transportation research recordonlinepubs.trb.org/onlinepubs/trr/1993/1410/1410.pdf ·...

TRANSPORTATION RESEARCH

RECORD No. 1410

Pavement Design, Management, and Performance

·Pavetnent Monitoring and Evaluation

A peer-reviewed publication of the Transportation Research Board

TRANSPORTATION RESEARCH BOARD NATIONAL RESEARCH COUNCIL

NATIONAL ACADEMY PRESS WASHINGTON, D.C. 1993

Transportation Research Record 1410 ISSN 0361-1981 ISBN 0-309-05557-1 Price: $28.00

Subscriber Category IIB pavement design, management, and performance

TRB Publications Staff Director of Reports and Editorial Services: Nancy A. Ackerman Associate Editor/Supervisor: Luanne Crayton Associate Editors: Naomi Kassabian, Alison G. Tobias Assistant Editors: Susan E. G. Brown, Norman Solomon Production Coordinator: Sharada Gilkey Graphics Coordinator: Terri Wayne Office Manager: Phyllis D. Barber Senior Production Assistant: Betty L. Hawkins

Printed in the United States of America National Research Council. Transportation Research Board.

Sponsorship of Transportation Research Record 1410

GROUP 2-DESIGN AND CONSTRUCTION OF TRANSPORTATION FACILITIES Chairman: Charles T. Edson, Greenman Pederson

Pavement Management Section Chairman: Joe P. Mahoney, University of Washington

Committee on Pavement Monitoring, Evaluation, and Data Storage ·

Chairman: Freddy L. Roberts, Louisiana Tech University Secretary: Richard B. Rogers, Texas Department of Transportation A. T. Bergan, Billy G. Connor, Brian E. Cox, Jerome F. Daleiden, Wade L. Gramling, Jerry J. Hajek, Scott A. Kutz, Kenneth J. Law, W. N. Lofroos, Et;lwin C. Novak, Jr., Dennis G. Richardson, Ivan F. Scazziga, Mohamed Y. Shahin, Roger E. Smith, Elson B. Spangler, John P. Zaniewski

Committee on Surface Properties-Vehicle Interaction Chairman: John Jewett Henry, Pennsylvania State University Secretary: James C. Wambold, Pennsylvania State University Robert A. Copp, Steven L. Cumbaa, Gaylord Cumberledge, Kathleen T. Diringer, Stephen W. Forster, Lawrence E. Hart, Carlton M. Hayden, Brian S. Heaton, Walter B. Horne, David L. Huft, Michael S. Janoff, Khaled Ksaibati, Kenneth J. Law, Georg Magnusson, Kenneth H. McGhee, James E. McQuirt, Jr., William

. G. Miley, Robert L. Novak, William D. 0. Paterson, Jean Reichert, Dennis G. Richardson, Richard B. Rogers, Roger S. Walker, Thomas J. Yager

Committee on Vehicle Counting, Classification, and Weigh-In-Motion Systems

Chairman: Perry M. Kent, Federal Highway Administration David Preston Albright, Shyamal Basu, Thomas F. Black, Harold R. Bosch, James Y. Campbell, Jr., Craig A. Copelan, Wiley D. Cunagin, Michael J. Dalgleish, Edward K. Green, H. K. (Kris) Gupta, ·Barbara Mason Haines, John L. Hamrick, Lawrence E. Hart, Loyd R. Henion, David L. Huft, Bernard Jacob, Clyde E. Lee, D. Keith Maki, William A. Mickler, John T. Neukam, Thomas Papagiannakis, Peter Sebaaly, Herbert F. Southgate, Alex T. Visser

Daniel W. Dearasaugh, Jr., Transportation Research Board staff

Sponsorship is indicated by a footnote at the end of each paper. The organizational units, officers, and members are as of December 31, 1992.

I I

Transportation Research Record 1410

Contents

Foreword

Accreditation of Strategic Highway Research Program Long-Term Pavement Performance Pavement Distress Raters Gonzalo R. Rada, John S. Miller, William Y. Bellinger, and Richard B. Rogers

Comparison of Pavement Surface Distress Measurement Systems Raymond K. Moore, G. Norman Clark, and Andrew]. Gisi

Obtaining Rut Depth Information for Strategic Highway Research Program Long-Term Pavement Performance Sites · Wade L. Gramling, George S. Suzuki, John E. Hunt, and Kazuhiko Muraoka

Adapting an Automated Data Collection Device for Use at an Airport Margaret Broten, George Schwandt, and Rudy Blanco

Maintenance Skid Correction Program in Utah Tracy C. Conti, fames C. McMinimee, and Prianka Seneviratne

Consistency of Roughness and Rut Depth Measurement Collected with 11 South Dakota Road Profilers Sanjay Asnani, Khaled Ksaibati, and Turki I. Al-Suleiman

APP ARE: Personal Computer Software for Automated Pavement Profile Analysis and Roughness Evaluation ]. Jim Zhu and Rajeev Nayar

Factors Affecting Repeatability of Pavement Longitudinal Profile Measurements Khaled Ksaibati, Sanjay Asnani, and Thomas M. Adkins

v

1

11

19

25

32

41

52

59

Automated Versus Manual Profilograph Correlation Carl B. Bertrand

Video Cameras for Roadway Surveillance: Technology Review, Test Methods, and Results Carl Arthur MacCarley, Daniel Need, and Robert L. Nieman

Traffic Sensing System for Houston High-Occupancy Vehicle Lanes Clyde E. Lee and Liren Huang

Cluster Analysis of Arizona Automatic Traffic Recorder Data I oe Flaherty

Vehicle Configuration Influences on Weigh-in-Motion Response P. E. van Niekerk and A. T. Visser

Influence of Vehicle Speed on Dynamic Loads and Pavement Response Peter E. Sebaaly and Nader Tabatabaee

Results of Weigh-in-Motion Project in France: 1989-1992 B. Jacob, C. Maeder, L. A. George, and M. Gaillac

67

80

89

93

100

107

115

Foreword

This Record compiles current information on pavement monitoring and evaluation; most of the papers were presented in sessions at the 1993 TRB Annual Meeting.

Rada et al. report on the development and implementation of an accreditation process by the Strategic Highway Research Program (SHRP) to ensure the quality of distress data being collected on Long-Term Pavement Performance (LTPP) sections by manual surveys. Moore et al. compare two pavement distress data collection devices, and Gramling et al. discuss the automated RoadRecon survey systems to obtain records of pavement surface distress and transverse profile of SHRP LTPP sites. Broten et al. take the reader to Chicago's O'Hare International Airport, where a demonstration project was conducted to determine the feasibility of using automated data collection equipment for distress data collection.

Conti et al. suggest an efficient process for priority ranking skid-deficient projects for restorative treatment. Asnani et al. analyze results from 11 South Dakota road profilers for consistency of roughness and rut depth measurements. Zhu and Nayar describe a new personal computer software package, APP ARE, that automatically determines the profile index from digitized profilograms. Ksaibati et al. look closely at the accuracy of profilometers, particularly at errors caused by human operators and environmental factors. Bertrand reports on an evaluation by the Texas Department of Transportation to correlate the outputs of the automated Cox profilograph, the automated McCracken profilograph, and a manual McCracken profilograph.

MacCarley et al. present a technology review and describe test methods and results of the application of video surveillance to roadway traffic monitoring, and Lee and Huang describe a data acquisition system that uses a pair of infrared light beam sensors and a microprocessor for traffic sensing on Houston's high-occupancy vehicle lanes. Flaherty uses cluster analysis statistical procedures to analyze data from Arizona automatic traffic recorders. Van Niekerk and Visser simulate dynamic pavement loadings of different vehicle configurations on a range of pavement roughnesses where weigh-in-motion devices could be used .. Sebaaly and Tabatabaee document the findings of a full-scale field research project aimed at evaluating the effect of vehicle speed on measured dynamic loads and pavement response. Finally, Jacob et al. present the objectives, organization, and main results of a national research and development project in France on weigh-in-motion techniques and devices.

TRANSPORTATION RESEARCH RECORD 1410

Accreditation of Strategic Highway Research Program Long-Term Pavement .Performance Pavement Distress Raters

GONZALO R .. RADA, ]O.HN S. MILLER, WILLIAM Y. BELLINGER, AND

RICHARD B. ROGERS

Distress surveys are one element of the monitoring effort currently under way by the Strategic Highway Research Program (SHRP) for the Long-Term Pavement Performance (LTPP) study. Because accurate data are key to the success of the L TPP study, SHRP has developed and implemented an accreditation process to ensure the quality of distress data collected from manual surveys. The purpose of SHRP's accreditation process is to provide a means for ensuring, to the extent possible, the quality and consistency of distress data being collected by the raters. The process consists of two parts, a written examination and a twopart field survey examination, and is being administered in a workshop situation. Although the process is still in its early stages, it is SHRP's intent that· all distress data for the LTPP study be collected by raters who have successfully completed the accreditation. The SHRP accreditation process and the results of its implementation to date are discussed. ·

Efforts of the Strategic Highway Research Program (SHRP) to monitor surface distress on the test sections under study in the Long-Term Pavement Performance (L TPP) research serve two primary purposes. The first is to provide a permanent, objective, high-resolution record of pavement condition over the full length and width of the sections under study; the second is to provide detailed, distress-specific condition data for use in the development of pavement performance prediction models.

To achieve these objectives, SHRP is making use of the PASCO Roadrecon photographic distress survey technology, which provides for high-resolution 35-mm black and white photographs and photographic transverse-profile measurements (1). The reduction of distress data from the PASCO film is accomplished through a computer-assisted interpretation process (2). The film interpretations and the initial quality assurance (QA) of the interpretations are performed under ·close supervision of experienced engineers and technicians in an office environment. Further QA of the film interpretations is performed at the SHRP regional coordination offices (RCOs) by the personnel most knowledgeable of the actual conditions at the sites.

In those instances in which the PASCO units cannot be used because of time constraints or the difficulty of getting

G. R. Rada and J. S. Miller, PCS/Law Engineering, 12240 Indian Creek Court, Suite 120, Beltsville, Md. 20705. W.Y. Bellinger, FHWALTPP Division, 6300 Georgetown Pike, McLean, Va. 22101. R. B. Rogers, Texas Department of Transportation, D-10, 40th and Jackson, Austin, Tex. 78731. ·

the PASCO survey vehicles to the site, a manual distres~ survey serves as the backup data collection method (3). These surveys do not have the same level of thorough supervision and QA checking as are available in the film interpretation process. Another important facet of the manual survey is that no permanent objective records, such as photographs· obtained in a consistent and controlled manner, are left behind to supplement the hand-drawn maps, observations, and interpretations (possibly subjective) of the rater.

Consequently, an accreditation process to develop consistency among raters has been established by SHRP. The specific purpose of this accreditation process is to provide a means for ensuring, to the extent possible, the quality and consistency of distress data that are collected for the L TPP program by the RCO raters. Although the process is still in its early implementation. phase, it is SHRP's intent that all distress data for the L TPP study be collected by raters who have successfully completed the accreditation.

This paper describes the L TPP accreditation process and its implementation to date. The first part of the paper presents an overview of the accreditation procedure, including its basis, components, and grading system. The next portion of the paper focuses on two workshops conducted by SHRP in May and June 1992 as part of the implem~ntation process. Particular emphasis is placed on th~ changes to the accreditation process that resulted from these workshops. Finally, the major conclusions to date and ·recommendations for improving the overall process and its implementation are presented in the last portion of th_is paper.

ACCREDITATION PROCEDURE

Achieving the desired consistency in distress data collection requires a basis for the actual identification, measurement, and recording of distresses. Pavement distresses are defined and measurement and recording requirements are established in SHRP's Distress Identification Manual for the Long-Term Pavement Performance Studies (DIM) (4). Manual distress surveys are performed using the procedures published in an appendix of the DIM. This appendix contains instructions for performing manual surveys, standard map symbols for recording distress occurrences, map sheets, and distress data summary sheets. The maps are prepared in the field by the rater and ·an distress quantities· are then summarized and re-

2

corded on distress survey summary sheets appropriate for the pavement type.

The importance of the distress data to the goals of the L TPP program require minimum levels of experience and expertise for the personnel performing the surveys. To participate in the accreditation process, and hence future distress data collection activities, RCO raters must have the following: high school education (or equivalent), _previous training in distress surveys (either formal.or informal); and familiarity with the LTPP DIM and field data collection procedures. Previous field experience (minimum of 1 year) is highly desirable but not mandatory.

The actual accreditation process consists of two major parts (5): a written examination and a two-part field survey examination. The written examination is intended to test the general knowledge of the rater. The examination consists of the following:.

• Identification of distresses from slides: 60 slides are shown to the RCO raters, covering various distress types on asphaltsurfaced, jointed concrete and continuously reinforced concrete pavements. The raters are allowed 20 sec to identify the distress type(s) shown in each slide. This portion of the examination is 20 min in length and is worth 25 percent of the total written exam.

•Knowledge of distress types, severities, and measurement procedures: RCO raters are required to answer a total of 10 short-answer questions covering the description of distress types, severity-level definitions, or field measurement procedures, or all of these. This part of the examination is 45 min in length and is worth 60 percent of the total written exam.

• Interpretation of distress maps: RCO raters are required to summarize distress types and quantities from a map sheet. This portion of the examination is 20 min in length and is worth 15 percent of the total written exam.

If a rater fails to achieve the minimum grade, review sessions are held and a reexamination using different questions is conducted. If after the second attempt the rater cannot pass the written examination, the rater is not accredited.

The field survey examinations are intended to measure the capabilities of the raters in observing and recording distress data. They are conducted on two 150-rri pavement sections: one surfaced with asphalt concrete and one with portland cement concrete. These sections will have been surveyed in detail by a committee of experienced raters, including the accreditation workshop leaders and other knowledgeable personnel, to determine the extent and types of distresses present. The results of the committee surveys are considered the ground truth or the "actual values" against which the individual rater's results will be compared for grade.

Each RCO rater is required to independently perform a distress survey of the sections included in the accreditation process. These surveys are performed using LTPP procedures; that is, detailed scaled mapping of the section followed by reduction of the mapped quantities and completion of the appropriate distress summary forms. At each accreditation section, the RCO raters are allowed 3 hr to complete the survey and to reduce the distress data for each section.


Grading the distress surveys is accomplished by comparing the individual rater's results to the actual values determined by the committee of experienced raters. The point value system used for grading consists of a maximum of 10 points for each distress type actually in the section. These 10 points are distributed among the individual severity levels, where applicable, as well as the total quantity of the distress type. Accuracy in identifying and recording distress determines the number of points received for each severity level and for the total. In turn, accuracy is determined by comparing the variance of the rater's results to those from the committee of experienced raters (i.e., ground truth quantities). The point values received on the basis of variance between committee and rater observations are calculated as follows:

( actual - rater) x Qt W t

Points = 1 - actual y g

where

actual = quantity of distress from committee survey, rater = quantity observed and recorded by rater, and

QtyWgt = quantity weight factor applied to total and to each severity level (QtyWgt = 7 for total quantity of distress and 0.5, 1.0, and 1.5 for the total quantity of low, moderate, and high-severity distress, respectively).

As an example, assume a rater recorded 60 m2 of alligator cracking in the section. Of this total, 20 m2 was of low severity and 40 m2 was of medium severity. The actual values totaled 80 m2 : 30 m2 of low severity and 50 m2 of moderate severity. Thus, the rater's grade for this particular distress is as follows:

Points Variance

Total Quantity Actual Rater (%) Possible Actual

Distress 80.0 60.0 25.0 7.0 5.3 Low severity 30.0 20.0 33.3 0.5 0.3 Moderate

severity 50.0 40.0 20.0 1.0 0.8 High severity 0.0 0.0 0.0 1.5 1.5

The sum of the points from rating the variance is 7 .9 out of a total possible of 10 points. Also note that if there is no distress at one severity level, the correct determination that no distress is present is given full credit.

The total number of points received for each distress type is then weighted for the significance of the distress. Table 1, for example, presents the weight factors for all distress types for AC-surfaced pavements. Thus, for the above example, an additional weighting factor of 5 (for alligator cracking, see Table 1) is applied to the number of points computed on the basis of the rater variance. Or the maximum possible number of points for this distress type is equal to the distress weight (5) times the maximum number of points gained from complete accuracy in recording (10), or 50 points. When the variance rating is weighted for the distress type, the points-received value becomes 39.5 out of a maximum possible 50.

Deductions also are imposed on the sum of points received for all distress types when the rater misses a distress type or records a distress type not identified by the committee. This deduction is set at 2 percent of the total number of possible points in the section and is assessed for each occurrence of

Rada et al. 3

TABLE 1 AC Pavement Distress Assessment Parameters

DISTRESS TYPE

Cracking. 1. Alligator (Fatigue) Cracking 2. Block Cracking 3. Edge Cracking 4. Longitud. Cracking - Edge

Length Length Sealed

4. Longitud. Cracking - Other Length Length Sealed

5. Reflection Cracking at Joints Number Length (Transv. Joints) Length Sealed (Transv.) Length (Longt. Joints) Length Sealed (Longt.)

6. Transverse Cracking Number Length Length Sealed

Patching and Potholes 7. Patch/Patch Deterioration

Number Area

8. Potholes Number Area

Surface Deformation 9. Rutting

10. Shoving Number Area

Surface Defects 11. Bleeding 12. Polished Aggregate 13. Raveling and Weathering

Miscellaneous Distress 14. Lane-to-Shoulder Dropoff 15. Lane-to-Shoulder Separation

Length Length Sealed

16. Water Bleeding and Pumping Number Length

missed or made-up distress. The final grade is the ratio of adjusted points received (points received minus deductions) to the maximum possible in the section:

m n

2: (DstWgti x points) - 2: deductsk 15 + ~·=_1=--~~m~~~~~~~--"k~=~l~~~~

where

grade; DstWgti

pointsi pointsmax,j =

deductsk =

2: DstWgt X pointsmax .. i j= 1

final grade for section surveyed by ith rater, distress weight applied to jth distress, points received by rater for jth distress, maximum possible number of points for jth distress, and deduct points for kth distress (missed or not identified).

A complete example of the field accreditation scoring system is presented in Table 2 for an AC-surfaced pavement.

UNIT WEIGHT

Square Meters 5 Square Meters 5 Meters 5

Meters 5 Meters 0.5

Meters 5 Meters 0.5

Number 5 Meters 3 Meters 0.5 Meters 3 Meters 0.5

Number 5 Meters 3 Meters 0.5

Number 2 Square Meters 1


Millimeters 2


Square Meters 0.5 Square Meters 0.5 Square Meters 0.5

Millimeters 2

Meters 2 Meters 0.5

Number 2 Meters 1

Also note that a constant (i.e., 15 points) was introduced in the equation to allow for, in a very crude fashion, "reasonable" deviations· from the ground truth values so that the raters are not unduly penalized. This value was established on the basis of the results of the pilot workshop, which are discussed later. However, because experience with the anticipated variability of the measurements is gained through future implementation of the accreditation process, it is recommended that a measure of variability (e.g., actual value ± one standard deviation for each distress type) be included in the scoring system, instead of using a somewhat arbitrary constant. In the interim, it is recommended that, for each accreditation site, a constant (say, 10 to 20 points) derived on the basis of the variance results of a committee of experienced raters be used for grading purposes.

In terms of the overall accreditation grade, the written examination is worth 20 percent of the total score, whereas the field survey portion is worth 80 percent. To receive accreditation, a rater must achieve a combined 75 percent grade for the written and field examinations, but no less than 70 percent on either portion. The passing grades noted are expected to affirm the competence of the raters in distress data collection.

4 TRANSPORTATION RESEARCH RECORD 1410

TABLE 2 Example of Accreditation Grading System for Field Surveys: AC-Surfaced Pavements

Rater Observations Distress Type Units Total Low Mod.

Cracking 1. Alligator cracking - Square 17.8 0.4 17.4

Area Meters 4. Longitudinal Cracking: Meters 4.7 4.7 o.o

Edge - Length 4. Longitudinal Cracking: Meters 86.3 21. 3 39.7

Other - Length 4. Longitudinal Cracking: Meters o.o o.o 0.0

Other - Length Sealed 6. Transverse cracking - Number 52 21 17

Number 6. Transverse Cracking - Meters 77.7 20.0 29.0

Length 6. Transverse Crac~ing - Meters 0.0 o.o o.o

Lenath Sealed

* Based on control group survey

ACCREDITATION WORKSHOPS

The accreditation of RCO raters is being administered by SHRP in a workshop situation. The raters are brought to a single location for 1 week of classroom and field work. The workshop agenda covers classroom sessions, field survey exercises, field survey examinations, and written examination.

Classroom sessions are limited in scope because of the level of experience required for attendance. Primary emphasis is on any changes or revisions to the DIM and field procedures. However, a general review of distress types is conducted using slides and video to reinforce the attendees' knowledge of the most current DIM and field procedures. Time is available for questions and any discussion required to help raters clearly understand the subject matter.

Field survey exercises are conducted as a calibration of the raters. For each pavement type, short pavement sections (60 to 90 m) will have been selected and surveyed by the committee of experienced raters before the start of the workshop. On each of these sections, the RCO raters are required to identify and measure the distresses present. Sections in the early portion of the field exercises are more complex to identify the level of experience of the raters and areas of conftision and error that should be addressed in additional review and discussion. The next test section consists of fewer examples of distresses, with additional time spent in detailed walkdown of the site and discussion of the individual distresses. The objective of these surveys is to determine the individual rater's bias and, as necessary, retrain or correct that individual's misperceptions. The use of field surveys is superior to photographs or video in this determination.

The field survey and written examinations are, as indicated earlier, intended to appraise the capabilities of RCO raters in observing and recording distress data and to assess their specific knowledge of the field procedures and distress definitions. The field survey examinations are conducted after completion of the field survey exercises for each pavement type, scheduled at the end of the second and fourth days, whereas the written examination is administered on the last day of the workshop.

To date, the SHRP accreditation workshop has been conducted twice, both times in Reno, Nevada. Reno was selected for climatic reasons (very little rainfall), thus minimizing de-

Scorinq suminarv Actual Quantities * Possible Points Deduct

Hiqh Total Low Mod. Hi ah Points Received Points

o.o 14.9 1.9 13.0 o.o 50.0 39.2 o.o

o.o 0.0 o.o o.o o.o 0.0 o.o 3.7

25.3 89.1 22.1 44.0 23.0 50.0 47.6 0.0

o.o 0.8 0.8 o.o o.o 5.0 0.0 3.7

14 56 25 20 11 50.0 44.3 0.0

28.7 74.6 21.8 28. 2 24.6 30.0 28.2 o.o

o.o o.o o.o o.o 0.0 o.o o.o o.o

Totals: 185.0 159.3 7.4 Grade: 97.1

lays or postponement of field activities associated with the workshop. The first workshop took place in May 1992 and the second one in June 1992. Both of these workshops are discussed next, along with a sum.mary of the major observations and conclusions.

