sec 1505 446 2672 894 run 2 8 min 39 sec 246 494 3088 987 run 3 9 min 40 sec 588 604 2829 1059 run 4 15 min 45 sec 492 920 3442 1244 on ramp 2 0 35 173 69 on ramp 3 34 29 47 74 on ramp 4 16 47 13 55 off ramp 1 16 54 128 94 off ramp 2 17 42 136 68 off ramp 3 64 39 7 93 off ramp 4 29 40 221 108 Table 4 2. VSP Model Hand Calculations (I 4 segment ) NO (mg) HC (mg) CO (mg) fuel (g) run 1 788 401 43221 1095 run 2 871 411 53393 1 112 run 3 973 444 62038 1187 run 4 983 493 48909 1394 on ramp 2 43 23 1943 64 on ramp 3 54 28 3193 74 on ramp 4 36 21 1606 56 off ramp 1 67 39 2543 108 off ramp 2 53 34 1600 94 off ramp 3 71 43 2860 117 off ramp 4 71 41 2444 117

PAGE 58

58 Table 4 3. CORSIM Predictions with VSP Implementation (I 4 Segment) NO HC CO fuel mg/mile mg mg/mile mg g/mile mg mi/gal grams Run 1 Not modeled Run 2 100.0 840.1 49.8 418.7 5.7 47468.4 20.6 1145.7 Run 3 99.3 833.8 49.5 416.2 5.6 46888.8 20.8 1137.9 Run 4 99.1 832.8 49.5 416.0 5.6 47157.6 20.8 1138.5 General congestion 117.5 987.3 60.9 511.8 6.0 50148.0 16.7 1412.0 Table 4 4. Simulation Travel Times (I 4 Segment) Run Travel time Run 2 9 minutes, 13 seconds Run 3 9 minutes, 8 seconds Run 4 9 mi nutes, 3 seconds General congestion 13 minutes, 54 seconds Table 4 5. Average Difference from Field Vehicle for All Modes (I 4 segment) Simulated Vehicle Driver Type Average Difference (s) Run 2 10 9.4 Run 3 1 7 16.5 Run 3 2 8 13.6 Run 3 3 1 33 Run 3 4 5 14.9

PAGE 59

59 Table 4 6. Average VSP Mode (I 4 Segment) Vehicle Driver type Average VSP Mode Field run 1 N/A 3.25 Field run 2 N/A 6.52 Field run 3 N/A 5.91 Simulated run 2 10 6.51 Simulated run 3 7 6.53 Simulated run 3 8 6.22 Simulated run 3 1 5.97 Simulated run 3 5 6.14 Average of all simulated run 3 N/A 6.22 Table 4 7. Simulation Travel Times (Newberry) Vehicle Driver Type Travel Time (s) 1 5 37 2 8 35 3 2 38 4 6 34 Table 4 8 Average Difference From Field Vehicle Average fo r All Modes (Newberry) Simulated Vehicle Driver Type Average Difference (s) 1 5 2.7 7 2 8 2.3 4 3 2 2.98 4 6 2.80

PAGE 60

60 Table 4 9. Average VSP Mode (Newberry) Vehicle Driver Type Average VSP Mode Field N/A 3.2 Simulated Vehicle 1 5 6.32 Simulated Vehic le 2 8 5.6 Simulated Vehicle 3 2 5.26 Simulated Vehicle 4 6 6.59 Figure 4 1. Run 1 VSP mode frequencies (I 4 segment )