Pilot Workshop

Before the pilot workshop, various planning activities were undertaken by SHRP to ensure the success of this and future workshops . .To assist in these activities, an accreditation committee was formed to finalize all workshop plans, materials, and selection of survey sections. This committee was composed of representatives from SHRP, SHRP contractors, FHW A, and Texas Department of Transportation. The initial activity consisted of visits to potential test sites to assess their suitability for use in the workshop. This effort resulted in the selection of the following sections:

•Lemon Drive, Lemon Valley, Nevada-complex and simple sections, AC-surfaced pavements.

• McCarren Boulevard, westbound (SHRP General Pavement Study Section 321021); Reno, Nevada-accreditation section, AC-surfaced pavement.

•Interstate I-80 westbound, Hirschdale, California-complex section, jointed concrete pavement.

• US-395, southbound, Reno, Nevada-simple and accreditation sections, jointed concrete pavement.

Once the site selections had been made, it was decided that a control group was necessary to provide detailed distress surveys for the test sections chosen for the workshop. This group consisted of the accreditation committee members and one representative from each of the four SHRP regions, each ostensibly the most knowledgeable distress rater at that RCO office. Their results, as noted earlier, would serve as the ground truth distress data against which the individual rater's results would be compared for grade.

The pilot workshop took place during May 1992. Although the ultimate goal of the workshop was the accreditation of RCO raters, several other objectives were targeted during this workshop:

Rada et al.

• To determine the feasibility of the planned accreditation workshops,

• To assess the grading system developed for the accreditation of the RCO raters,

• To assess RCO rater variability, and • To establish actual or ground truth distress data for use

in near-future workshops.

To allow for site selection and general setup activities before the arrival of the RCO raters, this workshop was limited to 3 days. The accreditation committee met during the 2 days before the pilot workshop to select the survey sites and to finalize workshop plans and materials. For the 3-day workshop, this committee was joined by the remainder of the control group members, the RCO representatives.

Classroom activity, significantly shortened to fit within the 3-day time frame, was limited to review of the DIM revisions and new data forms. Field activities consisted of individual distress surveys of the complex sections, conducted by each member of the control group, followed by a walkdown of the sites and discussion of the distresses observed. Although it was intended that a group survey would be conducted on the complex sites, insufficient time was available because of the tight schedule and traffic control restrictions.

A thorough review and discussion of the results for the complex sections led to a number of changes to the SHRP distress identification manual (i.e., DIM), which. were aimed at eliminating the ambiguity associated with some of the distress definitions. Once these issues had been resolved, individual surveys were performed by the control group raters on the accreditation sites (time did not permit individual or group surveys of the simple sections). The results of the accreditation site surveys, and in particular rater variability, were also reviewed and discussed before conduct of the control group surveys for the accreditation sites, which was the last field activity for each pavement type. Both the accreditation committee and RCO raters walked the sections as a group, identifying all distresses present and mapping them. Where disagreements occurred, the alternate viewpoints were discussed by the group before reaching a final decision.

The last pilot workshop activity was the written examination, which included the identification of distresses from slides and short-answer questions relating to distress definitions and field procedures. Although initially envisioned as an openbook exam because RCO raters must have the DIM with them when conducting LTPP surveys, it was decided to proceed with a closed-book examination. This did not seem to be a problem, as all RCO raters scored very high in the exam (more than 90 percent). It was also decided by the accreditation committee to add to the examination a question dealing with the interpretation and summary of distresses from maps.

Overall, the pilot accreditation workshop is considered a success, as all targeted objectives were satisfactorily completed. The concept of accreditation workshops for SHRPLTPP RCO raters was shown to be feasible; measures of rater variability were established (albeit limited); the grading system was shown to work satisfactorily, although changes will be required in the future to account for the inherent variability associated with subjective ratings; and ground truth values were established for the accreditation sites that will be used in future workshops. Besides the ambiguity problems asso-

5

ciated with the DIM and their impact on the individual accreditation surveys, the only other difficulty encountered in the workshop was the change in measurement units from English, to which the RCO raters were accustomed, to the International System of Units. This problem, however, was quickly overcome after the first couple of surveys.

First Full-Scale Workshop

The first full-scale workshop took place during June 1992. Attendance was limited to three persons per region (i.e., total of 12 participants) to allow for adequate discussion time and to make field surveys easier. As a result, the classroom time was more than adequate to answer questions, and the field activities were completed within the time allotted.

On Day 1 of the workshop, the first activity consisted of approximately 3 hr of review and discussion of distresses in asphalt-surfaced pavements. Overhead transparencies of the definitions and sketches along with slides of actual examples of distresses were used to instruct attendees on the DIM. The final portion of the classroom session consisted of a presentation on specific procedures recommended for performing surveys, including the sequence of activities, setup of the forms, presurvey walkdown, and use of equipment all intended to help the raters perform systematic and reliable surveys in an efficient manner. Field activities commenced on the afternoon of Day 1 with individual surveys of a complex site located on Lemon Drive, north of Reno. A 90-m section was surveyed by each individual for comparison to the. pilot workshop findings and to assess the relative abilities of the attendees. The time required for this exercise necessitated that detailed evaluation of individual results be conducted the following day.

On Day 2, the complex site on Lemon Drive was reviewed during a walkdown along with discussion directed at resolving differences in distress identification. In addition, examples of some distress types found outside the test sections were examined and discussed as a supplement to the classroom slide presentation. This activity was followed by individual surveys of a 60-m section, which allowed the raters to use'the "adjusted" definitions; that is, the corrections and clarifications resulting from the complex section surveys and discussions. The afternoon of Day 2 consisted of the individual surveys of the 150-m accreditation site on McCarren Boulevard. All raters were given 3 hr to map and reduce the distresses for entry on distress data sheets.

Day 3 of the workshop consisted of morning classroom sessions for review of the results of the asphalt-surfaced pavement accreditation surveys followed by separate sessions on distresses found in continuously reinforced pavement and jointed portland cement concrete pavement. Overhead transparencies of the definitions and forms and slides of typical distresses were used in the same way as that for the asphaltsurfaced pavement classroom sessions to familiarize the raters with changes to the DIM. The afternoon of Day 3 consisted of field activities conducted in I-80 in California. A 90-m section of this highway was marked out for individual surveys of a complex jointed concrete pavement. Before the surveys, a group walkdown and discussion were held to orient the raters to the site conditions. Sufficient time was available for the raters to complete their surveys and then conduct another

6

group walkdown to discuss and compare the findings. This served to identify differences in interpretations of the defects observed as well as reinforcing and supplementing the classroom presentation.

Day 4 consisted of field activities on jointed concrete sections at the US-395 site in Nevada. In the morning, a 60-m section was surveyed by each rater. The results of the morning survey were discussed in a walkdown before departing the site and returning to the classroom. In the afternoon, the raters were returned to the site to perform individual surveys of the 150-m accreditation section, also on US-395.

Day 5 activities were conducted in the classroom. These consisted of a review of the jointed concrete pavement accreditation surveys followed by the written examination. The examination included identification of distress types from slides, short-answer questions on distress identification, and reduction of distresses from map sheets. Once the examination was completed, the raters were allowed to grade their own papers during a presentation and discussion of the correct answers to all questions.

Results

The distress survey results for the complex and accreditation sites at both workshops are given in Tables 3 through 6: Tables 3 and 4 summarize the results for the AC-surfaced pavement


sections, whereas Tables 5 and 6 summarize those for the jointed concrete pavement sections. In each table, the mean and standard deviation for each distress type-severity level combination are given, along with the ground truth values where available. Because the pilot workshop was limited to 3 days, time permitted the conduct of group surveys for the accreditation sites only, not the complex and simple sections. Thus, only Tables 4 and 6 contain ground truth distress data. Time constraints during the pilot workshop also prohibited the conduct of individual or group surveys on the simple sections, so no results exist. Surveys on these simple sections were performed during the full-scale workshop but are not discussed in this paper because of the limited data; that is, no basis for comparisons and very little for discussion other than noting that rater variability was generally very good, as indicated by the small standard deviation.

From the information contained in these tables, the following observations and conclusions are made:

•AC-Surfaced Pavement-Complex Section (Table 3): In general, there is good agreement between the results from both workshops for this section. Significant differences do occur in the amount of alligator cracking and longitudinal cracking identified by the raters at each workshop. These differences, however, are almost exclusively caused by changes made to the DIM during the pilot workshop before the full workshop surveys. Because the DIM did not clearly distin-

TABLE 3 Between-Rater Statistical Summary: AC-Surfaced Pavement, Complex Site

Means Standard Deviations Severity Actual Pilot First Pilot First

Distress TvPe Units Level Values * Workshop Workshop Workshop Workshop Alligator Cracking - Area sq. Meter Low n/a 5.1 a.4 5.4 9.4

Moderate n/a a.a· 14.6 3.9 6.4 Hiqh n/a 0.9 2.a 0.7 3.3 Total n/a 14.a 25.a 6.1 11.6

Longitudinal Cracking: Other - Meters Low n/a 7.3 0.3 5.0 0.5 Length Moderate n/a 7.7 o.o 9.4 o.o

Hiqh n/a 1. 5 0.5 1. 5 1. 2 Total n/a 16.4 o.a 11.a 1.2

Longitudinal Cracking: Other - Meters Low n/a 0.2 0.4 0.5 0.6 Length Sealed Moderate n/a o.o o.a o.o i.a

Hiqh n/a 0.1 0.0 0.2 o.o Total n/a 0.3 1.2 0.5 2.1

Transverse Cracking - Number Number Low n/a 41.6 - 40.7 10.9 a.1 Moderate n/a 19.a 1a.o 10.9 10.2 Hiah n/a 3.4 0.7 4.3 o.a Total n/a 64.a 59.3· a.a 4.3

Transverse Cracking - ~ength Meters Low n/a 39.3 51.9 15.3 14 Moderate n/a 45.6 46.6 19.9 20.2 Hi ah n/a 10.2 2.4 13.7 2.7 Total n/a 95.1 100.a 19.5 17.5

Transverse Cracking - Length Meters Low n/a 7.3 14.a 6.a 7.3 Sealed Moderate n/a 14.4 19.9 11. 3 16.5

Hiqh n/a 5.4 1.2 12.9 1. 7 Total n/a 27.1 35.9 22.5 23.6

Patch/Patch Deterioration ~ Number Low n/a 0.1 0 0.3 0 Number Moderate n/a 0.1 0 0.3 0

Hiqh n/a 0.6 1 0.5 0 Total n/a o.a 1 0.6 0

Patch/Patch Deterioration - Sq. Meter Low n/a <0.1 0 <0.1 0 Area Moderate n/a <0.1 0 <0.1 0

Hi ah n/a 0.4 0.3 0.6 0.2 Total n/a 0.5 0.3 0.6 0.2

Ravelling and Weathering - Sq. Meter Low n/a 104.0 o.o 147.3 0.0 Area Moderate n/a 50.0 54.0 11i.a 120.a

Hi ah n/a 26.3 o.o a1.1 o.o Total n/a 1ao.3 54.0 152.6 120.a

TABLE 4 Between-Rater Statistical Summary: AC-Surfaced Pavement, Accreditation Site


Distress Tvpe Units Level Values * Workshop Workshop Workshop Workshop Alligator Cracking - Area sq. Meter Low 1.9 10.0 4.8 9.4 5.2

Moderate 13.0 6.0 15.1 6.8 11.8 Hiqh o.o 0.7 3.0 0.2 4.2 Total 14.9 16.1 22.9 9.3 11.1

Longitudinal Cracking: Edge - Meters Low o.o 6.3 o.o 8.6 o.o Length Moderate o.o 2.3 o.o 3.3 o.o

Hiqh o.o 24.9 0.0 35.9 o.o Total o.o 33.5 o.o 46.6 o.o

Longitudinal Cracking: Other - Meters Low 22.1 27.2 31.4 3.0 13.8 Length Moderate 44.0 26.7 25.1 8.3 18.2

Hiqh 23.0 26.9 34.6 6.6 21.6 Total 89.1 .so.a 88.8 3.2 14.6

Transverse Cracking - Number Number Low 25 27.0 28.2 3.4 7.5 Moderate 20 11. 5 11.8 4.7 4.7 Hiqh 11 14.7 11.3 3.0 4.4 Total 56 53.2 51.9 6.1 6.8

Transverse Cracking - Length Meters Low 21.8 24.5 27.7 3.4 10.6 Moderate 28.2 19.1 21.5 6.1 10.3 Hiqh 24.6 33.5 23.4 7.9 10.4 Total 74.6 77.1 70.8 8.0 24.0


TABLE 5 Between-Rater Statistical Summary: Jointed Concrete Pavement, .Complex Site


Distress Type ·Units Level Values * Workshop Workshop Workshop Workshop Longitudinal Cracking - Meters Low n/a 18.7 25.3 9.6 6.4 Length Moderate n/a 51. 5 56.9 21. 2 12.2

Hiqh n/a 29.8 11.4 20.2 8.4 Total n/a 100.0 93.6 4.2 8.3

Longitudinal Cracking - Meters Low n/a 0.5 0.3 0.8 0.6 Length Sealed Moderate n/a 0.2 0.6 0.5 1.0

Hiqh n/a o.o 0.1 o.o 0.3 Total n/a 0.7 0.9 0.8 1.3

Transverse Cracking - Number Low n/a 7.6 10.2 1.6 2.4 Number Moderate n/a 10.0 9.3 4.3 1.8

Hiqh n/a 2.6 0.6 2.7 1.4 Total n/a 20.2 20.0 2.4 2.6

Transverse cracking - Meters Low n/a 12.0 17.5 5.8 5.0 Length Moderate n/a 26.1 31.3 7.6 6.3

Hiqh n/a 8.5 1. 7 9.8 3.7 Total n/a 46.6 50.5 5.0 3.9

Transverse Cracking - Meters Low n/a 0.1 o.o 0.1 o.o Length Sealed Moderate n/a 0.6 o.o 0.8 o.o

Hiqh n/a o.o 0.0 0.0 0.0 Total n/a 0.6 0.0 0.8 o.o

Joint Seal Damage of Number Low n/a 0.6 0.6 1.2 1.2 Trans •. Joints - Number Moderate n/a 8.4 5.3 3.9 4.0

Hiqh n/a 5.8 6.2 3.5 4.8 Total n/a 14.8 12.1 0.4 5.5

Joint Seal Damage of Number Total n/a 2 1.8 0 0.4 Lonq. Joints - Number Joint Seal Damage of Meters Total n/a 61. 5 53.1 4.2 20.9 Lonq. Joints - Lenqth Spalling of Longitudinal Meters Low n/a 0.2 3.3 0.4 10.4 Joints - Length Moderate n/a o.o 3.8 0.0 12.5

Hiqh n/a o.o 0.5 o.o 1. 7 Total n/a 0.2 7.5 0.4 24.5

Spalling of Transverse Number Low n/a 4.4 0.6 5.6 1.0 Joints - Number Moderate n/a o.o 0.2 0.0 0.6

High n/a o.o 0.2 o.o 0.6 Total n/a 4.4 0.9 5.6 1.0

Spalling of Transverse Meters Low n/a 11.1 0.3 20.7 0.6 Joints - Length Moderate n/a o.o 0.2 o.o 0.5

Hiqh n/a o.o 0.1 o.o 0.1 Total n/a 11.1 0.5 20.7 0.7

Popouts - Number Number Total n/a 220.6 114.8 609.6 590.l * Based on control group survey

8 J:RANSPORTATION RESEARCH RECORD 1410

TABLE 6 Between-Rater Statistical Summary: Jointed Concrete Pavement, Accreditation Site

Severity Actual Distress Tvpe Units Level Values *

Corner Breaks - Number Number Low Moderate Hiah Total

Longitudinal cracking - Meters Low Length Moderate

Hi ah Total

Longitudinal Cracking - Meters Low Length Sealed Moderate

Hi ah Total

Transverse Cracking - Number Low Number Moderate

Hi ah Total

Transverse Cracking - Meters Low Length Moderate

Hi ah Total

Transverse Cracking - Meters Low Length Sealed Moderate

Hi ah Total

Joint Seal Damage of Number Low Trans. Joint - Number Moderate

Hiah Total

Spalling of Longitudinal Meters Low Joints - Length Moderate

Hiah Total

Spalling of Transverse Number Low Joints - Number Moderate

Hiah Total

Spalling of Transverse Meters Low Joints - Length Moderate

Hi ah Total

Pooouts - Number Number Total Based on control group survey

guish between low-severity alligator cracking and longitudinal cracking (other than construction) in the wheelpath, it was collectively decided at the pilot workshop that single, longitudinal cracks within the wheelpath should be defined as lowseverity alligator cracking. Thus, the differences shown in Table 3 reflect the impact of the DIM change on the survey results. The only other major difference occurred in the amount of raveling and weathering, where significantly higher quantities were identified by the raters at the pilot workshop. This difference is attributed to the lack of familiarity of the raters with the construction materials used at the site and the effect of studded tires on the pavement surface, both of which were explained before the full workshop surveys.

•AC-Surfaced Pavement-Accreditation Section (Table 4): Unlike the previous section, ground truth distress data were available for this accreditation site in addition to the results from the individual surveys performed during the two workshops. With the exception of longitudinal cracking (edge or construction), there is excellent agreement among all raters within and between workshops, as reflected by the similar means and low standard deviations. Even the differences in longitudinal cracking are somewhat misleading in that only one (out of four) RCO rater at the pilot workshop incorrectly

2 7 0 9

11.3 4.5 3.0

18.8 0.0 o.o 3.0 3.0

0 1 2 3

0.00 3.70 7.40 11.1 o.o 3.7 7.4

11.1 32

0 0

32 15.0 o.o o.o

15.0 2 0 0 2

0.6 o.o o.o 0.6

18

Means Standard Deviations Pilot First Pilot First

Workshop Works hoc Workshop Workshop 4.6 2.8 3.1 1.9 4.6 5.8 2.8 2.1 0.8 0.3 1.2 0.6

10.0 8.8 1.1 1.2 10.0 11.8 3.6 1.1 3.3 3.6 2.9 1.4 1. 3 0.3 1.1 1.1

14.7 15.7 4.9 1.1 0.0 o.o o.o o.o 0.5 1.6 1.0 1.1 1.3 0.2 1.1 0.7 1.8 1.8 0.9 1.1 0.8 0.8 0.4 0.6· 0.6 1.1 0.8 1.0 1.4 1. 7 0.8 1.6 2.8 3.5 0.4 1. 7 3.0 2.5 1.5 1.8 1.5 3.7 1.9 3.7 5.3 4.8 3.1 4.0 9.8 11.0 2.9 2.1 o.o o.o 0.0 o.o o.o 2.5 o.o 3.1 3.7 4.4 3.3 3.9 3.7 6.9 3.3 3.1

31.8 32.1 0.4 0.3 0.2 o.o 0.4 o.o o.o o.o o.o o.o

32.0 32.1 o.o 0.3 17.4 12.2 11.2 5.6 0.2 1.1 0.3 0.8 o·.o 0.2 o.o 0.3

17.5 13.5 11.2 6.0 5.4 0.5 3.4 1.1 2.6 0.0 3.3 o.o 0.2 0.2 0.4 0.6 8.2 0.7 5.1 1.3 2.3 0.1 1.1 0.2 0.9 o.o 1.2 o.o 0.1 0.1 0.1 0.1 3.3 0.2 1. 7 0.3 1.8 28.2 2.2 14.3

identified this distress type, instead of alligator cracking, which was the "correct" distress. There are also small differences in the quantities of alligator cracking shown in Table 4, but these are almost entirely because of differences in the way widths were defined by the raters for low-severity alligator cracking (generally a single crack). Guidelines for measuring these widths were developed during the full workshop as a result of the observed differences and will be implemented in future workshops.

• Jointed Concrete Pavement-Complex Section (Table 5): Survey results for this section were similar for both the pilot and first full workshops. The major differences between the two were in the quantities of joint spalling and popouts identified by the RCO raters in the respective workshops. These differences are also attributed to the DIM changes that took place after the pilot surveys but before the full workshop surveys; thus, they were to be expected. Otherwise, the two sets of surveys are in excellent agreement. Furthermore, it is noted that the standard deviation is generally low for most distress-type-severity level combinations, indicating consistency among all raters.

• Jointed Concrete Pavement-Accreditation Section (Table 6): The results for this section are similar to those of the

Rada et al.

complex section in that, with few exceptions, they were similar for both workshops. In addition, the results are also similar to the ground truth values established by the control group, particularly those from the full workshop. In general, the major differences between the pilot survey results and those of the control group and the full workshop surveys are a result of changes to the DIM made after the pilot surveys. To a lesser degree, some of the differences can simply be attributed to rater variability. Overall, the results of these accreditation surveys are quite good in that they show very consistent, uniform results among the raters, that is, similar means and low standard deviations for most distress-type-severity level combinations.

Overall, the rater variance for the first full-scale workshop was slightly lower than that of the pilot workshop. It is hypothesized by the authors that this is because of (a) changes made to the DIM identification and quantification procedures during the pilot workshop; (b) greater emphasis by the instructors at the full-scale workshop on certain distress types that were found to be a problem during the pilot workshop; and (c) changed pavement conditions (surface temperatures at the pilot workshop were significantly lower than those at the full-scale workshop).

Looking now at how these survey results translate into grades, and hence the accreditation of the 16 RCO raters who participated in the workshops, Table 7 summarizes the scores received by the raters for each accreditation site and the written examination as well as the final (composite) accreditation grade. All scores are based on a scale of 0 to 100, with 100 being excellent. For the most part, the scores are in the goodto-excellent range (80 to 100 percent). Also, the composite score for all RCO raters exceeded 75 percent, whereas their individual survey and written examination grades exceeded 70 percent, thus satisfying the accreditation criteria established by SHRP. These results were by no means unexpected as all raters involved in the workshops had 2 or more years of experience in the conduct of field distress surveys using

TABLE 7 Accreditation Workshop Scores

RCO Rater Flexible Workshop ID Section

Pilot Workshop 1 82 2 82 3 99 4 72

Average: 84 Std. Dev.: 10

Full Workshop 1 85 2 73 3 79 4 70 5 99 6 77 7 86 8 85 9 84

10 97 11 92 12 75

Average: 84 Std. Dev.: 9

Combined Average: 84 Statistics Std. Dev.: 9

9

SHRP procedures. The workshop results were also· encouraging in terms of the consistency of the distress data being collected by the RCO contractors.

SUMMARY AND CONCLUSIONS

The purpose of SHRP's accreditation process is to provide a means for ensuring, to the extent possible, the quality and consistency of distress data being collected by the RCO raters. The process consists of two parts: a written examination and a two-part field survey examination. The successful completion of these examinations will identify the rater as possessing the knowledge, competence, and accuracy to provide distress data of acceptable reliability for inclusion in the L TPP data base. Although the process is still in its early stages, it is SHRP's intent that all distress data for the L TPP study be collected by raters who have successfully completed the accreditation.

The SHRP accreditation process is being administered in a workshop situation, involving both classroom and field work. To date, the workshop has been conducted on two separate occasions, both in Reno, Nevada. The first, a pilot workshop, took place in May 1992, and the other in June of 1992. Although the ultimate objective was the accreditation of RCO raters, several other objectives were targeted during these workshops and successfully completed:

• The concept of accreditation workshops for RCO raters was shown to be feasible;

• Preliminary measures of rater variability were established;

• The accreditation grading system was shown to work satisfactorily; although changes will be required to account for the subjective nature of distress surveys; and

• Ground truth distress values were established for two accreditation sites that will be used in future workshops.

Rigid Written Final Section Examination Grade

80 92 83 70 90 79 81 96 91 88 97 83 80 94 84

6 3 4 99 98 93 80 93 80 71 96 79 79 90 78 88 88 92 83 80 80 97 94 92 76 83 81 81 90 84 94 91 95 73 93 85 77 89 79 83 90 85

9 5 6 82 91 85

8 5 6

10

In terms of the ultimate objective, all 16 RCO raters who attended the workshops successfully completed the accreditation process (i.e., satisfied the accreditation. criteria established by SHRP). This is not surprising since the RCO raters who participated in the accreditation workshops have had several years of experience in the conduct of SHRP distress surveys. Thus, another measure of the success of the accreditation process will come as additional workshops are conducted involving less-experienced personnel.

Another important outcome resulting from the initial accreditation workshops were revisions to the SHRP distress identification manual. In all cases, the changes to the manual were made to eliminate as much as possible the ambiguity associated with some of the distress definitions. Further revisions to the manual may be required as experience with distress surveys is gained.

Finally, although the accreditation process has proven quite successful so far, improvements can be made in a number of areas:

• Revision of the accreditation scoring system for the field examination to incorporate the inherent variability associated with subjective distress surveys; that is, a measure of the anticipated variability, as determined from several workshops, should be included in the scoring system.

• Inclusion of a continuously reinforced concrete pavement section as part of the field examinations. Such a section was not included in the initial workshops for two reasons: (a) there were no such sections within the vicinity of Reno and (b) there were time constraints.

• Because many distresses have the tendency to take on certain appearance characteristics on the basis of climatic (regional) conditions, it may be worthwhile to establish several accreditation sites (e.g., one for each SHRP RCO) throughout the country, with the workshops alternating from one site


to another. This would expose the RCO raters to different appearances of the same distress type.

ACKNOWLEDGMENTS

The work described in this paper was performed by PCS/Law Engineering under contract to SHRP, National Research Council. The authors gratefully acknowledge the cooperation and assistance of the SHRP Western Regional Coordination Office Contractor, Nichols Consulting Engineers, Chtd.

REFERENCES

1. Permanent Distress Record Collection and Transverse Profile Analysis Using PASCO USA's ~oadrecon Survey Systems. Strategic Highway Research Program, National Research Council, Washington, D.C., May 1992.

2. Distress Interpretation from 35mm Film for the LTP P Experiments. Strategic Highway Research Program, National Research Council, Washington, D.C., June 1992.

3. Manual for Distress Surveys. Strategic Highway Research Program, National Research Council, Washington, D.C., May 1992.

4. Distress Identification Manual for the Long-Term Pavement Performance Studies. Strategic Highway Research Program, National Research Council, Washington, D.C., June 1989.

5. Accreditation of Manual Distress Survey Raters. Strategic Highway Research Program, National Research Council, Washington, D.C., June 199i

Publication of this article does not necessarily indicate approval or endorsement by the National Academy of Sciences, the United States government, or AASHTO or its member states of the finding, opinions, conclusions, or recommendations either inferred or specifically expressed herein.

Publication of this paper sponsored by Committee on Pavement Mon~ itoring, Evaluation, and Data Storage.

TRANSPORTATION RESEARCH RECORD 1410 11

Comparison of Pavement Surface Distress Measurement Systems

RAYMOND K. MOORE, G. NORMAN CLARK, AND ANDREW J. GISI

Two state-of-the-art (late 1989) pavement distress data collection devices were used to evaluate 15 bituminous-surfaced test sections 152 m (0.1 mi) long. The Infrastructure Management Services (IMS) road surface tester was a laser-based system that produced a comprehensive array of rutting and cracking statistics for nominal pavement segments 0.16 km (0.1 mi) long. The PAVEDEX PAS-I system rec~rded the pavement surface condition on videotape, which was later visually analyzed by PA VEDEX technicians using Kansas Department of Transportation (KDOT) pavement management system network-level distress identification criteria. Data from the two systems were compared with distresses measured and mapped by KDOT engineering technicians using a ground survey. The average maximum rut depth measured by the IMS laser system provided a relatively precise estimate of rut depth severity. Only one linear correlation between IMS cracking data and KDOT distress data was significant at the 5 percent level. Given the comprehensive array of ten IMS cracking width and depth measurements, the absence of linear association with KDOT field data was unexpected. PAVEDEX video data generally detected the presence of transverse and fatigue cracking but had difficulty in assigning the correct KDOT severity code because perceived roughness associated with transverse cracking and differences between hairline and spalled fatigue cracking are used as criteria. Transverse cracks with secondary cracking were interpreted to be block cracking. As a general conclusion, the study indicated that the current KDOT distress rating criteria are not compatible with the capabilities of the two distress measurement systems.

The Kansas Department of Transportation (KDOT) conducts an annual network-level pavement condition survey that requires 4 months and evaluates 17 700 km (11,000 mi) of in-service pavement subdivided into nominal 1.6-km (1-mi) sections. Distress data are collected using a sample of three randomly selected 30.5-m (100-ft) segments in each 1.6-km (1-mi) section of pavement. The randomized selection process is repeated each year, so it is highly improbable that the same segments are evaluated on a year-to-year basis. Therefore, the annual distress survey is based on about a 6 percent sample.

PURPOSE OF STUDY

KDOT is interested in new technology that would reduce distress survey time at perhaps lower cost because the current process is labor intensive as it is based on visual inspections. The use of equipment has the potential to reduce or eliminate both rater-to-rater and time-dependent variations in the field

R. K. Moore, Department of Civil Engineering, University of Kansas, Lawrence, Kan. 66045. G. N. Clark and A. J. Gisi, Materials and Research Center, Kansas Department of Transportation, Topeka, Kan. 66611.

data. Near the end of the survey, visual rating consistency may also become compromised by the tediousness of the process. Furthermore, equipment that can travel at highway speeds eliminates the safety problems associated with temporary work zones that interrupt normal traffic flow.

Data Collection Equipment and Methodology

Two state-of-the-art (late 1989) pavement distress data collection devices using different technologies were selected for the study. The Infrastructure Management Services (IMS) road surface tester was a laser-based system and produced a comprehensive array of rutting and cracking statistics for nominal pavement segments 0.16 km (0.1 mi) long. The PA VEDEX PAS-I video system recorded the pavement surface condition on videotape. Subsequently, the tapes were visually analyzed by PAVEDEX technicians using KDOT pavement management system (PMS) crack distress criteria. The data were reported for nominal 0.16-km (0.1-mi) pavement segments using the standard format developed for the agency's annual network-level _survey. Rutting was not measured by the PAVEDEX system in late 1989.

KDOT Pavement-Type Classifications

Three KDOT PMS pavement types have asphalt concrete surfacing. FDBIT refers to full-design bituminous pavement; construction of the pavement section was based on conventional thickness design and prudent engineering practice. PDBIT is assigned to flexible pavement sections that have evolved through repeated application of surface treatments or overlays. Although contemporary resurfacing mixture and thickness design are based on current engineering practice, the complete pavement cross section has been constructed with only partial reliance on formal design procedures. COMP refers to 'composite pavements; a bituminous overlay has resurfaced the original rigid pavement.

KDOT Distress Code Definitions

The KDOT PMS uses a distress classification system that includes rutting, transverse cracking, and fatigue cracking. Definitions for the distress classifications discussed in the reported research are as follows:

•Code 1 transverse cracking has no roughness; cracks 6 mm (0.25 in.) wide or wider with no secondary cracking;

12

or any width crack with secondary cracking less than 1.2 m (4 ft) per lane.

• Code 2 transverse cracking is associated with any width crack with noticeable roughness caused by depression, bump, or wide crack width [in excess of 25 mm (1 in.)] or cracks that have more than 1.2 m (4 ft) per lane of secondary cracking but no roughness.

• Code 3 transverse cracking is assigned to any crack width with significant roughness caused by a depression or bump. Secondary cracking will be more severe than it is in Code 2.

Secondary cracks generally develop parallel and within 150 mm ( 6 in.) of the main transverse crack as a depression begins to form under the action of traffic. The length of secondary cracking referenced in the distress criteria is the summed length of secondary cracks immediately adjacent to the transverse crack across the full roadway width. Roughness is a subjective evaluation as perceived by the evaluator traveling along the pavement segment. Code 1 fatigue cracking is hairline alligator cracking with nonremovable segments, and Code 2 fatigue cracking indicates spalling of the cracks around the nonremovable segments. The detection of spalling generally requires a close visual examination of the fatigue cracks.

EXPERIMENT DESIGN

The two distress measuring systems were used on 15 bituminous concrete-surfaced pavement segments 152 m (500 ft) lorig, whose distresses identified by KDOT engineering technicians were meticulously recorded on crack maps. The objective was to compare the data obtained using the two devices with the distress patterns recorded on the crack maps. Statistical significance for correlation studies and analyses of variance was defined using an alpha level of 5 percent.

Test Section Descriptions

The 15 in-service test sections included 5 of each of the three KDOT asphaltic concrete-surfaced pavement types (COMP, PDBIT, and FDBIT). A typical surface distress diagram or crack map is shown in Figure 1.

Data Set Decriptions

Basic IMS and PAVEDEX data used nominal 0.16-km (0.1-mi) pavement segments. that corresponded in length to the 152-m (500-ft) KDOT test sections. In those cases when these segments did not exactly correspond with the KDOT test sections (i.e., the KDOT test section was overlapped by two IMS or PA VEDEX segments), the average IMS or PA VEDEX data for the two segments were used as the appropriate comparative statistics .

. The s~lected IMS data array relating to rutting and surface cracking consisted of the following elements:

• Average rut depth (in.) in left wheel path; •Average rut depth (in.) in right wheelpath; •Average maximum rut depth (in.) in both wheelpaths;


3/16" 3/16"

~----IA'lr-3/16"

3/16"

1/4" 400

FIGURE 1 Distress diagram (US-54, Woodson County).

1/4"

1/4" 5/18"

•Crack count [3 to 6 mm deep; 2 to 4 mm wide (0.12 to 0.24 in. deep; 0.10 to 0.16 in. wide)];

•Crack count [3 to 6 mm deep; 4 to 6 mm wide (0.12 to 0.24 in. deep; 0.16.to 0.24 in. wide)];



•Crack count [greater than 6 mm deep; 2 to 4 mm wide (greater than 0.24 in. deep; 0.10 to 0.16 in. wide)];


•Crack count [greater than 6 mm deep; 6 to 12 mm wide (greater than 0.24 in. deep; 0.24 to 0.47 in, wide)];


• Total crack count (left wheelpath plus right wheelpath) minus the number of cracks detected by both sensors; and

• Total crack count (both inside of wheel path sensors) minus the number of cracks detected by both sensors.

The relevant PA VEDEX test section data for COMP, PDBIT, AND FDBIT pavement types were as follows:

•Number of Code 1 transverse cracks, •Number of Code 2 transverse cracks, • Number of Code 3 transverse cracks, •Lineal feet of Code 1 fatigue cracking, and • Lineal feet of Code 2 fatigue cracking.

KDOT test section data discussed in this paper for COMP, PDBIT, and FDBIT pavement types were as follows:

Moore et al.

•Average rut depth (in.) in left wheelpath on the basis of five string line measurements taken at 30-m (100-ft) intervals;

•Average rut depth (in.) in right wheelpath on the basis of five string line measurements taken at 30-m (100-ft) intervals;

• Maximum of the two average rut depth measurements (in.) taken in the wheelpaths;

•Number of Code 1 transverse cracks; •Number of Code 2 transverse cracks; •Number of Code 3 transverse cracks; •Number of uncoded transverse cracks; •Total number of transverse cracks; •Lineal feet of Code 1 fatigue cracking; • Lineal feet of Code 2 fatigue cracking; and •Total lineal feet of fatigue cracking.

RESEARCH RESULTS

The figures use alphanumeric codings to represent the large number of duplicate data points found in the data sets. Numbers 1 through 9 plotted on a scatter diagram represent that number of duplicate data points. For larger numbers, the alphabet is used with A assigned to 10 data points and progressing in alphabetical order to Z, the code for 35 duplicate data points. An asterisk is used if the number of duplicates· exceeds 35.

IMS Rutting Data

A correlation analysis between the IMS and KDOT rutting data indicated three significant linear associations:

r (IMS average left wheelpath rutting KDOT average left wheelpath rutting) = 0.66

r (IMS average right wheelpath rutting, KDOT average right wheelpath rutting) = 0.84

r (IMS average maximum rut depth in both wheelpaths, maximum of the two average KDOT rut depths taken in the wheelpaths) = 0.86

The data pairs are graphed in Figures 2 through 4. A line of equality is also shown to assist in data interpretation: Figure 2 shows that the average of the KDOT string line measurements tends to be greater than the IMS laser data for the left wheelpath. A total of 10 of the 15 data points are right of the line of equality. Figure 3 shows the same trend for the right wheel path with 10 data points to the right of the line of equality. When maximum rut depth values are considered, the bias is less obvious. Only eight data points are to the right of the line of equality in Figure 4. The major conclusion drawn from these data suggests that the IMS average maximum rut depth provides a relatively precise estimate of rut depth severity.

Analysis of variance (ANOV A) was used to determine if the differences between the IMS rut depth data and the KDOT string line measurements were significantly affected by pavement type (e.g., COMP, PDBIT, FDBIT). In these analyses, the differences between IMS and KDOT data using the left wheelpath data, right wheelpath data, and the maximum rut data were tested. All three ANOV As indicated that pavement type was not a significant factor.

7

E E

6 £ i5.. (I)

a 5 5 a: a.. ~ 4 _J

Q) II> cu _J

(/) 3 ~ (I)

~2 (I) > c(

Average KDOT LWP Rut Depth, mm

FIGURE 2 IMS average left wheelpath rutting versus average KDOT left wheelpath string line rut depth.

10

9

E E 8 £ i5.. (I) 7 a 5 a: 6 a.. ~ a:

5 Q) II> cu _J

4 (/)

~ (I) 3 O> ~ (I) > 2 c(

2 4 6 8 10 12 14

Average KDOT RWP Rut Depth, mm

FIGURE 3 IMS average right wheelpath rutting versus average KDOT right wheelpath string line rut depth.

IMS Cracking Data

13

A similar correlation study was conducted using IMS and KDOT crack data statistics. Only one linear association was significant:

r [total crack count; both wheelpath sensors; 3 to 6 mm deep; 12 to 25 mm deep (0.12 to 0.24 in. deep, 0.47 to 0.98 in. wide), number of Code 1 transverse cracks] = -0.55.

14

10

E E 9 Ci) 0... 8 ~

~ 7 @.

~ 1 a> 6 I 0 1 '5 1 a: 5 1

E ::I E 4 ·x CIS

::::?! a> 3 ~ 1

1 a> 2 > <(

en ~ 1 1

0

0 2 4 6 8 10 12 14 Maximum Average KDOT Rut Depth (Both WPs), mm

FIGURE 4 IMS average maximum rut depth in both wheelpaths versus maximum of two average KDOT string line rut depths taken in wheelpaths.

As the number of Code 1 transverse cracks increases, the number of IMS-detected wheelpath cracks 3 to 6 mm deep and 12 to 25 mm wide decreases. This appears to be consistent with KDOT severity coding, which considers roughness. Code 1 transverse cracking is not associated with perceptible roughnf'.SS. As the transverse cracks become wider, roughness increases and a Code 1 severity rating would no longer be appropriate.

Given the comprehensive array of IMS crack width and depth measurements, the absence of a significant number of correlations with KDOT field data was unexpected. This finding suggests that a basic incompatibility exists between the KDOT distress rating criteria and the IMS cracking statistics. The simplicity inherent in visual distress rating criteria does not take advantage of the laser system ability to accurately measure crack widths and depths. If full benefit of a laserbased system is to be realized, a new distress rating system will be needed to replace the existing manual system that is based on visual observation. For example, cracking characteristics such as perceived roughness, spalled cracks, and nonremovable pieces used in the KDOT distress rating system have no "laser meaning." Crack severity and extent will require redefinition in terms of the IMS statistics such as cracking width, depth, and density.

PA VEDEX Cracking Data

PA VEDEX data were analyzed on the basis of two pavement section lengths. Because the basic data used 0.16-km (0.1-mi) pavement section lengths, these data were compared with the KDOT data summed over the entire 152-m (500-ft) test section. In addition, PA VEDEX data were also reported using the 30-m (100-ft) station lengths. This was done because KDOT


currently uses randomly seleded 30-m (100-ft) segments to collect cracking data during the annual condition survey.

The KDOT test sections were simply subdivided into five 30-m (100-ft) segments, and the appropriate cracking data were summed for each segment. However, PAVEDEX data for each 0.16-km (0.1-mi) section were divided by 5.28 creating five pseudo 30-m (100-ft) segments with identical cracking statistics. These were compared with the five 30-m (100-ft) KDOT segments to study the effect of the potential computational bias on the correlations. PA VEDEX indicated that this computational approach would consistently underestimate the actual lineal feet for a given distress level because of a dilution effect. PAVEDEX also expected the frequency of reported occurrences (i.e., the number of transverse cracks) to be higher than the KDOT data. (D. L. Bender, PA VEDEX, to A. J. Gisi, KDOT, personal communication, Feb. 1990). This bias creates two possible sources of variation between PA VEDEX and KDOT data that do not directly relate to the ability of the technology to detect surface distress.

Although this second analysis increased the number of correlate data pairs by a factor of five, the additional data are not truly independent. The KDOT data for each 30-m (100-ft) segment within the 152-m (500-ft) test section are independent; however, the five pairs of correlates use identical PAVEDEX data. This creates an additional source of computational variation between PAVEDEX and KDOT data, which, again, does not relate to the ·ability of the technology to detect surface distress.

Correlation Analyses

The statistically significant correlation coefficients (a measure of linear association) involving PAVEDEX distress measurements and KDOT extent-severity codes for the three pavement types are given in the following text. As an aid to interpretation, the correlations are subdivided into three general categories: correlations between (a) like distresses and severity codes, (b) like distresses but different severity codes, and ( c) different distresses. The significant correlations associated with both the 30-m (100-ft) and 152-m (500-ft) pavement segment length data bases are shown in the following list. Those linear associations significant for the 152-m (500-ft) pavement segment data base are shown in bold type. If a significant linear association between PA VEDEX (or PDX) and KDOT data existed for the 152-m data set consisting of 15 pairs of correlates, it was also significant using the 30-m (100-ft) pavement segment length data base consisting of 75 pairs of correlates.

Correlation interpretation can be complicated by a nonuniform distribution of data in the factor space. Several of the statistically significant correlations within the 152-m (500-ft) pavement segment data exhibited an elongated data cluster. Many data points were associated with the zero level of one or both of the correlates in conjunction with one or two outlying points. This situation is identified by an asterisk adjacent to the correlation coefficient. Although the correlation may be statistically significant, the engineering inferences are not robust. This finding also placed added importance on the correlations developed from the 30-m (100-ft) data sets.

Moore et al.

• Correlations between like distresses and severity codes:

r (number of PDX Code 1 transverse cracks, number of KDOT Code 1 transverse cracks) = 0.29

r (number of PDX Code 2 transverse cracks, number of KDOT Code 2 transverse cracks) = 0.67, 0.52

r (number of PDX Code 3 transverse cracks, number of KDOT Code 3 transverse cracks) = 0.93*, 0.26

r (lineal feet of PDX Code 1 fatigue cracking, lineal feet of KDOT Code 1 fatigue cracking) = 0.84, 0.83

• Correlations between like distresses and different severity codes:

r (number of PDX Code 1 transverse cracks, number of. KDOT Code 2 transverse cracks) = 0.27

r (number of PDX Code 1 transverse cracks, number of KDOT Code 3 transverse cracks) = 0.23

r (number of PDX Code 1 transverse cracks, total number of KDOT coded and uncoded transverse cracks) = 0.30

r (number of PDX Code 2 transverse cracks, number of KDOT Code 3 transverse cracks) = 0.96, 0.82

r. (number of PDX Code 3 transverse cracks, total number of KDOT coded and uncoded transverse cracks) = 0.24

r (number of PDX Code 3 transverse cracks, number of KDOT Code 2 transverse cracks) = 0. 70*, 0.38

r (lineal feet of PDX Code 1 fatigue cracking, total lineal feet of KDOT fatigue cracking) = 0.84, 0.84

r (lineal feet of PDX Code 2 fatigue cracking, lineal feet of KDOT Code 1 fatigue cracking) = 0.66*, 0.61

r (lineal feet of PDX Code 2 fatigue cracking, total Hneal feet of KDOT fatigue cracking) = 0.64*, 0.60

•Correlations between different distresses:

r (number of PDX Code 1 transverse cracks, lineal feet of KDOT Code 1 fatigue cracking) = -0.53, - 0.48

r (number of PDX Code 1 transverse cracks, total lineal feet of KDOT fatigue cracking) = -0.53, - 0.49

r (number of PDX Code 2 transverse cracks, lineal feet of KDOT Code 1 fatigue cracks) = -0.25

r (number of PDX Code 2 transverse cracks, total lineal feet of KDOT fatigue cracks) = -0.26

r (lineal feet of PDX Code 1 fatigue cracking, number of KDOT Code 1 transverse cracks) = -0.55, -0.33

r (lineal feet of PDX Code 1 fatigue cracking, number of KDOT Code 2 transverse cracks) = -0.31

r (lineal feet of PDX Code 1 fatigue cracking, total number of KDOT coded and uncoded transverse cracks) = -0.37

r (lineal feet of PDX Code 2 fatigue cracking, number of KDOT Code 2 transverse cracks) = -0.23

15

r (lineal feet of PDX Code 2 fatigue cracking, total number of KDOT coded and uncoded transverse cracks) = -0.24

r (PDX block cracking code, number of KDOT Code 2 transverse cracks) = 0.59, 0.47

r (PDX block cracking code, number of KDOT Code 3 transverse cracks) = 0.80, 0.48

r (PDX block cracking code, total number of KDOT coded and uncoded transverse cracks) = 0.38

Discussion of Data

The array of significant positive correlations between like distresses and severity codes suggests that the PA VEDEX data appear to be sensitive to all three severity codes of transverse cracking and to Code 1 fatigue cracking. Increases in PAVEDEX cracking occurrence data were associated with similar increases in KDOT data for both 152-m (500-ft) and 30-m (100-ft) pavement data sets. This finding is encouraging given the difficulty in interpreting video images using KDOT distress rating criteria.