PAGE 61

61 Figure 4 2. Run 2 VSP mode frequencies (I 4 segment )

PAGE 62

62 Figure 4 3. Run 3 VSP mode frequencies (I 4 segment )

PAGE 63

63 Figure 4 4. VSP mode frequencies, all field runs (Newberry Road segment)

PAGE 64

64 Figure 4 5. VSP mode frequencies, simulated vehicle s and average of field vehicles (Newberry Road segment)

PAGE 65

65 CHAPTER 5 SUMMARY AND RECOMMENDATIONS In this project, the v alidity of implementing the VSP based emissions estimating approach was investigated. The author, with the help of McTrans was able to i mplement the model ing approach successfully into CORSIM, using emissions and fuel use values derived from previous and concurrent research by the NCSU team After the successful implementation of the VSP modeling approach, field data collection was conduct ed. From the collected data, the research team was able to determine emissions and fuel use rates for each VSP mode for the field vehicle, a 2004 Honda Pilot. These data were applied to the VSP model in CORSIM, and comparisons of simulation data to field d ata commenced. The original plan was to compare emissions and fuel use outputs of the field vehicle from the portable emissions measuring device (PE MS) to those of the simulation. Testing was done for the I 4 route, in which each trip lasted between 8 and 16 minutes. It was concluded, though, that small scale totals of second by second emissions data from the PEMS should not be used for comparison. The PEMS data can be somewhat noisy, especially over short time intervals. It was ultimately determined that comparisons should only be made with PEMS data aggregated over much longer periods of time, on the order of at least 3 hours While the PEMS data are fairly unpredictable over short periods of time, that does not mean that the average emissions and fuel u se rates derived from the PEMS are invalid. These rates are derived from a large sample size of data Therefore, although comparing PEMS totals over the short term to simulation totals using the VSP model is unproductive, the VSP model is still reasonably representative of actual

PAGE 66

66 emissions for individual simulations over short time periods. The inaccuracy l ie s in the PEMS data for certain seconds, not in the simulation estimated data. For example, there are sections in the PEMS data that have zero emissions for certain pollutants while the vehicle is moving. The PEMS data are not going to be accurate for every second, but when the data are summed over large periods of time (over 10 hours for this project), the average emissions and fuel use rates should be v alid. A recommendation for future research would be to build a much larger CORSIM network, in which it takes a simulated vehicle 3 hours or more to drive a certain route. More meaningful c omparisons could then be made between the PEMS data from the actual driving of this route and the simulation data. A limitation of this approach is simply the amount of effort required to create a network that large. To build a network that could take three or more hours to drive but would require less effort to build, a circuitous route could be set up in CORSIM. For e xample, the westbound section of the I 4 route could have been coded into CORSIM, as well as the exit and entry ramps and roads that connect the ends of the westbound segment to the ends of the eastbound seg ment. The difficulty here would be finding a simulated vehicle that continues to drive east and west on I 4 for 3 or more hours. After it was determined that comparing PEMS data to simulation data over short time periods was not productive, the focus swit ched to investigating the percent of time spent in each VSP mode f or simulated and field vehicles. I t was found that there were differences in the amount of time spent in each mode between the field vehicle and the simulated vehicles The differences are l ikely due to the car following algorithms built into CORSIM. The differences also depended on the driver type of the simulated

PAGE 67

67 vehicle. A useful topic for a future study would be an investigation of how well the car following model in CORSIM replicates re al world driving behavior. The study could use the VSP mode approach, similar to the testing done in this study. The analysis could start with just one driver type. Researchers could calibrate the input values of a particular driver type and then analyze v ehicles from that driver type. Calibration could be based around trying to match VSP mode frequencies. If one was able to calibrate a driver type so that the VSP mode frequencies matched up more closely with a field driver, than the emissions of that field driver could be more closely predicted in CORSIM as well. Cold starts were another issue that was examined in this project. Cold starts contribute a significant amount of additional vehicle emissions, and before this project CORSIM had no way of accoun ting for cold starts. Cold start data were collected by the NCSU team, and the average cold start excess emissions w ere computed for 30 vehicles for each pollutant. CORSIM now has the ability to add this average cold start excess to each pollutant once the cold start vehicle is generated in the network. The amount of vehicles set to cold start is 10%. The emissions estimation process in CORSIM has undergone a complete makeover through this project. The VSP based emissions model not only helps to account for grade in emissions estimating, but it also provides more up to date data for CORSIM and its users to use. Adding the cold start logic to the CORSIM code prevents the simulator from significantly underestimating emissions and fuel use in networks that expe rience a lot of cold starts. Overall, the emissions and fuel use outputs in CORSIM

PAGE 68

68 should be much more accurate after the VSP and cold start implementation into the code.