The array of significant positive correlations between like distresses and different severity codes indicates that increases in PA VEDEX transverse and fatigue cracking occurrence data were associated with increases in KDOT data for both 152-m (500-ft) and 30-m (100-ft) pavement data sets, although the severity codes were not consistent. This suggests that the video image interpretation was sensitive to the general presence of transverse and fatigue cracking. Given the difficulty in converting the KDOT severity code criteria (based on perceived roughness and fragment spalling) into an image analysis format, the presence of positive linear associations between numbers of transverse cracks detected and fatigue cracking extent are encouraging, altho~gh the KDOT severity codes did not match.

The significant correlations between different distresses were both positive and negative for both 152-m (500-ft) and 30-m (100-ft) pavement data sets. The positive correlations between the PA VEDEX block cracking code and KDOT Code 2 and Code 3 transverse cracks may be caused by the video image analysis misinterpreting extensive interconnected secondary cracking associated with severe transverse cracking as block cracking.

The negative correlations indicate that as the PAVEDEX variable increases, the KDOT variable decreases and vice versa. The negative correlations between lineal feet of fatigue cracking and the number of transverse cracks suggest that extensive fatigue cracking does not occur simultaneously with transverse cracking, which seems to imply a mutual independence of distress types.

However, it may also indicate an interaction involving pavement age that can be explained in terms of causal factors; transverse cracking is an environmental distress caused by cold-weather temperature cycling, whereas fatigue cracking is traffic related. Younger pavements may experience enough temperature cycles for the development of Code 1 (hairline) transverse cracking but may not experience fatigue cracking. Hence, fatigue cracking would not be associated with transverse cracking. Older pavements certainly would experience

16

increased levels of fatigue cracking because of a larger number of axle loads; more severe transverse cracking would also be likely. Hence, for these pavements fatigue cracking would not be associated with the less severe Code 1 (hairline) transverse cracking. The combination of these two explanations could explain· the large number of negative correlations between transverse and fatigue cracking.

Direct Data Comparisons

Selected scatter diagrams comparing PA VEDEX and KDOT data for specific distresses and severities with statistically significant correlation coefficients were developed using both data sets. Lines of equality rather than simple linear regression models are shown on each figure to aid in directly comparing KDOT and PAVEDEX data.

Test Section Length of 152 m (500 ft)

Data pairs for Code 1 fatigue cracking are shown in Figure 5. Seven test sections did not exhibit fatigue cracking. PAVEDEX data for one test section indicated Code 1 fatigue ·cracking, although no fatigue cracking at that severity level was noted by KDOT engineering technicians. It does not necessarily mean that the PA VEDEX data indicated fatigue distress where none existed. The remaining eight" data pairs indicate that KDOT cracking data generally exceeded corresponding PAVEDEX data.

For Code 1 transverse. cracking, the trend illustrated in Figure 6indicates that the PAVEDEX number of occurrences is usually greater than the KDOT data. In single cases, the KDOT data did not indicate the presence of Code 1 transverse cracking, although the PAVEDEX data did, and vice versa. For Code 2 transverse cracking, as shown in Figure 7,

E <ii 

150

:§ 125 ci) c: 32

~ (.) 100

 :l Cl

~ u.

 "C 0 (.)

75

b ·so 0 ~

x w 0 ~ 25 <( 0...

0 50 100 150 200 250 KDOT Code 1 Fatigue Cracking, lineal m

FIGURE 5 KDOT Code 1 fatigue cracking measured by PA VEDEX versus KDOT Code 1 · fatigue cracking, 152-m test sections.

300


the general trend is reversed, with KDOT -data exceeding PAVEDEX data. For three test sections, KDOT data indicated the presence of Code 2 cracking, whereas PA VEDEX data were void of the distress at that severity level. One section exhibited the reverse.

These data indicate that the PA VEDEX video-based data detected the presence of fatigue cracking with perhaps some difficulty in assigning the correct severity, which is based on differences between nonspalled and spalled cracks. If the data shown in Figures 6 and 7 are taken together, it is apparent

40

Cl 35 c: 32 (.)

~ (.) 30

 ~ > (/) 25 c: I'll .= 20

"C 0

(.)

I-0 15 0 ~

x w 10 0 w > <( 0... 5

0 5 10 15 20 25 KDOT Code 1 Transverse Cracking

FIGURE 6 Number of KDOT Code 1 transverse cracks measured by PA VEDEX versus number of KDOT Code 1 transverse cracks, 152-m test sections.

Cl c: 32 25 (.) 113

0 ~ 20 > (/) c: ~ I-C\I 15 "C 0

(.)

I-0 0 10 ~

x w 0 w > 5 <( 0... 1 1

1/1 0 -1 1

30

0 5 10 15 20 25 30 KDOT Code 2 Transverse Cracking

FIGURE 7 Number of .KDOT Code 2 transverse cracks measured by PA VEDEX versus number of KDOT Code 2 transverse cracks, 152-m test sections.

Moore et al.

that although the PA VEDEX data were sensitive to the presence of transverse cracking,.the severity was generally underestimated. This is not surprising since the difference in KDOT severity criteria between Code 1 and Code 2 transverse cracking is a function of perceived roughness rather than a visual characteristic.


Figures 8 through 10 plot data using a 30-m (100-ft) pavement segment as the basis. Recall that within the five data pairs developed for each 152-m (500-ft) KDOT test section, the PA VEDEX data cannot be considered statistically independent and may also indicate a computational bias.

Figure 8 illustrates the PA VEDEX Code 1 transverse cracking data and the KDOT Code 1 transverse cracking data plotted with respect to a line of equality. Given the technique used to subdivide th~ PA VEDEX data into five pseudo 30-m (100-ft) segments, a single value of PA VEDEX data associated with several different KDOT values produced the obvious horizontal alignment within the data set plotted in the figure. Furthermore, it is apparent that the PA VEDEX data tend to overestimate the number of transverse cracks as compared with the KDOT crack mapping data.

Figure 9 indicates that the PA VEDEX data for Code 2 transverse cracking tends to underestimate the KDOT Code 2 transverse cracking data. The "L" plotted at (0,0) signifies 21 data points.

These two figures taken together suggest that the PA VEDEX video analysis classified many KDOT Code 2 transverse cracks as Code 1 cracks. This classification would overestimate the number of .Code 1 cracks and underestimate the number of Code 2 cracks. These trends are similar to those observed in Figures 6 and 7. The potential effect of computational bias

10

9 Cl c: :i: (.) 8 ~ () Q)

7 en Qj > en c: 6 ~ Q) 5 -g ()

lo Cl :::.::: x w Cl w > <( a..

4

4

0 2 3 4 5 6 7 8 9 KDOT Code 1 Transverse Cracking

FIGURE 8 Number of KDOT Code 1 transverse cracks measured by PAVEDEX versus number of KDOT Code 1 transverse cracks, 30-m test sections.

17

(i.e., overestimation of the number of transverse cracks), if it exists, is not apparent. Although the correlation for Code 3 cracking was statistically significant; not enough Code 3 data were obtained from the test sections for a meaningful discussion of a scatter diagram.

Figure 10 illustrates the relationship between the PA VEDEX Code 1 fatigue cracking data and KDOT Code 1 fatigue cracking data. The asterisk plotted at (0,0) indicates over 35 points. These data clearly indicate that the PA VEDEX data underestimate the lineal feet of Code 1 fatigue cracking as measured by KDOT field personnel. Furthermore, seven of the data

E ca Q)

,§

7 Cl c: :i: (.) 6 ra 0 Q)

~ 5 Q)

> en c: ra ~ 4 C\I Q)

"O 0 () 3 I-0 Cl :::.::: 2 x w Cl 3 w > 15/~ <( 2 a..

0 i- ~ 4 l l 4 2 2 2

0 2 3 4 5 6 7 KDOT Code 2 Transverse Cracking

FIGURE 9 Number of KDOT Code 2 transverse cracks measured by PA VEDEX versus number of KDOT Code 2 transverse cracks, 30-m test sections.

30

25 C> c: :i: (.) ra 0 20 Q) ::J Cl

~ LL

15 Q)

"O 0 ()

I-0 10 Cl :::.::: x w Cl w 5 > <( a..

5 ll l l

11. l 0 l

0 10 20 30 40 50 KDOT Code 1 Fatigue Cracking, lineal m

FIGURE 10 KDOT Code 1 fatigue cracking measured by PA VEDEX versus KDOT Code 1 fatigue cracking, 30-m test sections.

8

2 3

60

18

points that fall above the line of equality indicate the presence of Code 1 fatigue cracking in the PAVEDEX data but not in the KDOT data. This does not necessarily mean that PA VEDEX detected nonexistent fatigue cracking, but that PA VEDEX assigned a higher severity code to the distress than KDOT. This would not be unexpected since the difference between Code 1 and Code 2 fatigue cracks relates to the presence of spalling, and video image interpretation was inconsistent in assigning the correct distress severity.

Crack Analysis-Significance of Pavement Type

A one-way ANOV A was p~rformed using pavement type as the independent variable. The response variables were the differences between compatible PAVEDEX and KDOT data related to fatigue and transverse cracking.


None of the ANOV As indicated that pavement type was a significant factor. Therefore, pavement type (COMP, PDBIT, and FDBIT) did not influence the relative precision of the PA VEDEX unit in identifying the number of specifically coded (by severity) transverse cracks or lineal feet of specifically coded (by severity) fatigue cracking in both wheelpaths.


Pavement type was a significant factor affecting the differences between KDOT and PA VEDEX Code 1 transverse cracking, Code 1 fatigue cracking, and Code 2 fatigue cracking measurements. Since the null hypothesis is an equality of means (differences in this case) for each pavement type, the mean differences for the significant distresses were as follows:

Distress

Code 1 TC Code 1 FC Code 2 FC

COMP

2.50 1.20 0.00

PD BIT

0.30 -58.26

25.68

FD BIT

0.56 -15.08 -3.80

The transverse cracking data are numbers of cracks; the fatigue cracking data are in lineal feet. The differences were developed by subtracting KDOT data from PA VEDEX data. Positive quantities mean that KDOT data were larger than PA VEDEX data and vice versa. These data illustrate the magnitude of variation in the mean differences over the three pavement types and the basis for the statistical significance of pavement type in the analysis of variance. It is not clear whether pavement type or some other unknown concomitant


factor, such as the computational technique used to create PAVEDEX data for 30-m (100-ft) segments, caused this variation in the differences between the PA VEDEX and KDOT data. No apparent physical reason associated with pavement type was evident from examination of the KDOT crack map data. However, these findings may suggest that subdivision of the PAVEDEX data using 0.16-km (0.1-mi) pavement segment length into smaller units should be done with caution.

CONCLUSIONS

1. The average maximum rut depth measured by the IMS laser system provided a relatively precise estimate of rut depth severity.

2. ·The differences between IMS rutting data and KDOT field measurements were not affected by KDOT PMS pavement type (COMP, FDBIT, PDBIT).

3. Only one correlation between IMS cracking data and KDOT cracking distress measurements was significant. Given the comprehensive array of IMS crack width and depth measurements, the absence of linear association with the field data was unexpected.

4. PAVEDEX video data appeared to be sensitive to the presence of transverse and fatigue cracking, as suggested by numerous statistically significant correlations with KDOT field measurements. However, the video interpretation had difficulty in assigning the correct severity code because the KDOT PMS distress rating system uses perceived roughness associated with transverse cracking and spalling associated with fatigue cracking as criteria. This may also indicate a technician training deficiency.

5. The PA VEDEX video data interpretation also appeared to classify severe transverse cracking with secondary distress as block cracking. This could also be.a training deficiency.

6. The differences between PA VEDEX cracking data and KDOT field measurements were not affected by KDOT PMS pavement type (COMP, FDBIT, PDBIT) if analyzed on the basis of a 152-m (500-ft) pavement segment length, but KDOT pavement type was significant if the data were analyzed on the basis of a 30-m (100-ft) pavement segment length.

ACKNOWLEDGMENT

The research reported herein was conducted under the auspices of an FHWA contract, entitled Evaluation of Selected Devices for Measuring Pavement Distress.

Publication of this paper sponsored by Committee on Pavement Monitoring, Evaluation, and Data Storage.

Obtaining Rut Depth Information for Strategic Highway Research Program Long .. Term Pavement Performance Sites

WADEL. GRAMLING, GEORGES. SUZUKI, JOHN E. HUNT, AND

l<AZUHIKO MURAOKA

The use of automated pavement condition survey systems has been a goal of highway managers for many years. With the advent of the Strategic Highway Research Program's (SHRP's) LongTerm Pavement Performance study the need for permanent, highresolution pavement distress records arose. To meet this need through the use of state-of-the-art technology, SHRP chose to use automated RoadRecon survey systems to obtain permanent high-resolution records of pavement surface distress and transverse profile. The survey system and procedures used in the field to record the transverse profiles on 35-mm film, the procedures used in the office to process the films, the methods used to analyze the transverse profiles to report the rut depth data, and the quality-control procedures employed are all discussed. The data for the first three rounds of measurement on many of the sites are available. The results of some preliminary analysis of these data are also described.

The Strategic Highway Research Program (SHRP) announced the initial request for proposals for the Long-Term Pavement Performance (LTPP) program in 1987. Included in the request was a project to collect pavement surface distress as a major component of the monitoring of selected sections of pavement on in-service highways. The data were required to be obtained in a rigorously consistent fashion _and were to be of an accuracy and precision required by the needs of the LTPP studies.

The objective of the pavement distress data collection work was to provide high-resolution visual records of the pavement condition and high-quality measurements of the rut depth of each wheel track in the outside lane of each L TPP test section.

A contract was awarded in the fall of 1987 to PASCO USA for the collection of pavement distress data. New RoadRecon survey units, equipped with RR-70 and RR-75 survey systems, were designed and constructed to perform the data collection. The RR-70 and RR-75 survey systems are used to collect surface distress and rut depth data, respectively. On completion of construction, the performance of the units was thoroughly evaluated and SHRP L TPP operational guidelines were established. This effort is described in another paper (Gramling et al., unpublished data, July 1992).

On completion of the construction and evaluation of the RoadRecon units, routine survey operations began on March 8, 1989, near Sedalia; Missouri.

PASCO USA, Inc., 4913 Gettysburg Road, Mechanicsburg, Pa. 17055.

This paper describes the survey methods used in the field to obtain transverse profiles and the procedures used in the office to process, analyze, and report the rut depth data. The initial three rounds of data for most of the SHRP L TPP sites are available and the contents are described.

RR-75 SURVEY SYSTEM

The RoadRecon-75 (RR-75) survey system consists of a 35-mm pulse camera synchronized with a strobe projector. The pulse camera is mounted on a boom that extends from the rear of the survey vehicle with its lens pointed directly down and perpendicular to the pavement and focused on the surface. The projector is mounted on the center of the rear bumper at a set distance above the roadway. A glass plate with a hairline etched in its surface is positioned in front of the projector lens so that when the strobe is triggered a shadow of the hairline is projected onto the pavement directly under the center of the camera. The position of the camera and the angle of the projector are carefully maintained. The projected hairline shadow provides a thin, sharp-edged image on the film caused by the intensity of the projector flash.

In operation, the RR-75 pulse camera is triggered at selected intervals to photograph the projected hairline. The 35-mm film frame exposure then contains the hairline image across the pavement lane for a width of 4.7 m (15.5 ft) accurately reflecting the profile of the pavement.

Nightly Survey Setup Procedures

To obtain high-quality film exposure and resolution, which are controlled by artificial illumination and the angle of lighting, survey operations are performed only during nighttime hours. Before beginning operations each night, quality control procedures are followed by the survey crew to ensure the safe operation of the vehicle and to verify the condition of the survey system setup. Detailed checklists are completed by the crew to ensure that all required items are inspected. An illumination check is run on the front lighting arrangement, which is part of the RR-70 system, and a set of calibration blocks is used to verify that the RR-75 system is in proper position. The tests are run with the vehicle static at a convenient level parking area before starting the night's production survey.

20

Survey Operations

The SHRP L TPP program includes a large number of carefully marked and documented "in-service" pavement sections across the United States and Canada. There are General Pavement Study (GPS) sections that represent the most common typical pavement types found and the Special Pavement Study (SPS) sections that are fewer in number and are selected to investigate specific pavement questions. All of the GPS and SPS sections have been selected to fit into a preplanned experimental matrix.

SHRP LTPP GPS sections are 150 m (500 ft) in length with a lead-in of 150 m (500 ft) and ending marks at 76 m (250 ft). The sections are marked with white paint on the roadway surface. Blue reflectors and signs denoting the upcoming section have been placed along the right shoulder. Figure 1 shows a typical GPS test section layout (1).

Other SHRP functions have been directed at formally establishing, documenting, and marking SHRP sites for inclusion in the research. The current list of approved sites is used to develop a survey routing that requires the least amount of travel time to systematically move from site to site and cover logical geographic parts of the United States and Canada. Surveys are done usually in the northern areas during the summer, in the middle states during the autumn and spring, and in the southern and west coast states during the winter.

In survey operations the crew goes through the setup procedure before approaching the site to be surveyed. When the survey crew approaches the site the systems are in operating

White paint stripe. 250" past and of Monitoring Sita

Oelineator located at and of Monitoring Sita with 2 blue refectors

Monitoring { Seel ion

Delineator located al begining of Monitoring Sita with 3 blue refectors

White paint stripe

!---+

+

t

+

! Sign B indicating start

: of monitoring site \ : ~arallal with fence . , line

:ei

.; 1 \__Spika or nail

Sign A located soo· in advance of Test Section

FIGURE 1 General layout of test section showing sign locations.


position and are ready to be activated to begin filming. Each 150-m (500-ft) test section has a 150-m (500-ft) lead-in, which is the first mark encountered; the test section is also marked at each 30.5-m (100-ft) station. When the crew sights the first mark the systems are activated and filming begins. When the vehicle reaches a point in advance of the section start mark, the operator resets the systems to take the first transverse profile shot at the begin mark. After the reset, the RR-75 system automatically records a frame at the beginning and at each subsequent 15.25-m (50-ft) interval. The survey continues through the 76-m (250-ft) runout marked with a paint stripe following the test section. This results in a continuous RR-70 film record of the 376-m (1,250-ft) length marked and a series of RR-75 transverse profiles, which include the 11 profiles at 15.25-m (50-ft) intervals required through the actual test section.

The crew makes visual observations of the RR-75 filming operation during the survey run by watching for, and counting, projector flashes to verify that the profile shots are on the targets and that there are 11 sequential shots taken within the test section during the survey run.

Because the test section survey costs are relatively minor compared with the travel costs to get to the section, each test section is filmed at least twice to provide a backup record. If one of the two runs is questionable, additional runs are made until the crew is satisfied that they have observed two good runs. On completing a section survey the crew proceeds to the next section on the schedule and the survey process is repeated.

SHRP L TPP SPS sites are much fewer in number than the GPS sites. However, each site may contain 10 or more 150-m (500-ft) test sections. Occasionally some of the SPS sections are longer than the 150-m (500-ft) GPS standard. Usually there is a GPS section associated with each SPS site, and there may be a number of supplemental sections that have been added by the state or province.

The distress survey procedures used at the SPS sites are the same as those used for GPS sites as far as possible. At some SPS sites the spacing between the various sections does not permit resetting the RR-75 system. For those particular sections involved, the 15.25-m (50-ft) interval for profiles is maintained, but the quality-control requirement for locating the transverse profile within a few meters of the 30.5-m station marks is waived.

When an SPS site that includes an associated GPS section is surveyed, separate surveys are performed. The GPS section is surveyed using normal GPS procedures and then another survey is performed to record the SPS sections, including the GPS section. Any agency supplemental SPS sections are surveyed and processed in the same manner as that for L TPP sections.

Periodic Checks

At the end of each week's survey operations, or after 20 sites have been surveyed, the crew performs a quality-control check. The quality-control check is done by the crew placing a standard resolution board on the pavement and making an additional survey run. The unit is also stopped at a level parking area and the calibration blocks are filmed by manually acti-

Gramling et al.

vating the strobe projector. The exposed film containing the routine survey runs, the resolution board, and the calibration blocks is then shipped to headquarters for processing and analysis.

Office Operations

Film Processing

On receipt of the exposed film at headquarters the film is developed using an automatic film developer and then viewed by a technician at a film processing work station. The work stations consist of a 1.2-m ( 48-in.) light table and film winders. The negative RR-70 film is reviewed for quality by checking longitudinal distortion, lateral lane placement, and exposure. The RR-75 film is checked to locate longitudinal profile locations and to ensure that 11 frames were recorded within the test section without skips. The profile location is identified using conventional stationing, with the first station coinciding with the begin mark of the 150-m (500-ft) test section. The paint marks are usually visible in the RR-75 frames, enabling an accurate offset station to be assigned to each profile hairline. If the offsets are more than + 1.2 m (4 ft) to -0.6 m ( -2 ft) the profile hairline placement is identified by using event marks found on the RR-70 film that are automatically placed when the RR-75 system shots are triggered. The event marks are related to the section paint marks to determine. the offset of the profiles and to establish the stationing.

After the RR-75 film has been checked for quality and completeness, and profile stations have been assigned, the film is edited by trimming and splicing, labeled, and moved to an RR-75 work station for digitizing.

Digitizing

The digitizing work station combines a film motion analyzer (FMA), a personal computer, and custom software to record transverse. profile information and compute rut depth values from the RR-75 films. The RR-75 film is mounted on the FMA film transport, which back-projects the profile hairline image on a digitizing screen. The operator enters the section identification (ID) information and proceeds to digitize each film frame profile into the computer data base using a cursor to input coordinates.

The operator begins the digitizing operation by identifying the first point to be digitized. This point is located along the shoulder edge of the pavement and is identified by following a _set hierarchy of criteria.

For concrete pavements, the operator identifies the laneshoulder joint and then digitizes the first point on the pavement lane's surface adjacent to the joint. If this is not visible, the first point is digitized on the pavement's surface adjacent to the outside edge of the pavement's edgeline.

For bituminous surfaced pavements, the operator first checks to see whether a lane-shoulder joint is present. If so, then the first point is digitized on the pavement lane's surface adjacent to the joint. If a joint is not readily apparent, then the operator's second check is for an obvious difference in the surface texture between the pavement lane and the shoul-

21

der. If this difference exists, then the operator digitizes the first point on the lane's surface adjacent to the location of the texture change. If a texture difference cannot be found, then the operator's third check is the pavement's edge line, and the first point is digitized on the pavement's surface adjacent to the outside edge of the edge line.

Once the first digitizing point has been identified and digitized, the operator digitizes between 24 and 29 additional points along the hairline image at approximately 15-cm (6-in.) intervals across the pavement's lane.

After the required profile points are entered, the computer calculates the rut depths and displays the profile for the operator's review and acceptance. After the section's last profile is entered the operator can view all 11 profiles overlaid to judge if any anomalies might be obvious. If the operator is satisfied, the file is accepted and moved to quality assurance.

Quality-Assurance Procedures

The quality-assurance procedures for the profile data involve checking the identifying header information against the log information used to track individual data records for each section scheduled and surveyed in the field. An historical check is also made after the first survey is recorded for a section. Each subsequent round of survey information for a section is checked to verify that the latest rut depth values are logical and reasonable when compared with prior sets of survey data. If anomalies are found, the prior RR-75 film records are available if needed to verify data.

After the sections pass quality-assurance reviews, an ASCII data file, a section summary report, and a set of transverse profile plots are generated for each test section. These records are forwarded for SHRP entry into the TRB IMS data base.

Data Format

The data output from digitizing the profiles from the RR-75 frames is provided in a set of 11 profile plots that contain the section ID, the date surveyed, and the station of the profile. The profile plot graphically shows a standard reference plane established by the lane edges, the profile, and the high points of the profile determined by the wire method of analysis (2). The maximum values for rut depths are given for the center and shoulder half of the lane, and the locations of the values are plotted. Figure 2 shows a standard profile plot with the center of the lane higher than the baseline, and Figure 3 shows a standard profile with the center lower than the baseline.

A summary data sheet that contains a tabulation of the stations for each profile and the maximum rut depths for each half lane is also supplied. The 11 sets of data are summarized with an average rut depth for each half lane, the maximum and minimum values are shown, and the standard deviation for the rut depths through the section is given. Figure 4 shows a typical summary sheet.

An ASCII data file is also supplied containing the data along with positive copies of the edited RR-75 film. The original negative film is packaged for placement in the TRB L TPP archives and could be used in the future to perform any additional analysis desired.

22

(1111)

70

60

50

40

30

20

iO

0

-iO

-20

-30

-40

-50

-60

SECTION i2i370 DATE SURVEYED : 02/05/9i DATE DIBITIZED : 05/06/9i

iO

(Canter)

STATION UNIT OPERATOR

2

u

2+99 H CAMERA: H CRONEL

3 (M)

(Shoulder)

FIGURE 2 Standard profile in which center of lane is higher than baseline.

(1111)

70

60

50

40

30

20

iO

0

-10

-20

-30

-40

-50

-60

SECTION i2i370 DATE SURVEYED : 02/05/9i DATE DIBITIZED : 05/08/91

i4

(Canter)

STATION UNIT OPERATOR

2

i3

0+50 H CAMERA: H CRONEL

3 (Shoulder) !Ml

FIGURE 3 Standard profile in which center of lane is lower than baseline.

Results to Date

The initial SHRP L TPP efforts to collect distress information, including rut depths, have been completed with three rounds of survey data collected for a large majority of the GPS sections. The following table gives the distribution by region of the 763 GPS sections surveyed:

Region

North Atlantic Southern North Central Western

Number of Sites

134 258 199 172

The same procedures were used on both portland cement concrete and asphaltic concrete sections filming both RR-70


*** LTPP Section ID 121370 ***

Date surveyed 02/05/91 Unit fl camera fl

operator Date Diqitized

No.

01 02 03 04 OS 06 07 08 09 10 11

Station

o+oo o+so 1+00 1+50 1+99 2+49 2+99 3+49 3+98 4+48 4+98

CROWEL 05/06/91

Rut Depth(mm)

Leftside

11 14 10 10

9 16 10

8 8 9 7

Rightside

13: 13 11

9 8 8

11 10

9 10

8

--------------------- summary -------------------------Maximum Minimum Averaqe SD

16 7

10.2 2.7

FIGURE 4 Sample summary report.

13 8

10.0 1.8

for continuous distress records and RR-75 for rut depths. Table 1 shows the pavement types included in the total GPS experiment. Table 2 contains the average rut depths by region for each of the GPS pavement types. Table 3 contains the average of the standard deviations for each of the GPS pavement types by region.

SUMMARY

The initial 5 years of the SHRP LTPP program has resulted in the development of equipment, survey methods, data analysis procedures, and standards of quality for the automated collection of pavement distress and rut depth determination. The survey procedure produces a permanent 35-mm film record.

Three rounds of rut depth survey data have been collected for most of the GPS sections, and the data have been processed and supplied to the TRB IMS data base. Similarly, two rounds of rut depth data are available on most of the SPS sites.

A review of the average rut depth data would lead to several observations about the SHRP GPS sites. There is not much difference in average rut depths between the center half lane and the shoulder half lane. The average rut depths for all pavement types and regions are well below Vi in. and would be considered acceptable by most pavement condition criteria. There are differences between pavement types and between regions. There are also GPS sections with rut depths well above Vi in., but they are found throughout the pavement types and across the regions. Some agencies have a greater number of GPS sites with larger rut depths than many of the other agencies. The significance of this observation is beyond the scope of this paper.

These data also indicate that the average rutting on concrete pavements was found to be less than 4 mm for the GPS pavement types including jointed plain, jointed reinforced, con-

TABLE 1 GPS Experiment Pavement Types

Pavement Type Description

1 2 3 4 5 6A 6B 7A 7B 9

Asphalt on Granular Base Asphalt on Bound Base Jointed Plain Concrete Jointed Reinforced Concrete Continuously Reinforced Concrete Existing Asphalt Overlay of Asphalt Planned Asphalt Overlay of Asphalt Existing Asphalt Overlay of Concrete Planned Asphalt Overlay of Concrete Unbonded Concrete Overlay of Concrete

TABLE 2 Average Rut Depths by Region

Regions North Southern North Western

I Atlantic Central Pavt. Types C-AVG S-AVG C-AVG S-AVG C-AVG S-AVG S-AVG C-AVG

1 9.0 8.7 6.0 8.3 6.1 6.9 7.0 8.9

2 5.3 5.7 4.5 6.4 9.7 9.9 5.3 5.5

3 2.6 3.4 2.1 3. 2 2.5 3.0 3.6 4.0

4 3.2 3.4 1. 7 2.6 2. 3 2.6 --- ---

5 2.8 3.4 2.1 3.2 2.8 3.5 3.8 4.0

6A 12.5 13.3 5.1 6.5 6.9 7.2 6.9 6.9

6B 6.6 5.3 5.7 6.4 3.7 2.8 4.9 5.7

7A 7.2 6.0 8.3 9.0 4.3 5.7 11.1 10.7

7B 3.9 3.6 1. 3 2.5 2.7 2.9 --- ---

9 2.1 2.3 1. 9 2.3 2.5 3.2 3.5 4.6

Note: Depths given in millimeters C-AVG The average rut depths of the center side. S-AVG = The average rut depths of the shoulder side.

TABLE 3 Average GPS Standard Deviation of RUT Depths by Region

Regions North Southern North Western

I Atlantic Central Pavt. Types C-STD S-STD C-STD S-STD C-STD S-STD S-STD C-STD

1 1. 8 2.1 1. 6 2.2 1. 8 1. 9 1.8 1. 9

2 1. 2 1. 2 1. 6 1.8 1. 7 1. 9 1. 5 1. 7

3 1. 0 1. 4 1.1 1. 4 0.8 1. 0 1.1 1. 0

4 1.1 1. 2 0.7 1.1 0.8 0.9 --- ---5 0.9 1. 4 0.8 1. 2 0.8 1. 0 1.1 1. 0

6A 2.6 2.3 1. 6 1. 6 1. 3 1. 3 1. 4 1. 6

6B 1. 7 1. 7 1. 5 1. 7 1. 3 0.9 1. 4 1. 5

7A 1. 4 1. 0 1. 7 1.9 1. 5 1. 9 2.3 1. 9

7B 1. 4 1. 0 0.5 1. 0 1. 5 1. 9 --- ---

9 0.8 1.1 0.7 0.8 1. 0 1.1 1. 1 1. 4

Note: Depths given in millimeters. C-STD The average standard deviation of the rut depths

of the center side. S-STD The average standard deviation of the rut depths

of the shoulder side.

24

tinuously reinforced, and concrete overlays. Isolated GPS sections were surveyed that had greater rut depths, but these were scattered.

The averages of the standard deviations by region and pavement type generally are less than 2 mm. This would indicate that the rutting is found to be fairly uniform through each of the 500 GPS sections. When the generally lower average rut depths of concrete pavements are considered with the standard deviation values it appears that most of the rut depths might be associated with construction procedures producing the transverse cross-section variables observed as rutting. A closer inspection and analysis of the data within the GPS sections, including the transverse location of maximum rut depth in each half lane and the transverse profiles, will provide insight into these data.

The quality-assurance procedures and consistency of data would indicate that the .intended accuracy and precision required in obtaining the rut depth data have been achieved.


ACKNOWLEDGMENT

The work described in this paper was performed as part of the LTPP study of SHRP, National Research Council.

REFERENCES

1. Guidelines for Signing and Marking of General Pavement Studies' (GPS) Test Sections. Operational Memorandum SHRP-LTPP-OM-002. SHRP, National Research Council, Highway Research Program, Washington, D.C., July 1988.

2. Gramling, W. L., J.E. Hunt, and G. S. Suzuki. Rational Approach to Cross-Profile and Rut Depth Analysis. In Transportation Research Record 1311, TRB, National Research Council, Washington, D.C., 1991, pp. 173-180.



Adapting an Automated Data Collection Device for Use at an Airport

MARGARET BROTEN, GEORGE SCHWANDT, AND RUDY BLANCO

O'Hare International Airport, in Chicago, Illinois, is one of the busiest and largest airports in the world. It is critically important to maintain all the O'Hare airside and landside pavements in operational and safe condition. Accurate and current distress data are needed to document the present condition of the pavement, to determine maintenance and rehabilitation needs, and to form t~e .basis of a pavement management system. However, it is very difficult to collect pavement condition information safely and accurately in this extremely congested environment. A demonstration project was conducted on one runway and one parallel taxiway to determine the feasibility of using automated data collection equipment for distress data collection. The demonstration project was successful, and a full-scale data collection effort was undertaken to film the remaining runways and parallel taxiways. This approach allowed the data to be collected between 11 :00 p.m. and 5:00 a.m. In addition, data collection was fast and efficient. Modifications to the procedures and equipment used to collect distress information on roads had to be made to use this equipment successfully on airfield pavements. A method for maintaining a straight pass down the runway had to be devised. A lighting system had to be developed because the data collection occurred during the nighttime hours. Finally, data interpretation was modified to obtain an acceptable estimate of a pavement condition index.

O'Hare International Airport is one of the busiest facilities in the world, and it provides a critical link in the United States' air transportation system. Figure 1 shows the layout of the airside pavements at O'Hare. It is very important to keep the pavements at this airport in safe and operational condition. In addition, because any pavement closure is extremely costly, it is important to apply timely maintenance and rehabilitation to the pavements to red~ce any required closures.

Accurate and current distress information is needed to effectively manage the pavements at O'Hare International Airport. Distress data are used to document current condition to identify maintenance and rehabilitation needs, to estimat~ repair quantities, and to provide the basis for a pavement management system (PMS). However, it is difficult to collect pavement distress data safely and efficiently, without dis-rupting operations, at a facility like O'Hare. ·

Because of heavy traffic conditions during the daytime hours at O'Hare, the only available time for distress data collection on the runways and taxiways is between 11:00 p.m. and 5:00 a.m. It was estimated that it would take several months to manually collect distress data on all the runways and taxiways

M. Broten, ERES Consultants, Inc., 439 Edgewood Drive, Ambler, Pa. 19002. G. Schwandt, City of Chicago, 20 North Clark, Suite 2900, Chicago, Ill. 60602. R. Blanco, PaveTech, Inc., P.O. Box 639, Norman, Okla. 73070.

under these conditions, which would be a hardship on the operations staff at O'Hare and would be a difficult task for a survey crew.

This situation led to the conduct of the demonstration project and full-scale implementation project discussed in this paper. The city of Chicago contracted with ERES Consultants, Inc. (ERES), to conduct a pavement evaluation of all airside and landside pavements at O'Hare and to implement a PMS for O'Hare. An automated data collection device was evaluated for its feasibility to collect distress data on runways and taxiways at O'Hare during a demonstration project. The successful completion of this demonstration project led to the evaluation of all runways and parallel taxiways at O'Hare using this equipment.

DEMONSTRATION PROJECT

A PaveTech video inspection vehicle (VIV) was used to collect distress data on a runway and its parallel taxiway for the purpose of evaluating the feasibility of using this type of equipment at O'Hare. There are several other automated data collection devices on the market today. However, it was determined that this equipment best fit O'Hare's requirements when the contract was signed.

Pavement Condition Index Procedure

Pavement management decisions are dependent on some method of evaluating pavement condition to document current pavement condition, predict future performance, priority rank needs, and determine required repair levels. The pavement condition index (PCI) procedure, outlined in the F AA's advisory circular (AC) 150/5380-6, entitled Guidelines and Procedures for Maintenance of Airport Pavements, is the standard used by the aviation industry to assess current pavement condition (1). The PCI, developed by the U.S. Army Corps of Engineers, is a numerical rating ranging from 0 to 100, with 100 being a pavement in new condition. The PCI provides an objective and repeatable indication of the condition of the pavement. A manual is used to identify individual distress types in asphalt and concrete pavements and to guide the determination of severity level and quantity of each distress present.

The first step in the procedure is to divide the airport pavements into branches, features (sections), and sample units. A branch is defined as an entity that serves a single function, such as a runway, a taxiway, or an apron. Sections are a

26

portion of the branch with consistent characteristics throughout their entire area or length. The area within a section should have been constructed at the same time, should receive approximately the same type and level of traffic, should have received the same type of rehabilitation and maintenance throughout the years since original construction, and should be a size that can be economically managed as a single entity; An example of a runway at O'Hare, which has been broken into sections, is shown in Figure 2.

Sections are further subdivided into sample units for the purpose of inspection. Guidelines have been established for

I "' FIGURE 1 O'Hare International Airport layout map.

... SICTION 1 !

+ 8

1C

19

FIGURE 2 Example of section identification map.


determining the appropriate size of sample units. As defined in FAA AC 150/5380-6, sample units for asphalt-surfaced pavements are approximately 5,000 ft2 (1,524 m2), and sample units for concrete pavements are approximately 20 slabs. Figure 3 illustrates runway sections that have been subdivided into sample units.

Once the network definition has been completed, the PCI inspection can be conducted. The measured data for distress type, severity, and quantity are used to calculate a composite index, the PCI. The procedure to perform this calculation is outlined in the FAA AC and is shown in Figure 4.

Equipment Description

The VIV uses video cameras to collect super-VHS video images of 100 percent of the pavement surface. The high-detail resolution video distress images were used for distress identification. Pavement profile, roughness, rutting, faulting, and texture information are collected automatically by the equipment.

The pavement profile is collected using a South Dakotatype profiler that operates using the inertial reference concept. As the vehicle travels along the pavement, an accelerometer mounted on the vehicle senses acceleration forces and generates an electrical signal proportional to the vehicle's vertical accelerations. The signal is filtered, digitized numerically, and doubly integrated to calculate the vehicle's up-anddown movement and to establish the position of the vehicle as a function of time. A distance-measuring sensor, in this case an ultrasonic transducer, measures the distance between the accelerometer and the pavement surface by emitting a short burst of high-frequency sound toward the pavement surface. The sound waves strike the roadway, reflect upward,

SBCl'ION 2 SECl'ION S

ac SC

2S SS

4l t ! i g SBCnON 4

Broten et al.

and are detected by the same transducer. The time between the sound generation and echo detection is proportional to the distance to the roadway surface. The roadway profile is then computed as the difference between displacement and the distance between the vehicle and the road surface.

Rutting is measured using the results from five ultrasonic transducers mounted in the vehicle bumper. They are collinear and equidistant from each other. At every foot these sensors measure the distance to the pavement surface.

The automatic faulting measurement device uses two sensors that measure the distance from the moving vehicle to the ground. The sensors are triggered simultaneously to eliminate any negative impact on the readings as a result of the up-anddown movement of the ve~icle. The sensors are located one behind the other close to the center of the vehicle to minimize any impact of the pitch movement of the vehicle. Once the sensors are simultaneously triggered, a computer reads the distance from the vehicle to the ground for each sensor. The computer compares these readings and calculates the difference between each sensor. The differences are then stored in the computer. The differences in height along the pavement become the raw data used to calculate faulting measurements at concrete slab joints.

1N

27

Preparation and Conduct of Demonstration Project

The automated equipment was originally developed to collect road condition data. It was designed to collect information during daylight, and a single pass of the equipment videotapes a width up to 13.8 ft (4.2 m). For use on the airside pavements at O'Hare, the equipment had to be adapted to operate under nighttime conditions on facilities that were up to 200 ft (61 m) wide.

A lighting system was added to allow data collection during nighttime for this airport application. The lighting system consisted of a modified trailer that rides behind the VIV and provided the lighting needed for the back distress cameras. The trailer contained a gasoline-powered generator, the lights, and an extendible and retractable modified "goose neck tongue." The goose neck was modified to prevent it from being videotaped.

Because of the width limitations of the filming equipment, the sample unit layouts were different from what they would have been for a manual inspection. Using conventional methods, a section that was 50 ft (15 m) wide would have sample units with dimensions of 50 ft (15 m) wide by 100 ft (30 m) long. However, because the equipment was set to videotape

1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41

2 4 8 8

1 3 5 7

2 4 e 8

1 3 5 7

2 4 e 8

W~M~~~~UU~OO~M~~~~

1C

9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41

W~M~~~~UU~OO~M~~~~

1S

9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41

W~M~~~~UU~OO~M~~~~

' Ill\\ NORTH

FIGURE 3 Example of sample unit identification map.

28

passes 12.5 ft (3.8 m) wide, this type of layout would have been inconvenient. Instead, sample units on asphalt-surfaced pavements were 12.5 ft (3.8 m) wide by 400 ft (122 m) long, which allowed a single video pass to be viewed to inspect a sample unit. Sample units for concrete pavements were 1 slab wide by 20 slabs long. Two filming passes had to be combined to inspect a concrete sample unit.

There was concern that at night it would be difficult to maintain a straight pass down the length of the runway. It is important to maintain a straight pass, because 7 to 12 (depending on the width of the facility) longitudinal passes are needed to provide complete coverage: The runway and parallel taxiway that O'Hare Operations identified as the most convenient for the demonstration project were painted at 200-ft (60-m) intervals. Each filming lane was marked, and the paint marks were staggered so that the effective spacing was 100 ft (30 m). These paint marks were to be used by the vehicle driver to maintain a straight line as each lane was filmed.

In addition, sample units were identified; these were to be inspected manually for the purposes of comparing the manual results with the videotape results. Unfortunately, because of


the unforeseen scheduling of a rubber removal project on the demonstration runway, the marked runway and taxiway were not available to the crews during the scheduled demonstration period. The operations staff was able to obtain closure for the crews on a different runway and parallel taxiway. However, these facilities were not premarked in any way.

Because of the change in schedule, the first night of filming was a test run. The driver attempted to maintain straight passes visually. The parallel taxiway was filmed, and a few passes on the runway were made. A review of this film revealed that not all the passes were straight, and in addition, light rainfall had resulted in glare occurring on the film.

The crew was adaptable and devised an alternative way of maintaining straight passes. They set up traffic cones before filming to aid the driver. Orange cones with reflective tape were used to mark the lanes. One cone approximately every 200 ft ( 60 m) proved to be sufficient to guide the driver. This approach worked so well that is was adapted for the full-scale implementation later.

The use of cones was much more cost-effective than prepainting the facilities. Painting the originally identified demonstration runway and taxiway took a crew of three people

STEP 1. DIVIDE PAVEMENT FEAWRE INTO SAMPLE UNITS.

STEP 2. INSPECT SAMPLE UNITS: DETERMINE DISTRESS TYPES AND SEVERITY LEVELS AND MEASURE DENSITY.

ASPHALT CONCRETE

LIGHT LONGl'NDIN~~SE~ ~ STEP 8. DETERMINE

EDIUM MEDIUM CORNER PAVEMENT AWGATOR • • · SPAWNG CONDITION

STEP J. DETERMINE DEDUCT VALUES. RATING OF FEATURE.

wc&j~) ~ t :::::> Iii -Q I

O O. 1 DENSITY " 100 (Log Scale)

111~GATOR AC ~ b

~ 0

o. 1 DENSITY " 100 (Log Scale}

STEP 4. COMPUTE TOTAL DEDUCT VALUE (lDV) a+b.

STEP 5. ADJUST TOT AL DEDUCT VALUE.

e~100 ~, 3 t> cov ~t; -- q • ~ber o!t~ 8a owr ~lnta

Q OO TDV-a+b 100 200 TOTAL D£DUCT VALUE

Excellent

85 Very Good

To Good

55

40 Poor

25 Very Poor

To Foiled 0

STEP 6. COMPUTE PAVEMENT CONDITION INDEX (PCI) = 100 - CDV FOR EACH SAMPLE UNIT INSPECTED.

STEP 7. COMPUTE MEAN PCI OF ENTIRE SECTION FROM PCls OF SAMPLE UNITS.

NOlE: FOR DETAll.£0 PROCEDURE SEE •cONOl110N SURVEY PROCEDURES, NAVY AND MARINE CORPS AIRAELD PAVEMENTS, •NAVFAC INTERIM GUIDE. OCTOBER, 1985.

FIGURE 4 PCI procedure.