PAGE 69

69 APPENDIX A FIGURE S FROM DATA COLLECTION Figure A 1. Gas zeroing tubes and weathe r station (Photo courtesy of Jack Hulsberg) Figure A 2. PEMS device connection running from side window to tailpipe (Photo courtesy of Jack Hulsberg)

PAGE 70

70 Figure A 3. Side view of instrumented Honda Pilot (Photo courtesy of Jack Hulsberg) Figure A 4. PEMS connection to tailpipe (Photo courtesy of Jack Hulsberg)

PAGE 71

71 Figure A 5. OBD connection (Photo courtesy of Jack Hulsberg) Figure A 6. GPS devices (Photo courtesy of Jack Hulsberg)

PAGE 72

72 Figure A 7. PEMS computer and laptop connected to OBD (Ph oto courtesy of Jack Hulsberg) (A)

PAGE 73

73 (B) Figure A 8. Gainesville route s (Martin, 2013) A) Entire route driven in Gaineville, B) Route in Gainesville used for testing in CORSIM. (A) (B) Figure A 9. University Avenue near starting point A) Front view, B) Rear view. (Photos courtesy of Jack Hulsberg)

PAGE 74

74 (A) (B) Figure A 10. Newberry Road nea r Oaks Mall A) Front view, B) Rear view. (Photos courtesy of Jack Hulsberg) (A) (B) Figure A 11. Ramp to get on I 75 N A) Front view, B) Rear view. (Photos courtesy of Jack Hulsberg)

PAGE 75

75 (A) (B) Figure A 12. US 441 A) Front view, B) Rear view. (Photos courtesy of Jack Hulsberg) (A) (B) Figure A 13. Ri ght turn onto NW 34 th Blvd A) Front view, B) Rear view. (Photos courtesy of Jack Hulsberg)

PAGE 76

76 (A) (B) Figure A 14. NW 34 th Blvd A) Front view, B) Rear view. (Photos courtesy of Jack Hul sberg) (A) (B) Figure A 15. NW 39 th Ave A) Front view, B) Rear view. (Photos courtesy of Jack Hulsberg)

PAGE 77

77 (A) (B) Fig ure A 16. Right turn onto NE 15 th St A) Front view, B) Rear view. (Photos courtesy of Jack Hulsberg) (A) (B) Figure A 17. NE 15 th St A) Front view, B) Rear view. (Photos courtesy of Ja ck Hulsberg)

PAGE 78

78 (A) (B) Figure A 18. NE 16 th Ave A) Front view, B) Rear view. (Photos courtesy of Jack Hulsberg) (A) (B) Figure A 19. Left onto NW 13 th St A) Front view, B) Rear view. (Photos courtesy of Jack Hulsberg)

PAGE 79

79 (A) (B) Figure A 20. NW 13 th S t. A) Front view, B) Rear view. (Photos courtesy of J ack Hulsberg) (A) (B) Figure A 21. Right onto University Ave A) Front view, B) Rear view. (Photos courtesy of Jack Hulsberg)

PAGE 80

80 Figure A 22. Orlando route (Martin, 2013) (A) (B) Figure A 23. On ramp for I 4 E at S Orange Blossom Trail A) Front view, B) Rear view. (Photos courtesy of Jack Hulsberg)

PAGE 81

81 (A) (B) Figure A 24. Merging onto I 4 E from S Orange Blossom Trail A) Front view, B) Rear view. (Photos courtesy of Jack Hulsberg) (A) (B) Figure A 25. I 4 E A) Front view, B) Rear vie w. (Photos courtesy of Jack Hulsberg)

PAGE 82

82 (A) (B) Figure A 26. Ramp to get on I 4 W from Maitland Blvd A) Front view, B) Rear view. (Photos courtesy of Jack Hulsberg) (A) (B) Figure A 27. I 4 W A) Front view, B) Rear view. (Photos courtesy of Jack Hulsberg)