Broten et al.

four nights (6 hr available per night) to complete. Setting up the cones took less than 1 hr each night, and picking up the cones took about 114 hr. In one night, the parallel taxiway was refilmed, and the runway was filmed.

While one crew was filming the runway and taxiway, another crew was manually inspecting selected sample units for later comparison with film inspections. Unfortunately, there was some difficulty with the lighting equipment, and because of the last-minute change in the available runway and taxiway, the sample units were not premarked on the pavement. This made later comparisons difficult and not as reliable as planned.

Interpretation of Demonstration Project Film

The distress data were collected automatically using videotape; however, the distress identification process was conducted manually. This approach was selected because complete control of the data evaluation process was desired. The videotape was reviewed manually, with the aid of a computerized workstation.

The runway and taxiway films were reviewed by the same person who manually inspected the sample units during the demonstration project. The use of the same person allowed a direct comparison to be made between the automated and manual condition evaluation results because a subjective element exists in the rating of distress· severity levels.

The operation of the workstation used to review the videotapes is not difficult. It is harder to review videotapes collected during the night because the cameras that normally collect perspective information are not operational. It is not possible to provide sufficient lighting to these cameras. Thus, other than the distance measurement that is provided for each frame of film, there is no other easy way to determine exactly where the pavement is located that the technician is reviewing, which can be disorienting for the inspector. Under daylight conditions, the technician would use the perspective views to locate landmarks such as intersections and signs. ·

For the results to be useful for an airport application, it was determined that some alterations had to be made to the method in which the videotape is reviewed for road projects. Changes had to be made to accommodate concrete slabs 25 ft (7.6 m) wide. Also, inc.orporation of the automatic faulting and rutting measurements had to be made. Finally, recommendations for future videotaping were made on the basis of the film reduction process. Each of these items is discussed below.

Portland Cement Concrete Slab Review

The PCI procedure for portland cement concrete slabs requires that approximately 20 slabs per sample unit be inspected. The slabs on the runway and taxiway evaluated during the demonstration project have dimensions of 25 x 25 ft (7.6 x 7.6 m). However, the automated equipment provides a width coverage of 12.5 ft (3.8 m), which results in only onehalf of a slab being filmed in one pass. To perform a PCI in accordance with the FAA guidelines, it is necessary to evaluate an entire slab, not just a portion of a slab. A procedure was developed for merging the distress data identified by the

29

technician on the two consecutive film passes. The distress data are then filtered to eliminate any duplicate distress calls created by merging the results of two film passes.

Incorporation of Faulting and Rutting Measurements into Distress Review

Faulting and rutting measurements are gathered automatically by the VIV. This information must be merged with the distress data at the sample unit level. A method is used whereby the faulting and rutting can be identified at the sample unit level through the use of the distance measurements collected on each pass. The distress data and rutting and faulting measurements are then manually combined for each sample unit.

Recommendations for Full-Scale Implementation

A review of the film showed that in some situations the slab joints were not captured on film. This has serious ramifications on the PCI calculations because slab edges often are where the majority of distress occurs. For the full-scale implementation to be successful, corrections needed to be made in the data collection process to ensure that full slab coverage was achieved during videotaping.

The determination of distress severity was not difficult in most situations. However, it was felt that it would be beneficial to incorporate a way to determine actual crack width. A template was developed for the full-scale implementation that was spray painted onto each of the facilities evaluated. This template showed lines of%, 1/z, %, and 1 in. (0.64, 1.27, 1.91, and 2.54 cm) thick. In that way, the person evaluating the film can "calibrate" his or her eye with respect to crack thickness.

Because of the filming width constraint of 12.5 ft (3.8 m), it was determined that the videotaping process was not efficient for collecting distress data on parking lots, aprons, or short connectors with significant fillets. In these situations, it is difficult to accurately track the exact location of each piece . of film and to merge the results of the film reduction later into a condition index. A manual survey was recommended and utilized for these sections.

Comparison of Manual Inspections with Video Inspections

As discussed previously, an attempt was made to inspect selected sample units during the demonstration project both manually and through video interpretation. The purpose of this evaluation was to determine the feasibility of using automated data collection, followed by manual distress identification using the videotape, to obtain a reasonable estimate of a PCL Because of the unforeseen change in the project runway and taxiway, the locations of the sample units were not preinarked. In some cases, the filming of an area took place after the crew had manually marked and inspected a sample unit. The locations of these sample units showed up on the film, and a comparison was relatively simple. However, some areas were filmed before the manual survey crew con-

30

ducted their inspection, and it was not always possible to locate the exact area on film that was manually surveyed.

The problems encountered during this test limit the statistical significance of the results. However, it was possible to draw some general conclusions from the data. Overall, the major disadvantage of performing a PCI survey by reviewing videotapes is the lack of three-dimensional depth perception provided by the tapes. During a manual survey, the depth perception helps an inspector identify the severity of several distresses, including joint seal damage and weathering and raveling. The automated data collection procedure was more time consuming in the data interpretation steps but required far less field time. The major disadvantage of using manual inspection is the fact that distresses can be missed, particularly when conducted under inadequate lighting conditions or during distracting aircraft a~tivity. In addition, no permanent record of pavement condition is collected during a manual survey, as it is with the automated equipment.

Success of Demonstration Project

The results of the demonstration project were presented to the city of Chicago. In addition, city officials were shown the collected film from the demonstration project. City officials were very satisfied with the data collected and believed that the data more than met their needs. The alternative, which involved manually collecting the distress data, was not considered to be viable because of the conditions under which the inspection would have to be conducted. It was felt that although the measurement of certain distress types, such as low-severity weathering and raveling, was difficult using videotape, the benefits of the system offset the disadvantages. Members of the maintenance department were extremely pleased that they could obtain accurate measurements of total crack length, rather than having to rely on the extrapolation of inspected sample unit distresses to the section level. In addition, a permanent record of the pavement surface has the potential to provide more useful information than just distress data. For example, the maintenance department staff expressed an interest in using the film to calculate the area of paint striping on each facility. The engineering staff anticipates that the videotapes will be useful during communications with consultants, upper management, and maintenance personnel. The videotapes will make it much easier to locate distresses and effectively demonstrate the need for maintenance and rehabilitation work.

FULL-SCALE IMPLEMENTATION

A full-scale implementation was conducted at O'Hare International Airport June 22 through June 30, 1992. The automated equipment was used to videotape the remaining runways and parallel taxiways. Manual field crews collected distress information on the aprons, short connector taxiways, and the parking lots.


Conduct of Project

On nights when there were no traffic delays or equipment malfunctions, on the average the crew videotaped one runway and one parallel taxiway. The coning system devised during the demonstration project continued to work very well, and the lighting system developed for this nighttime application operated perfectly. The weather was excellent throughout the project and did not create any delays. However, because a film of water on the pavement surface results in a glare from the lights on the videotape, if rain had occurred, videotaping would have been halted.

Coordination with the operation's personnel worked smoothly. If a facility to be filmed was closed and did not cross any other active facilities, the crew was escorted to one end of that facility. The crew was then allowed to conduct the filming passes, while keeping alert for any unexpected traffic. In situations in which the crew was required to cross active facilities, an airport escort in a vehicle obtained clearance for the crew when needed. The filming was completed without incident.

Interpretation of Film

A computerized workstation for analyzing the videotapes was installed in the ERES office, and PaveTech personnel provided a training course in its use. The engineering technicians and project engineers who have the responsibility of conducting manual PCI surveys for the firm were assigned to the workstation. A high level of quality control was desired throughout the data reduction process, and a quality-assurance plan was implemented for the data interpretation process. Sample units that had already been processed were selected at random for reinspection. In addition, no technician was required to spend more than 3 consecutive hr at the workstation because it was found that fatigue is a factor if more time than that is spent reviewing the film.

USES OF COLLECTED DATA

The collected pavement condition data are being used to document current pavement conditions; identify maintenance and rehabilitation needs; estimate repair quantities and costs; and form the basis of a state-of-the-art PMS. Each of these items is discussed briefly in the following sections.

Documentation of Current Conditions

Distress data were evaluated using the videotapes in accordance with FAA AC 150/5380-6. The distress types and severities were identified by the engineering technician reviewing the film. A computer system was used to determine the starting and end point of each distress, length, area, and the location of each distress on the filming pass. The system was also used to record the pavement distress data at the sample unit level.

Broten et al.

The distress data were then loaded into a computerized PMS. The program automatically determines the sample unit and section PCis for each facility. These data then form the heart of the computerized PMS.

Estimation of Repair Quantities and Costs

The city of Chicago now possesses videotapes that cover 100 percent of the runways and parallel taxiways at O'Hare. The videotapes can be analyzed to obtain accurate crack counts or any other maintenance measurement that may be required. Combined with cost information, these data can then be used to develop cost estimates.

Basis for PMS

The distress data collected were entered into a PMS developed by ERES. This system has been linked to a computerized map of O'Hare to facilitate data retrieval. The implemented PMS allows the city of Chicago to manage the pavements at O'Hare International easily and rapidly. It contains advanced budgeting and planning software that the city has found to be extremely useful.

CONCLUSION

The objective of this paper was to discuss the adaptation of an automated distress data collection device to the airport environment. A demonstration project, followed by a fullscale implementation, at O'Hare International Airport was presented. The automated data collection approach was selected by the city of Chicago to minimize interference with

31

aircraft operations at the airport while pavement management data were obtained. The videotaping was conducted at night so that peak traffic flow interruption at this extremely busy airport was eliminated.

Several modifications to the equipment and procedures normally used for roadway data collection had to be made. During a typical road condition inventory, the equipment is operated during the day, and the coverage of 13.8 ft (4.2 m) wide is more than sufficient to film a lane. The data collection at O'Hare International Airport was performed at night on facilities that are up to 200 ft (61 m) wide. The modifications necessary for an airport application were made and then tested during a successful demonstration project.

The videotapes were interpreted manually to estimate PCI values, which were entered into a customized PMS. The distress data were determined to be acceptable by the city. Certain distress types, such as joint sealant damage and lowseverity weathering and raveling, were difficult to identify because of the two-dimensional nature of the videotape. However, it was felt that adequate interpretation of the data was possible and that the advantages outweighed any disadvantages. The videotapes provide the city with physical documentation of pavement conditions to supplement their PMS, and the information has already proven very useful for the city.

REFERENCE

1. Guidelines and Procedures for Maintenance of Airport Pavements. Advisory Circular 150/5390-6. FAA, U.S. Department of Transportation, 1982.



Maintenance Skid Correction Program in Utah

TRACY C. CoNTI, JAMES C. McMINIMEE, AND PRIANKA SENEVIRATNE

Deficiencies in resistance to skidding in sections of highways exist and sometimes contribute to elevated accident rates. A more efficient process for priority ranking projects for restorative treatment of these defiCiencies given limited funding is presented. Projects for treatment of these deficiencies using the benefit-cost ratio are discussed. Surface treatments are recommended and the

. associated project costs are identified. Benefits are identified through expected accident reductions. The costs associated with accidents are computed and multiplied by an accident reduction factor (ARF) to find the expected benefits of the countermeasure. Two separate methods are used in calculating the benefits. The first method involves using a standard ARF that is applied to all projects regardless of functional class and traffic volume. The second method utilizes new expected ARFs specific to each project. These new factors are based on the assumption that the countermeasure will reduce accidents to the average accident rate of each project's particular functional class. Individual projects are then selected from the prioritized lists using a dynamic pro-gramming technique. ·

When the United States was established, much of its legal system was patterned after the British system. Among the concepts brought to the new country was the principle of sovereign immunity. The sovereign immunity concept came to have the following meaning as the result of U.S. Supreme Court rulings in the early 1800s: the government could not be sued unless it gave its express permission, and even when it allowed itself to be sued, it was not responsible for the acts of its employees. This defense was almost unbeatable, and as a result governmental units were rarely brought to court on tort issues. Sovereign immunity became the primary defense against torts for state governments for almost a century and a half.

LOSS OF SOVEREIGN IMMUNITY

Eventually the courts began to realize the unfairness in the sovereign immunity defense. Several states lost their sovereign immunity through court decisions, usually by their state supreme courts. Most states viewed these cases as flukes and continued business as usual. Then in the late 1960s and through the 1970s, most states lost their immunity status through not only court decisions but also individual state legislation. As a result, the number of suits against the states mushroomed. Especially ripe were the transportation departments in which

T. Conti and J. C. McMinimee, Utah Department of Transportation, 4501 South 2700 West, Salt Lake City, Utah 84119. P. Seneviratne, Utah Center for Advanced Transportation Studies, Utah State University, Logan, Utah 84322.

endless maintenance defects on the older roads seemed to breed lawsuits. The number of tort claims against state transportation departments exploded from about 2,000 cases per year to more than 27 ,000 per year from 1976 to 1986 (1).

TORT ISSUES

The breach of a legal duty is the major issue in most tort liability cases involving skid accidents. This negligence is the failure to exercise such care as a reasonable and prudent person would under the circumstances. Neglecting a duty can be either wrongful performance or the omission of a required act (2).

If an agency can demonstrate that there is a systematic approach to treating deficiencies in resistance to skidding in the network and that the process has been followed in the case in question, it is easier to prove that the agency acted reasonably within the externally imposed constraints. This means that the agency needs to have in place a mechanism to

1. Routinely monitor the condition of the facilities, 2. Identify the deficient elements of the network, 3. Prioritize and program the deficient elements for treat

ment, and 4. Select appropriate warning or interim measures when

deficiencies cannot be corrected immediately.

Two key factors limit the abilities of highway agencies to follow such a procedure. One is the lack of reliable and upto-date information of every highway element. The other is the lack of resources. The first deficiency limits the ability to accurately and correctly detect the problem. Even when detected, the second deficiency sometimes prevents corrective measures from being implemented.

Agencies that overcome the first barrier and decide to invest in safety-related projects are faced with yet another responsibility_, that is, to develop procedures for ranking improvement projects and allocating the limited funds in the most effective manner. Because one of the strongest types of evidence to demonstrate the standard of care or the plan used to correct safety problems is the agency's own guidelines and policies, it is important that logical project prioritization and programming are available.

The primary objective of this paper is to present a study that proposes changes to the approach currently taken by the Utah Department of Transportation (UDOT) to identify highrisk locations and schedule treatment in order of priority.

Conti et al.

So how does a state transportation agency proceed with a program to correct pavements that are deficient in their ability to resist skidding and protect itself from being sued? The following are requirements for a defendable program:

1. A periodic inventory of the skid index throughout the highway system,

2. Accident data and history on the system including possible contributing cause, and

3. A procedure to logically prioritize and program proposed improvements within the available funding on the basis of skid index and accident data.

CURRENT SKID CORRECTION PRACTICE IN UTAH

Skid Index Studies

UDOT's skid correction program has been in operation since the late 1960s. A locked-wheel skid trailer was purchased recently to enhance the frequency of monitoring and inventory updating. At present, one test is taken at least every 2 years at each milepost with the trailer traveling at the posted speed limit but not exceeding 55 mph. Each measured skid index is then adjusted to a standard speed of 40 mph using a computerized speed correction method. When the corrected skid indexes approach a specified critical value, the operators

33

slow the testing speed to 40 mph and increase the number of tests to four per mile. Through a literature survey of skid studies and a poll of other western states, UDOT has established 35 as the critical skid index.

Any pavement with an index below 35 is considered substandard, and steps are taken to correct the condition. Pavements having skid indexes measured between 35 and 45 are considered marginal and those pavements above 45 are classified as standard. In the 1990-1991 inventory, of the 4,713 mi tested, 203 mi (4 percent) were substandard and 747 mi (16 percent) were marginal. These data, along with information on the structural adequacy, ride index, and pavement distress, are compiled by the planning division.

Because most deficiencies in a pavement's ability to resist skidding are corrected by district maintenance forces or with maintenance construction contracts, the district directors are notified of the deficient or substandard sections in the network under their respective jurisdictions. The data supplied for each section are the skid index, accident rate, percentage of wet weather accidents, and annual average daily traffic (AADT). Locations where the accident rate is higher than expected are noted. Tables 1 and 2 are examples of the information provided to the district directors. The example provided is from District 6, one of the six districts in Utah and comprises Utah, Juab, Daggett, Uintah, Wasatch, and Duchesne counties. Of the 1,051 mi of highway in District 6, 95 (9 percent) were classified as substandard.

TABLE 1 Highway Sections Deficient in Skid Resistance Identified in District 6

DISTRICT 6

PROJECT STATE MILEPOST DESCRIPTION FUNCTIONAL AADT NUMBER ROUTE CLASS

1 6 166.0 - Main St. to Ma.Col/M.Art 7,968 166.8 1000 East Mi.Co.IP.Art

2 28 23.5 - South of Principal 2,051 29.0 Levan Arterial

3 40 16.0 - Heber City Principal 9,619 19.5 Main Street Arterial

4 40 111.0 - SR-87 to 400 So. Principal 5,716 115.3 Roosevelt Arterial

5 40 147.0 - 700 So. Vernal Principal 5,775 152.0 To Rd. Left Arterial

6 41 0 - 4.8 So Nephi Int. Major 3,047 To 1-15 Collector

7 115 3.0 - 3.7 Rd. Crossing to Art/Maj. Col 1,598 SR-147 Urb Col/Min Col

8 121 0 - 1.5 SR-40 to Dry Major 1,441 Gulch Collector

9 121 36.0 - Highline Canal Maj. Col/ 1,715 39.0 to 1150 West Min Art

10 208 0 - 2.0 SR-40 to Major 167 M.P. 2 Collector

11 265 2.72 - 800 E. to Major 32,072 3.34 Canterville Rd. Collector


TABLE 2 Skid Index, Accident Rate, and Suggested Treatment of Identified Projects

DISTRICT 6

SKID 3YEAR %WET PROJECT INDEX AVG ACC WEATIIER TREATMENT NUMBER AVG (MIN) RATE* ACCIDENTS

1 37 (34) 3.868 3.70 PMSC

2 34 (20) 1.942 0.00 CHIP SEAL

3 39 (29) 3.608 13.53 PMSC

4 34 (29) 1.433 5.13 PMSC

5 34 (29) 1.328 9.52 PMSC

6 30 (24) 2.997 8.33 CHIP SEAL

7 26 (18) 3.266 0.00 CHIP SEAL

8 30 (27) 5.912 0.00 CHIP SEAL

9 38 (23) 7.986 2.22 CHIP SEAL

10 24 (19) 0.000 0.00 CHIP SEAL

11 23 (23) 2.159 6.38 PMSC

TOTAL µ.= 31 (25) 3.13 4.44

STD. DEV_. u= 5.1 (5) 2.25 4.61

* RA TE IN PER MILE PER l\.1ILLION VEHICLES

Programming Maintenance Skid Correction

The district director, together with the maintenance engineer and the district pavement management team, develops the maintenance program under which the skid deficiencies will be addressed. The maintenance program is integrated with the construction and rehabilitation programs. However, the fact that UDOT does not allocate specific funds to correct skid deficiencies means that not all sections will be treated immediately. Thus, the maintenance engineer is faced with the decision either to change or channel a certain amount of the funds from the regular maintenance budget or to postpone the corrective action until the deficient sections are part of the routine maintenance schedule.

Project Selection

Projects are selected on the perception of what projects are the worst candidates. The factors that ·are evaluated in determining these are the skid index, the accident rate, and the age and condition of the pavement surface. Funds required to address these are channeled from the maintenance budget, depending on the size of the maintenance budget and the maintenance engineer's gut feeling about its effect on the longterm implications on the regular maintenance program.

Surface Treatment Alternatives

Before scheduling a maintenance activity to correct the skid index, a review is made to determine whether the segment under consideration is programmed for rehabilitation or reconstruction. If the section needs rehabilitation and is pro-

grammed, it is determined whether the treatment can be deferred until the project begins. If the rehabilitation is scheduled too far in the future, a temporary treatment is considered or the rehabilitation project is moved forward to correct the problem earlier.

The common treatments on asphalt pavements in Utah are a chip seal, a slurry seal, and a plant mix seal coat. On concrete pavements the best option is grinding. Many defects such as rutting, minor cracking, and early raveling can be corrected for little additional cost while correcting the skidding problem. Treatment selection for sections deficient in their ability to resist skidding is based on two factors: (a) pavement type and (b) traffic counts. UDOT has written guidelines for specific treatments for specified AADTs. The deficient sections are addressed as dictated by these written strategies. The costs and locations of applicable treatments are shown in Table 3.

Proposed Modifications to Program

To improve the practice of addressing pavement sections deficient in their ability to resist skidding in Utah, the authors would like to propose some modifications to the current program. If a more systematic strategy were adopted, it may be able to direct a specific amount from the maintenance budget and invest it in a set of sections deficient in their ability to resist skidding that would reap the maximum benefit. For this purpose, it is suggested that the following programming process be adopted.

Step 1. All substandard and marginal projects would be ranked on the basis of expected benefit-cost (B/C) ratio. In the present case the benefits were estimated on the basis of expected reductions in accidents of different severities. To demonstrate the importance of employing appropriate acci-

Conti et al. 35

TABLE 3 Treatment Costs and Criteria for Surfaces Deficient in Skid Resistance

I I COST

I I TREATMENT PER LANE APPUCATION

MILE

PMSC $14,800 Asphalt Pavements Where AADT > 4,000

CIIlP SEAL $4,500 Asphalt Pavements Where AADT < 4,000

SLURRY $5,100 Asphalt Pavements In Shade Areas, SEAL Mountains, Or Intersections

CONCRETE $21,100 Concrete Pavements Surlace GRINDING Texturing

dent reduction factors, different project prioritization schemes were compared. One uses the standard factor currently in use by UDOT. The other uses a factor derived using the accident rates of functionally similar road sections and the expectation that a treatment would cause accident rates to trend toward the average.

Step 2. Step 2 involves defining what the funding constraints are for this program. It is recognized that UDOT will not have sufficient funds to address all projects at one given time. It was therefore decided to pursue this program on the basis of the estimated cost of correcting all of the known skid index deficiency problem areas over 4 years. In this way one-fourth of the problem areas in any given year would be addressed.

Step 3. Step 3 involves the dynamic programming process. This process involves simply going down the list of projects and their costs and including as many projects under the given funding level so as to optimize the funding.

Dynamic Programming

Dynamic programming is a process used to maximize funds. In this process, prioritized projects are included in the pro-

gram so that all of the available funds are utilized. The process looks at possible combinations of projects to program the available funds. Projects are included or deferred on the basis of their being able to fit within the program. In this way funds that expire on the basis of the fiscal year are maximized.

CASE STUDY

To illustrate the proposed procedure and some of its pros and cons, data from District 6 of the Utah DOT are used in this paper. Table 4 gives the project costs for the 11 segments determined to be deficient in their ability to prevent skidding. It is assumed that these segments are not currently programmed for reconstruction or rehabilitation and need to be addressed. The objective of the exercise now is to determine the projects that could be corrected with funds transferred from the maintenance budget. Often this amount does not cover all projects. Thus, it should be assigned to the optimal set of projects.

The B/C ratio method was used to initially rank the identified projects. To do this the number and severity of each

TABLE 4 Total Project Costs for Required Treatment

DISTRICT 6

COST TOTAL PROJECT STATE LANE PER LANE PROJECT NUMBER ROUTE MILES MILE COST

1 6 1.6 $14,800 $23,680

2 28 11.0 $4,500 $49,500

3 40 17.5 $14,800 $259,000

4 40 8.6 $14,800 $127,280

5 40 20.0 $14,800 $296,000

6 41 19.2 $4,500 $86,400

7 115 1.4 $4,500 $6,300

8 121 3.0 $4,500 $13,500

9 121 6.0 $4,500 $27,000

10 208 4.0 $4,500 $18,000

11 265 2.48 $14,800 $36,704

TOTAL 94.78 $943,364

36

type of accident, with a dollar value for each accident type, must be known. The Accidents Record Division of UDOT provided the data given in Table 5.

Using these dollar values and the number of each severity type of accidents, a total value of accident costs for 1991 was found for the projects in District 6 (Table 6).

The expected benefit of investing in each project was considered to be the savings in the present value of the expected accident (PVAC). Two methods were explored to arrive at these expected savings. In one method a standard accident reduction factor (ARF) of 42 percent was used. In the other the expected accidents were expected to decrease to the average accident rate for the functional class of road to which each project belongs. The assumption that the treatment will reduce all-not just wet weather accidents-was arrived at because any skid correction project also will include other measures, such as new striping and shoulder dressing, that enhance safety.

Method 1

The ARF of 42 percent for resurfacing was arrived at using the data provided to UDOT by the Texas Highway Depart-


ment. A discount rate of 4 percent and an 8-year design life was used for estimating the PV AC. (UDOT's Division of Safety currently uses an interest rate of 8 percent and a 20-year design life in computing the PV AC of accidents.) It may be expressed numerically as

PVAC = ARF[(NSl x 4,500) + (NS2 x 25,200)

+ (NS3 x 48,300) + (NS4 x 228,600)

+ (NS5 x 2,722,500)) x (PIA 4 % 8)

where

ARF = accident reduction factor, NS# = number of accidents of each severity type,

and (Pl A 4 % 8) = present worth factor of annual costs using

4 percent interest rate and 8 years of treatment life.

The BIC ratios are found by simply dividing the PV ACs by the project costs. Table 7 shows the respective BIC ratios for the projects in District 6. The projects can now be ranked according to either BIC ratios or benefits only. The rankings under the two criteria are shown in Tables 8 and 9.

TABLE 5 Associated Costs of Each Accident Severity Type

I SEVERITY

I DESCRIPTION

I DOLLAR

I NUMBER VALUE

1 Property Damage Only $4,500

2 Minor Injury $25,200

3 Possible Incapacitating Injury $48,300

4 Incapacitating Injury $228,600

5 Fatal Injury $2,722,500

TABLE 6 Number of Accidents and Associated Costs for Each Project

DISTRICT 6

PROJECT NUMBER OF ACCIDENTS 1989 - 1991 TOTAL NUMBER DOLLAR

SEVERITY SEVERITY SEVERITY SEVERITY SEVERITY VALUE TYPE 1 TYPE2 TYPE3 TYPE4 TYPES

1 18 3 4 2 0 $807,000

2 20 1 2 1 0 $440,400

3 95 14 19 5 0 $2,841,000

4 22 4 6 7 0 $2,089,800

5 24 6 8 4 0 $1,560,000

6 28 8 6 6 0 $1,989,000

7 2 1 1 0 0 $82,500

8 12 1 . 1 0 0 $127,500

9 28 4 6 1 1 $3,467,700

10 0 0 0 0 0 $0

11 32 5 5 5 0 $1,654,500

Conti et al. 37

TABLE 7 B/C Ratio of Identified Projects Using Texas ARF

DISTRICT 6

PROJECT ACCIDENT PROJECT BENEFIT-NUMBER REDUCTION PVAC "'"' COSTS COST

FACTOR"' RATIO

1 0.42 $760,660 $23,680 32.12

2 0.42 $415,111 $49,500 8.39

3 0.42 $2,667,864 $259,000 10.34

4 0.42 $1,969,800 $127,280 15.48

5 0.42 $1,470,422 $296,000 4.97

6 0.42 $1,874,788 $86,400 21.70

7 0.42 $77,763 $6,300 12.34

8 0.42 $120,179 $13,500 8.90

9 0.42 $3,268,578 $27,000 121.06

10 0.42 $0 $18,000 0.00

11 0.4i $1,559,495 $36,704 42.49

* USING TEXAS REDUCTION FACTORS PROVIDED BY UDOT SAFETY DIVISION

""" PRESENT WORTII OF ACCIDENTS USING 8 YEARS & 4% INTEREST

Method 2

The basic assumption in Method 2 is that the existing accident rate will be lowered to the accident rate for similar sections in the network. The logic is that it is unreasonable to expect surface treatments to have the same effect at all sites but that the accident rates would return to the average accident rate for similar roads. The observed accident rate and the mean accident rate for that class of roads can then be used to compute an expected ARF (EARF) for each project. EARF is expressed as a ratio of the difference between observed and expected accident rate to observed accident rate.

TABLE 8 B/C Ratio Ranking of Projects Using Texas ARF

DISTRICT 6

COST PROJECT (lOOO'S NUMBER DOLLARS)

9 27.0

11 36.7

1 23.7

6 86.4

4 127.3

7 6.3

3 259.0

8 13.5

2 49.5

5 296.0

10 18.0

The functional class volume group and the 5-year average accident rate for each project's functional group are presented in Table 10. The information on accident rates for each functional class was furnished by UDOT's Traffic and Safety Division, and these rates were used to compute the EARFs given in Table 11; the expected PV AC for each of the projects is presented in Table 12.

Despite the low skid indexes, the accident rates on Projects 2, 4, 5, 10, and 11 are less than the averages for the respective groups. Method 2 produces zero benefits for the above projects, and if the B/C ratios as shown in Table 13 or pure benefits shown in Table 14 are used, they will not be programmed but

TABLE 9 Benefit-Only Ranking of Projects Using Texas ARF

DISTRICT 6

COST PROJECT (lOOO'S NUMBER DOLLARS)

9 27.0

3 259.0

4 127.3

6 86.4

11 36.7

5 296.0

1 23.7

2 49.5

8 13.5

7 6.3

10 18.0


TABLE 10 Average Accident Rates for Each Project's Functional Class

DISTRICT 6

PROJECT URBAN FUNCTIONAL VOLUME 5 YEAR NUMBER OR CLASS GROUP AVGACC

RURAL (xlOOO) RATE*

1 SU 16 5 - 10 3.74

2 R 02 0-5 3.08

3 R 02 5 - 10 2.00

4 R 02 5 - 10 2.00

5 R 07 5 - 10 2.27

6 R 07 2.5 - 5 1.96

7 R 08 1 - 2 2.76

8 R 07 0 - 2.5 2.54

9 SU 16 0 - 2.5 4.38

10 R 07 0 - 2.5 2.54

11 u 14 25 - 35 6.04

* RA TE IS PER :MILE PER lMILLION VEHICLES

still should be scheduled for treatment. Those projects displaying zero benefits will be prioritized on the basis of the associated costs of the accidents, with those having the highest accident costs being the higher priority.

Allocation of Funds to Feasible Projects Using Dynamic Programming

With the imposed funding limitations, suppose it will be possible to fund $235,000 worth of projects each year for the next 4 years. According to the current practice, funds will be allocated to projects each year, starting w.ith the one showing the highest B/C ratio. This practice does not result in global optimization. Thus, although the projects funded will have

TABLE 11 Expected Accident Reduction Factors

DISTRICT 6

OBSERVED AVERAGE PROJECT ACCIDENT ACCIDENT EXPECTED NUMBER RATE RATE ARF

1 3.868 3.74 0.03

2 1.942 3.08 -

3 3.608 2.00 0.45

4 1.433 2.00 -

5 1.328 2.27 -

6 2.997 1.96 0.35

7 3.266 2.76 0.15

8 5.912 2.54 0.57

9 7.986 4.38 0.45

10 0 2.54 -

11 2.159 6.04 -

the highest B/C ratios, the total return on the investment may not be a maximum. In this paper, a dynamic programming approach will be applied to allocate funding during the next 4 years. This practice permits the selection of the set of projects that will maximize the benefits. Effectively, all the projects on the list will be considered and the available funds will be allocated sequentially so that all the funds are depleted or the remaining funds are insufficient to fund a complete project. The following is an example of . the process by which

TABLE 12 Present Value of Accident Costs Using New ARF

DISTRICT 6

PROJECT ACCIDENT NUMBER REDUCTION PVAC **

FACTOR*

1 O.Q3

2 -3 0.45

4 -

5 -

6 0.35

7 0.15

8 0.57

9 0.45

10 -11 -

* USING FACTORS ESTIMATED FOR FUNCTIONAL TYPE

$59,933

$0

$2,841,573

$0

$0

$1,544,525

$28,685

$163,204

$3,514,034

$0

$0

** PRESENT WORTII OF ACCIDENTS USING 8 YEARS & 4 % INTEREST

Conti et al.

dynamic programming would allocate funds in Year 1 to projects ranked on the basis of B/C ratios in Table 13.

Project Number

9 6 8 7 1

Cost ($ thousands)

27.0 86.4 13.5 6.3

23.7

Remaining Amount

235 - 27 = 208.0 121.6 108.1 101.8 78.1

Up to this point both the current approach and the dynamic programming approach give similar results. However, a dilemma occurs when a project shows a negative or zero return.

TABLE 13 B/C Ratio Ranking of Projects Using New ARF

DISTRICT 6

COST RANKING PROJECT (lOOO'S B/C

NUMBER DOLLARS) RATIO

1 9 27.0 130.15

2 . 6 86.4 17.88

3 8 13.5 12.09

4 3 259.0 10.97

5 7 6.3 4.55

6 1 23.7 2.53

7 4 127.3 0.00

8 11 36.7 0.00

9 5 296.0 0.00

10 2 49.5 0.00

11 10 18.0 0.00

TABLE 14 Benefits-Only Ranking of Projects Using New ARF

DISTRICT 6

COST RANKING PROJECT (lOOO'S BENEFITS

NUMBER DOLLARS)

1 9 27.0 $3,514,034

2 3 259.0 $2,841,573

3 6 86.4 $1,544,525

4 8 13.5 $163,204

5 1 23.7 $59,933

6 7 6.3 $28,685

7 4 127.3 $0

8 11 36.7 $0

9 5 296.0 $0

10 2 49.5 $0

11 10 18.0 $0

39

There are two options at this point. One is to carry the $78,100 over to the next year and allocate the new $235,000 + $78,100 to Project 3 that will yield an approximately $1,500,000 return or to proceed to treat the maximum number of sections having B/C ratios equal to 0. For instance, the $78,100 is not enough to fund Project 4 but is enough to fund Project 11 and still have sufficient funds remaining for Project 10.

The benefits of investing in Projects 11 and 10 may be simply related to the pavement life. Because such benefits are uncertain, it may be worthwhile to adopt the option of carrying the money over and then investing in Project 3 in the following year. On the other hand, from a risk minimization (loss control) point of view, it may pay to treat as many deficient sections as possible, starting with the section with the highest accident costs. Even with this approach, dynamic programming permits the selection of the maximum number of projects as opposed to the traditional approach that will allocate the funding to projects only according to the B/C ratios.

Using the risk minimization criterion, Projects 11 and 10 will be funded in Year 1 with the $78,100 remaining after the first iteration shown above. This will leave $23,400 to be carried over to Year 2. This still does not provide enough funding to complete Project 3 in Year 2, which has the highest remaining B/C ratio. Continuing the risk minimization approach, the funds would be allocated in Year 2 as follows:

Project Number

4 2

Cost ($thousands)

127.3 49.5

Remaining Amount

(235 + 23.4) - 127.3 = 131.1 81.6

The remaining amount would again be carried over to Year 3:

Project Number

3 5

Cost ($ thousands)

259.0 296.0

Remaining Amount

(235 + 81.6) - 259 = 57.6 0.

This example of dynamic programming does not completely illustrate the advantages of the method. Because the costs of some projects exceed the entire budget for 1 year, the amount of flexibility is limited. This still is better than the current practice. The current practice would split the larger projects into smaller ones to fit into the budget. This practice results in higher construction costs because two or more contractors would have to mobilize to complete the smaller projects. Another common practice is transferring unused funds elsewhere, leaving insufficient future funding to complete the required work.

One way to maximize the advantages of this programming technique in this situation would be to award Project 3 at the end of the Fiscal Year 2. This way the carryover amount from Year 1 could be spent on the project until Year 3 funds become available. Doing this would allocate the funds at the earliest possible time; therefore they would not be lost elsewhere.

As deficient sections are treated and new ones located, this programming procedure should be performed again to ensure that the projects with the highest benefits are completed. A time frame of every 2 or 3 years would be sufficient to serve this purpose.

40

CONCLUSIONS AND RECOMMENDATIONS

Currently UDOT has the necessary manpower and equipment to survey the state highway system for the necessary data. The data, including skid index, pavement distress, structural adequacy, accident data, and ride index, are collected at acceptable intervals. By evaluating these data, UDOT can be aware of the condition of its facilities. There are still two major decisions to be made: the first is to decide whether to correct the deficiencies at the expense of a disrupted maintenance program; the second is to determine the optimal allocation of those funds among the various projects, assuming that the decision was made to allocate a portion of the maintenance budget.

There is no policy at present on the transfer of funds from regular maintenance to skid correction. However, it was shown that if funds are appropriated to skid correction, dynamic programming could be used to optimally allocate the funds first on the basis· of B/C ratio and after that on the basis of the number of accidents. These two criteria can be viewed as efforts by the agency to utilize tax dollars to maximize public safety.


ACKNOWLEDGMENTS

The authors convey their gratitude to Douglas Anderson, of UDOT for his assistance in assembling and interpreting much of the data used to prepare this report. The editing and formatting of the paper were capably provided by Bryan Lee, and data compilation was completed by Matt Swapp, both of UDOT.

REFERENCES

1. Turner, D. S., J. K. Davis, and B. T. Wood. Status Report: Tort Liability Among State Highway Agencies. In Transportation Research Circular 361: Tort Liability and Risk Management. TRB, National Research Council, Washington, D.C., 1990, pp. 77-111.

2. Lewis, R. M. NCHRP Synthesis of Highway Practice 106: Practical Guidelines for Minimizing Tort Liability. TRB, National Research Council, Washington, D.C., 1983.

Publication of this paper sponsored by Committee on Surface Properties-Vehicle Interaction.


Consistency of Roughness and Rut Depth Measurement Collected with 11 South Dakota Road Profilers

SANJAY ASNANI, KHALED KSAIBATI, AND TURKI I. AL-SULEIMAN

Pavement roughness has long been recognized as a primary indicator of pavement performance. To provide accurate and reliable roughness measurements, the South Dakota Department of Transportation (SDDOT) designed and constructed a profilometer system in 1982. This system was later improved and enhanced by adding more sensors for rut measurements. The increased interest in the road profiler resulted in the establishment in 1989 of the South Dakota Road Profiler User's Group (SDRPUG). During the Third Annual SDRPUG meeting in Minnesota in 1991, international roughness index and rut depth data were collected with 11 road profilers on 4 different pavement surfaces. These selected pavement types were concrete, bituminous, concrete-bituminous over concrete, and bituminous over concrete. Each road profiler was run three times over each test section. The collected data were then reduced and analyzed statistically. The main objective of the statistical analysis was to determine whether the differences in roughness and rut measurements obtained with the 11 road profilers were statistically significant. The experiment and the statistical analysis are described in detail. In addition, specific recommendations are provided for the need to establish calibration procedures to ensure consistency in roughness and rut depth measurements obtained nationwide.

AASHTO pavement design procedure is based on the functional performance of pavements. Functional performance is measured by the serviceability index that incorporates a number of parameters such as pavement roughness, cracking, rutting, and patching. Because roughness is an indicator of all other parameters, some highway agencies calculate pavement serviceability index (PSI) on the basis of roughness measurements only.

Highway agencies use roughness to monitor the condition and performance of their pavement networks. The existing conditions of pavements, measured by roughness, determine the distribution of available funds for highway allocation such as providing routine maintenance, major maintenance, or reconstruction of a pavement section. In addition, roughness measurements often are employed as the dependent factor relative to the evaluation of new or modified pavements, pavement maintenance, materials, or construction techniques.

During the past few decades, roughness response devices were the primary instruments for measuring roughness. Results from these devices were known to be affected by the

S. Asnani and K. Ksaibati, Department of Civil Engineering, University of Wyoming, P.O. Box 3295, University Station, Laramie, Wyo. 82071. T.I. Al-Suleiman, Department of Civil Engineering, Jordanian University of Science and Technology, Irbid, Jordan.

condition of shock absorbers, wear and pressure of tires, and vehicles. These uncertainties greatly reduced the level of confidence in the data and demanded that consideration be given to the development of a more accurate and positive apparatus.

In the early 1980s the South Dakota Department of Transportation (SDDOT) developed and built a highway profiling and rut depth measurement system (1). This equipment, referred to as a road profiler, operates at highway speeds and measures pavement profile only in the left wheelpath. Pavement profile can be converted to any computerized roughness statistic. Over the years, quantifying roughness from pavement profiles proved to be much more accurate and reliable than depending on the point response of a vehicle.

SDDOT shared the road profiler technology with several other highway agencies. The demand for road profilers has become so great that they are now manufactured commercially. Today 8 states have duplicated the road profiler in house and about 20 others have purchased commercially manufactured systems (2). The following two reasons are behind the fast spread of this technology:

1. The FHW A requirement that pavement roughness measurements be reported in international roughness index (IRI) units.

2. The relatively low cost of the road profiler when compared with other available technologies.

Because of the increasing interest in measuring road profile, users of the road profiler began meeting annually to discuss feasible system enhancements. The first meeting was held in South Dakota in 1989, the second was held in Wyoming in 1991, and the last meeting was held in Minnesota in 1991. Eleven road profilers from 11 states participated in the meeting in Minnesota. The main objective of this paper is to investigate repeatability and consistency of roughness and rut depth measurements obtained with these 11 road profilers.

DESIGN OF EXPERIMENT

One major objective of the Minnesota experiment was to run the participating road profilers on several pavement test sections and then conduct statistical analysis on the collected IRI and rut depth measurements. Figure 1 graphically shows the data gathering and analysis strategies for the experiment. Pavement sites used in this study were selected to represent the range of surface types encountered in Minnesota. These


MINNESOTA EXPERIMENT

ELEVEN ROAD PROFILERS

MEASUREMENT REPEAT ABILITY OF INDIVIDUAL ROAD PROFILERS

COMPARISON AMONG ROAD PROFILERS

CONCLUSIONS

FIGURE 1 Data gathering and analysis strategies.

pavement types were concrete, bituminous, bituminous over concrete, and concrete-bituminous over concrete. All test sections were 0.2 mi long and selected to represent a wide range of roughness and rut depths of pavements in Minnesota. The test sections were conveniently located around the St. Paul area. Table 1 shows the locations and types of the selected eight test sections.

the IBM-based road profilers are commercially manufactured with slight hardware and software modifications. Table 2 provides a list of the participating road profilers and their types.

DATA COLLECTION

Seven participating road profilers were Digital Equipment Corporation (DEC)-based and four were IBM-based. The original South Dakota road profiler is DEC based whereas

On the second day of the Minnesota meeting, all road profilers' operators were given detailed information about the locations of the test sections. Data were then collected by all

TABLE 1 Test Section Types and Locations

TEST SECTION NO. PAVEMENT TYPE LOCATION

1 CONCRETE 1-94 EAST

2 CONCRETE 1-94 WEST

3 BITUMINOUS C0-10 WEST

4 BITUMINOUS C0-10 EAST

5 BITUMINOUS OVER CONCRETE IS-694 NORTH

6 BITUMINOUS OVER CONCRETE IS-694 SOUTH

7 CO NC/BOC MN-5 EAST

8 CO NC/BOC MN-5 WEST

Asnani et al. 43

TABLE 2 Road Profilers That Participated in Minnesota Experiment

I ROAD PROFILER NUMBER II STATE II TYPE I 1 WYOMING (WY) DEC

2 NEBRASKA (NE) DEC

3 MINNESOTA (MN) DEC

4 WISCONSIN (WI) DEC

5 ILLINOIS (IL) DEC

6 NORTH DAKOTA (ND) DEC

7 SOUTH DAKOTA (SD) DEC

8 IOWA (IA) IBM

9 ALABAMA (AL) IBM

10 MONTANA (MT) IBM

11

11 road profile:i;-s at the same time. The collected data included pavement roughness expressed in IRI and rut depth measurements. Each road profiler was run three times on _each test section. In total, each road profiler made 24 runs. Tables 3 and 4 summarize in tabular form the collected roughness and rut depth data respectively.

DATA ANALYSIS

The main objectives of the statistical analysis were to

1. Investigate the repeatability of measurements for individual road profilers,

2. Compare results from different road profilers, and 3. Determine the effect of pavement type on the repeata

bility of road profilers.

The data collected during the experiment were adequate to satisfy the first two. objectives only.

Repeatability of Measurements of Individual Devices

Each road profiler was run three times on each test section. Roughness and rut depth measurements from all three runs were then averaged, and the standard deviations were calculated. Table 3 summarizes the averages and standard deviations for all systems on all test sections. It is clear from Table 3 that the standard deviations for all measurements were extremely low, which indicates that the overall repeatability of measurements for all road profilers is very good.

Comparisons Among Road Profilers

Roughness and rut depth measurements from all 11 road profilers were first examined visually without conducting any

IDAHO (ID) IBM

analysis. This preliminary examination indicated some variations in the results from various road profilers. As an example, Table 3 shows that the roughness of Test Section 1 is 1.41 when measured with the South Dakota road profiler and 1.11 when measured with the Idaho road profiler. Therefore, it was necessary to determine the statistical significance of these differences. The two-sample t-test was used in the cqmparison among the means. Basically, the measurements from any two road profilers were compared to see whether they were statistically equal. A 95 percent confidence level was used in the whole analysis to be within practical limits. To conduct the t-test the following assumptions were made:

1. The population samples are small. 2. Both the populations are normal with CT1 = CT2 = CT, and

the design is completely randomized.