PAGE 83

83 LIST OF REFERENCES Bishop, G.A. Starkey, J.R., Ihlenfeldt, A., and W.J. Williams (1998), IR Long path Photometry: a Remote Sensing Tool for Automobile Emissions Cohen, S. L. (2002), Application of Car Following Systems in Microscopic Time Scan Simulation Models Transportation Research Record : Journal of the Transportation Research Board TRR 1802, Washing ton, D.C Cohen, S. L. (2002) Application of Car Following Systems to Queue Discharge Problem at Signalized Intersections Transportation Research Record: Journal of the Transportation Research Board TRR 1802, Washington, D.C Elefteriadou, L. Washburn S.S ., Yin, Y. Modi, V., and C. Letter (2012), Variable Speed Limit (VSL) Best Management Practice Florida Department of Transporta tion: Tallahassee, Florida Ene Annual Energy Review 0384, Washingto n, DC Frey, H.C., Rouphail, N.M., Unal, A, and J.D. Colyar (2001 ), Emission Reductions through Better Traffic Management: An Empirical Evaluation Based on On Road Measurement FHWY/NC/2002 001, Prepared by North Carolina State University for North Carol ina Department of Trans portation: Raleigh, North Carolina Frey, H.C. Unal, A., and J. Chen (2002), Recommended Strategy for On Board Emission Data Analysis and Collection for the New Generation Mode Prepared by North Carolina State University for Offi ce of Transportation and Air Quality, U.S. Environmental Protection Agency: Raleigh, N orth C arolina Frey, H. C., Unal, A., Chen, J., Li, S., and C. Xuan ( 2002), Methodology for Developing Scale Motor Vehicle and Equipm ent Emission Estimation System EPA420 R 02 027 Prepared by North Carolina State University for the Office of Transportation and Air Quali ty, U.S. Environmental Protection Agency: Raleigh, North Carolina Frey, H.C., Unal, A., Rou phail, N.M., and J.D. Co lyar (2003), On Road Measurement of Vehicle Tailpipe Emissions Using a Portable Instrument Journal of the Air and Waste Management Association 53(8) North Carolina State University: Raleigh, North Carolina. Frey, H.C., Zhang, K., and N.M. Rouphail (2 0 08), Fuel Use and Emissions Comparisons for Alternative Routes, Time of Day, Road Grade, and Vehicles Based on In Use Measurements Environmental Science and Technology 42(7) North Carolina State University: Raleigh, North Carolina. Martin, Andrew (201 3), Route Builder : http://routebuilder.org/

PAGE 84

84 McTrans (2011), CORSIM Data Dictionary Statewide Transportation Engineering Warehouse for Archived Regional Data ( 2013 ), http://cce trc cdwserv.ce.ufl.edu/steward / U.S. Department of Transportation (2012), Transportation for a New G eneratio n, Strategic Plan for Fiscal Years 2012 2016 http://www.dot.gov/sites/dot.dev/files/docs/990 _355_DOT_StrategicPlan_508lowr es.pdf U. S. Environmental Protection Agency (2011) 1970 2011 Average Annual Emissions, All Criteria Pollutants in MS Excel, http://www.epa.gov/ttn/chief/trends/index .html. U.S. Environmental Protection Agency Federal Test Procedure Review Project: Preliminary Technical Report A/420/R 93 007: Washington, DC. Washburn, S. S and A. Kondyli (2006), Development of Guidelines for Driveway Location and Median Configuration in the Vicinity of Interchanges Florida Department of Transportation: Tall ahassee Florida Zhang, K., and H. C. Frey (2006), Road Grade Estimation for On road Vehicle Emission M odeling U sing LIDAR D ata Journal of Air and Waste Management Association 56 (6), North Carolina State Universit y: Raleigh, North Carolina

PAGE 85

85 BIOGRAPHICAL SKETCH Jack Hulsberg was born in Jacksonville, Florida He grew up there and graduated from Stanton College Preparatory School in 2006. He earned his B.S. in civil engineering from the University of Florida in 2011. A year later, he enrolled in the m r understanding of transportation engineering. He worked at ARCADIS in Jacksonville at the time of his thesis submittal and hopes to find a job near West Palm Beach, Florida, to be with his fiance, Sarah Jeck, after graduation.