The t-value was calculated with the following equation:

(1)

where

Y1 , Y2 sample means, n1 , n2 = sample sizes, and

SP = estimate of common variance CTi = CT~ = CT2•

The common variance SP was computed with the following equation:

~ = ~(n_1~----'l)~CT_i~+---'-(n_2_-~~l)_CT_~ P n 1 + n2 - 2

(2)

where Si and ~ are the two individual sample variances. In the analysis of IRI and rut depth data, the previously

described two-sample t-test was used. Means of IRI and rut

TABLE 3 IRI Data Collected at Minnesota Experiment

CONC MN SD IA AL ND MT ID WY 'NE WI IL TEST #1 IS 94 (EB)

RUN 1 1.26 1.36 1.20 1.19 1.21 1.14 1. 11 1.28 1.37 1.26 1.34 RUN 2 1.27 1.42 1.18 1.18 1.19 1.14 1.10 1.37 1.49 1.31 1.37 RUN 3 1.26 1.45 1.19 1.18 1.22 1.08 1.13 1.38 1.35 1.25 1.35

AVERAGE 1.26 1.41 1.19 1.18 1.21 1.12 1. 11 1.34 1.40 1.27 1.35 STD. DEVIATION 0.01 0.05 0.01 0.01 0.02 0.03 0.02 0.055 0.08 0.03 0.02

CONC MN SD IA AL ND MT ID WY NE WI IL TEST #2 IS 94 IWBI

RUN 1 1.50 1.56 1.40 1.46 1.38 1.31 1.53 1.51 1.53 1.38 1.45 RUN 2 1.44 1.64 1.42 1.46 1.39 1.26 1.53 1.60 1.45 1.37 1.47 RUN 3 1.49 1.59 1.41 1.43 1.39 1.27 1.38 1.53 1.46 1.37 1.49

AVERAGE 1.48 1.60 1.41 1.45 1.39 1.28 1.48 1.55 1.48 1.37 1.47 STD. DEVIATION 0.03 0.04 0.01 0.02 0.01 0.03 0.09 0.047 0.04 0.01 0.02

BIT MN SD IA AL ND MT ID WY NE WI IL TEST #3 CO 10 (WB)

RUN 1 4.47 4.38 4.30 4.43 4.53 3.88 4.21 4.58 4.23 4.51 4.53 RUN 2 4.56 4.31 4.27 4.54 4.36 3.93 4.24 4.51 4.44 4.46 4.43 RUN 3 4.54 4.28 4.34 4.32 4.41 3.97 4.24 4.48 4.43 4.53 4.38


BIT MN SD IA AL ND MT ID WY NE WI IL TEST #4 CO 10 (EBI

RUN 1 4.53 4.42 4.39 4.48 4.52 3.73 4.18 4.70 4.47 4.45 4.43 RUN 2 4.81 4.27 4.36 4.25 4.55 3.72 4.14 4.54 4.51 4.56 4.44 RUN 3 5.13 4.27 4.39 4.21 4.54 3.74 4.05 4.54 4.51 4.40 4.45


BOC MN SD IA AL ND MT ID WY NE WI IL TEST #5 IS 694 (NB)

RUN 1 1.07 1.10 0.99 0.95 1.25 0.86 0.87 1.03 1.03 0.99 1.05 RUN 2 1.09 0.98 0.99 0.89 1.19 0.86 0.91 1.03 1.06 1.02 1.09 RUN 3 1.07 1.04 0.99 0.93 1.03 0.85 0.89 1.01 1.05 1.01 1.14

AVERAGE 1.08 1.04 0.99 0.92 1.16 0.86 0.89 1.02 1.05 1.01 1.09 STD. DEVIATION 0.01 0.06 0 0.03 0.11 0.01 0.02 0.012 0.02 0.02 0.05

BOC MN SD IA AL ND MT ID WY NE WI IL TEST #6 IS 694 (SB)

RUN 1 1.16 1.08 1.00 0.88 0.97 0.87 0.90 1.05 1.07 1.05 1.09 RUN 2 1. 11 1.04 1.00 0.87 0.95 0.85 0.89 1.07 1.15 1.08 1.07 RUN 3 1.04 1.03 1.02 0.88 0.95 0.88 0.85 1.10 1.13 1.07 1.07

AVERAGE 1.10 1.05 1.01 0.88 0.96 0.87 0.88 1.07 1.12 1.07 1.08 STD. DEVIATION 0.06 0.03 0.01 0.01 O.Q1 0.02 0.03 0.025 0.04 0.02 0.01

CONC/BOC MN SD IA AL ND MT ID WY NE WI IL TEST #7 MN 5 (EBI

RUN 1 2.23 2.63 1.99 2.08 2.07 1.84 1.95 . 2.10 2.23 2.16 2.28 RUN 2 2.22 2.55 1.96 2.08 2.09 1.86 1.98 2.20 2.25 2.02 2.26 RUN 3 2.24 2.53 2.03 2.01 2.11 1.84 1.98 2.19 2.23 2.15 2.29

AVERAGE 2.23 2.57 1.99 2.06 2.09 1.85 1.97 2.16 2.24 2.11 2.28 STD. DEVIATION 0.01 0.05 0.04 0.04 0.02 O.Q1 0.02 0.055 0.01 0.08 0.02

CO NC/BOC MN SD IA AL ND MT ID WY NE WI IL TEST #8 MN 5 (WBI

RUN 1 2.15 2.16 1.92 2.07 2.03 1.72 2.02 2.30 2.11 2.04 2.19 RUN 2 2.27 2.12 1.95 1.98 2.06 1.74 2.04 2.37 2.11 2.06 2.18 RUN 3 2.19 2.16 1.97 2.09 2.02 1.73 2.00 2.33 2.09 2.03 2.15


Asnani et al. 45

TABLE 4 Rut Depth Data Collected at Minnesota Experiment

BIT MN SD ND TEST #1 CO 10 (WB)

RUN 1 0.64 0.54 0.67 RUN 2 0.64 0.49 0.71 RUN 3 0.63 0.49 0.66

AVERAGE 0.63 0.51 0.68 STD. DEVIATION 0.01 0.03 0.03

BIT MN SD ND TEST #2 CO 10 (EB)

RUN 1 0.53 0.43 0.62 RUN 2 0.55 0.41 0.64 RUN 3 0.55 0.36 0.63

AVERAGE 0.54 0.40 0.63 STD. DEVIATION O.Q1 0.04 0.01

BOC MN SD ND TEST #3 IS 694 (NB)

RUN 1 0.11 0.05 0.17 RUN 2 0.11 0.02 0.17 RUN 3 0.12 0.02 0.17

AVERAGE 0.11 0.03 0.17 STD. DEVIATION 0.01 0.02 0

BOC MN SD ND TEST #4 S 694 (SB)

RUN 1 0.09 0.02 0.12 RUN 2 0.08 0.01 0.13 RUN 3 0.09 0.03 0.13

AVERAGE 0.09 0.02 0.13 STD. DEVIATION 0.01 0.01 0.01

depths for all three runs on each test section were calculated and compared with each other. The test statistic t was then determined by using Equation 1, and finally its absolute value was compared with twz,ni+nz-z = 2.776 (for O'. = 0.05 and 4 degrees of freedom since n1 = n2 = 3). If ABS(t) > twz,ni +nz-Z• it would be concluded that the two means are statistically different. A large number of paired comparisons were made. As an example, roughness measurements from each road profiler were compared with the measurements from 10 other road profilers on eight test sections, which would result in 80 possible comparisons. The results from all of these comparisons are summarized in Tables 5 and 6 for roughness and rut depth measurements, respectively. It is clear from examining these tables that the road profilers produced equal IRI measurements in 35.5 percent of the cases and equal rut depth measurements in only 25. 7 percent of the cases. These extremely low percentages are alarming because all the systems are similar in design.

To find the reason behind the differences in measurements from the 11 road profilers, an additional statistical analysis was conducted. This analysis aimed at determining whether there are any linear relationships among IRI and rut depth data collected with different road profilers. A regular regression approach was used to establish these relationships. The following basic regression model (i.e., simple linear param-eters) was used in the analysis: ·

WY

0.65 0.64 0.64 0.64 0.01

WY

0.59 0.58 0.58 0.58 0.01

WY

0.16 0.16 0.17 0.16 0.01

WY

0.11 0.12 0.12 0.12 0.01

NE WI IL IA AL MT ID

0.58 0.63 0.63 0.72 0.65 0.58 0.59 0.60 0.64 0.60 0.70 0.66 0.58 0.58 0.60 0.65 0.61 0.72 0.66 0.60 0.56 0.59 0.64 0.61 0.71 0.66 0.59 0.58 0.01 O.Q1 0.02 O.Q12 O.Q1 0:01 0.02


0.50 0.51 0.55 0.61 0.57 0.43 0.48 0.49 0.52 0.54 0.62 0.53 0.43 0.48 0.47 0.51 0.54 0.61 0.54 0.42 0.44 0.49 0.51 0.54 0.61 0.55 0.43 0.47 0.02 O.Q1 O.Q1 0.007 0.02 0.01 0.02


0.05 0.11 0.12 0.11 0.09 O.Q1 O.Q1 0.05 0.11 0.12 0.11 0.09 0.02 0.00 0.05 0.12 0.12 0.12 0.10 0.02 0.00 0.05 0.11 0.12 0.11 0.09 0.02 0.00

0 0.01 0 0.007 0.01 0.01 0.01


0.03 0.07 0.06 0.08 0.05 0.00 0.00 0.03 0.08 0.06 0.08 0.05 0.00 0.00 0.03 0.08 0.06 0.08 0.05 0.00 0.00 0.03 0.08 0.06 0.08 0.05 0.00 0.00

0 0.01 0 0 0 0 0

(3)

where

Y; = mean of IRI or rut depth for three runs by one profiler,

X; = mean of IRI or rut depth by another road profiler,. and

B0 , B 1 = regression constants.

Tables 7 and 8 present summaries of the regression equations for IRI and rut depth measurements, respectively. These regression equations yield very high R-square (100% in some cases), which indicate almost perfect agreement among systems. Sample plots of the raw data used in the regression analysis are shown in Figures 2 and 3.

The t-test results can be now explained on the basis of the results from the regression analysis. Although all participating road profilers are similar in design, they should be calibrated against each other before making any attempts for comparisons. Unfortunately, the South Dakota-type road profilers are used by different highway agencies to create a national roughness data base without calibration. This national data base can be used to compare roughness measurements within any individual state. However, roughness measurement comparison for sections in various states will not be accurate without calibration.

TABLE 5 Results from IRI Comparisons

SYSTEM

MN

SD

IA

AL

ND

MT

ID

WY

NE

WI

IL

SECTION

CONC BIT BOC

CONC/BOC

ALL SECTIONS

CONC BIT

BOC CONC/BOC

ALL SECTIONS

CONC BIT

BOC CONC/BOC

ALL SECTIONS

CONC BIT BOC

CO NC/BOC

ALL SECTIONS

CONC BIT

BOC CONC/BOC

ALL SECTIONS

CONC BIT

BOC CONC/BOC

ALL SECTIONS

CONC BIT

BOC CONC/BOC

ALL SECTIONS

CONC BIT

BOC CO NC/BOC

ALL SECTIONS

CONC BIT

BOC CO NC/BOC

ALL SECTIONS

CONC BIT

BOC CONC/BOC

ALL SECTIONS

CONC BIT

BOC CO NC/BOC

ALL SECTIONS

POSSIBLE I COMPARISONS

to 20 20 20

80

20 20 20 20

80

20 20 20 20

80

20 20 20 20

80

20 20 20 20

80

20 20 20 20

80

20 20 20 20

80

20 20 20 20

80

20 20 20 20

80

20 20 20 20

80

20 20 20 20

80

GOOD COMPARISONS

8 13 IO 5

36

5 9 IO 3

27

4 8 5 4

21

6 16 3 IO

35

3 13 7 6

29

2 0 3 0

5

IO 2 4 5

21

10 10 10 5

35

10 14 9 3

36

4 14 8 9

35

8 11 9 3

31

.... ·.·• : ................. .

% GOOD COMAPRISONS

40 65 50 25

45

25 45 50 15

33.8

20 40 25 20

26.3

30 80 15 50

43.8

15 65 35 30

36.3

10 0 15 0

6.25

50 10 20 25

26.3

50 50 50 25

43.8

50 70 45 15

45

20 70 40 45

43.8

40 55 45 15

38.8

•·.······· \_::/-.,,{,····· ·•.· .. • .. .

··· I. c:: I ·<>•-•·•:''.'_~· ·· \ \···• ... .

Asnani et al. 47

TABLE 6 Results from Rut Depth Comparisons

~, SECTION POSSIBLE II GOOD L____J. COMPARISONS II COMPARISONS

UVVJJ

I COMPARISONS

MN BIT 20 6 30 35 BOC 20 7

SD

IA

AL

ND

MT

ID

WY

NE

WI

MN

BOTH BIT & BOC

BIT BOC

BOTH BIT & BOC

BIT BOC

BOTH BIT & BOC

BIT BOC

BOTH BIT & BOC

BIT BOC

BOTH BIT & BOC

BIT BOC

BOTH BIT & BOC

BIT BOC

BOTH BIT & BOC

BIT BOC

BOTH BIT & BOC

BIT BOC

BOTH BIT & BOC

BIT BOC.

BOTH BIT & BOC

Effect of Pavement Type on Repeatability of Measurements

As shown in Table 5, the percentages of good IRI comparisons were 50 and 31.8 percent on bituminous and concrete sections, respectively. These percentages may lead someone to believe that measurements on bituminous surfaces are more repeatable than measurements on concrete sections. But since all bituminous sections were rough and all concrete sections were smooth, the factor roughness level should be taken into consideration. In other words, the encountered differences could

40

20 20

40

20 15

35

20 15

35

20 19

39

20 15

35

20 15

35

20 20

40

20 15

35

20 20

40

13

2 7

9

2 5

7

10 2

12

6 2

8

7 4

11

7 3

IO

4 2

6

4 3

7

6 7

32.5

10 35

22.5

IO 33.3

20

50 13.3

34.3

30 10.5

20.5

35 26.7

31.4

35 20

28.6

20 10

15

20 20

20

30 35

be due to the effect of roughness level rather than pavement type. In this experiment, the selected sections did not reflect all roughness ranges. Therefore, no conclusive conclusions could be obtained with respect to the effect of pavement type on the repeatability of measurements.

CONCLUSIONS AND RECOMMENDATION

In this research, 11 South Dakota-type road profilers participated in collecting roughness and rut depth data in Min-

TABLE 7 IRI Calibration Equations

SYSTEMS REGRESSION EQUATION R-SQUARE(%)

MN SD IRir.rn = -0.229 + 1.11 IRl50 98.2

MN IA IRIMN = 0.0218 + 1.08 IRI,A 99.7

MN AL IRIMN = 0.0734 + 1.05 IRIAL 99.3

MN ND IRIMN = 0.0214 + 1.04 IRIND 99.7

MN MT IRIMN = -0.015 + 1.23 IRIMI' 99.0

MN ID IRIMN = 0.029 + 1.11 IRl10 99.1

MN WY IRIMN = -0.0687 + 1.04 IRIWY 99.6

MN NE IRIMN = -0.133 + 1.08 IRINE 99.7

MN WI IRIMN = 0.0185 + 1.04 IRIWI 99.6

MN IL IRIMN = -0.126 + 1.07 IRIIL 99.6

SD IA IRI50 = 0.262 + 0.95 IRI1A 98.1

SD AL IRl50 = 0.295 + 0.932 IRIAL 98.7

SD ND IRl50 = 0.262 + 0.918 IRINo 97.9

SD MT IRl50 = 0.215 + 1.09 IRIMT 98.7

SD ID IRl50 = 0.257 + 0.982 IRl10 98.5

SD WY IRI50 = 0.177 + 0.917 IRiwv 98.3

SD NE IRl50 = 0.115 + 0.96 IRINE 98.9

SD WI IRl50 = 0.254 + 0.921 IRlwi 98.3

SD IL IRl50 = 0.117 + 0.956 IRI1L 99.1

IA AL IRllA = 0.0465 + 0.975 IRIAL 99.6

IA ND IRllA = 0.0012 + 0.966 IRIND 99.8

IA MT IRl1A = -0.0381 + 1.14 IRIMT 99.6

IA ID IRl1A = 0.0056 + 1.03 IRl10 99.5

IA WY IRl1A = -0.082 + 0.962 IRlwv 99.6

IA NE IRIIA = -0.142 + 1.01 IRINE 99.8

IA WI IRllA = -0.004 + 0.967 IRIWI 99.9

IA IL IRl1A = -0.136 + 0.999 IRl1L 99.6

AL ND IRIAL = -0.0385 + 0.987 IRIND 99.4

AL MT IRIAL = -0.085 + 1.17 IRIMI' 99.8

AL ID IRIAL = -0.0429 + 1.06 IRI10 99.9

AL WY IRIAL = 0.129 + 0.985 IRlwv 99.7

AL NE IRIAL = -0.186 + 1.03 IRINE 99.5

AL WI IRIAL = -0.045 + 0.989 IRIWI 99.6

AL IL IRIAL = -0.184 + 1.02 IRIIL 99.8

ND MT IRIND = -0.0349 + 1.18 IRIMI' 99.3

ND ID IRIND = 0.0105 + 1.06 IRl10 99.2

ND WY IRIND = -0.08 + 0.994 IRIWY 99.3

ND NE IRIND = -0.143 + 1.04 IRINE 99.5

ND WI IRIND = -0.0003 + 0.999 IRIWI 99.7

ND IL IRIND = -0.138 + 1.03 IRIIL 99.6

MT ID IRIMT = 0.0405 + 0.899 IRl10 99.6

MT WY IRIMT = -0.0317 + 0.893 IRlwv 99.3

MT NE IRIMI' = -0.0844 + 0.877 IRINE 99.5

MT WI IRIMT = 0.035 + 0.844 IRlwi 99.8

MT IL IRIMT = -0.0821 + 0.873 IR111• 99.7

ID WY IRIID = -0.0804 + 0.933 IRIWY 99.7

ID NE IRllD = -0.132 + 0.972 IRINE 99.3

ID WI IRl10 = 0.0015 + 0.935 IRiwi 99.4

ID IL IRllD = -0.13 + 0.968 IRllL 99.5

WY NE IRIWY = -0.055 + 1.04 IRINE 99.6

(continued on next page)

TABLE 7 (continued)

SYSTEMS REGRESSION EQUATION R-SQUARE(%)

WY WI IRiwv = 0.0896 + 1.0 IRiwi 99.5

WY IL IRlwv = -0.0512 + 1.04 IRI1L 99.6

NE WI IRINE = 0.141 + 0.961 IRIWI 99.8

NE IL IRINE = 0.0063 + 0.993 IRIIL 99.8

WI IL IRIWI = -0.137 + 1.03 IRI1L 99.8

TABLE 8 Rut Depth Calibration Equation

I SYSTEMS I REGRESSION EQUATION I R-SQUARE(%) I MN SD RUTMN = 0.0685 + 1.32 RUT50 97.2

MN ND RUTMN = -0.045 + 0.963 RUTND 99.6

MN WY RUT MN = - 0.0452 + 1.03 RUTwv 99.7

MN NE RUTMN = 0.0616 + 0.969 RUTNE 100.0

MN WI RUT MN =· 0.008 + 0.999 RUTWI 99.6

MN IL RUTMN = 0.0107 + 0.998 RUTIL 99.6

MN IA RUT MN= 0.0123 + 0.986 RUTIA 94.6

MN AL RUTMN = 0.0319 + 1.05 RUTAL 94.9

MN MT RUTMN = 0.0955 + 0.95 RUTMT 99.2

MN ID RUTMN = 0.101 + 0.922 RUTm 99.9

SD ND RUT50 = -0.0806 + 0.716 RUTNo 98.7

SD WY RUT5D = -0.0797 + 0.766 RUTwv 98.2

SD NE RUT50 = 0.0011 + 0.712 RUTNE 96.8

SD WI RUT50 = -0.036 + 0.727 RUTwi 94.8

SD IL RUT5D = -0.038 + 0.738 RUT1L 97.6

SD IA RUT50 = -0.0453 + 0.755 RUT1A 99.5

SD AL RUT50 = -0.03 + 0.805 RUTAL 99.5

SD MT RUTsD = 0.0286 + 0.688 RUT MT 93.3

SD ID RUTsD = 0.03 + 0.676 RUTm 96.6

ND WY RUT ND = 0.0004 + 1.07 RUT WY 100.0

ND NE RUTND = 0.112 + 1.0 RUTNE 99.4

ND WI RUTND = 0.0578 + 1.03 RUTWI 98.6

ND IL RUT ND = 0.0583 + 1.04 RUTIL 99.8

ND IA RUTND = 0.0561 + 1.03 RUTIA 97.0

ND AL RUT ND = 0.0776 + I. I RUT AL 97.3

ND MT RUTND = 0.149 + 0.977 RUTMT 97.6

ND ID RUTND = 0.153 + 0.951 RUTID 99.1

WY NE RUTwv = 0.104 + 0.934 RUTNE 99.6

WY WI RUTwv = 0.0529 + 0.962 RUTwi 99.0

WY IL RUTwv = 0.0539 + 0.966 RUT1L 99.9

WY IA RUTwv = 0.0532 + 0.961 RUTIA 96.3

WY AL RUTwv = 0.0722 + 1.03 RUTAL 96.7

WY MT RUTwv = 0.138 + 0.913 RUTMT 98.2

WY ID RUTwv = 0.142 + 0.888 RUTm 99.3

NE WI RUT NE = -0.0555 + 1.03 RUTWI 99.7

NE IL RUT NE = -0.0524 + 1.03 RUTIL 99.5

NE IA RUT NE = -0.05 + 1.02 RUT IA 94.2

NE AL RUT NE = -0.0299 + 1.08 RUTAL 94.5

NE MT RUTNE = 0.0347 + 0.982 RUTMT 99.3

NE ID RUTNE = 0.0402 + 0.951 RUTm 99.9

WI IL RUTwi = 0.0042 + 0.995 RUTIL 99.1


50

5.00

Q:: ~ 4.00 G: 0 Q:: a..

~ 3.00 Q::

0 I <(

Q 2.00 :::!: 0 Q:: u..

~ 1.00

TABLE 8 (continued)

SYSTEMS

WI IA

WI AL

WI MT

WI ID

IL IA

IL AL

IL MT

IL ID

IA AL

IA MT

IA ID

AL MT

AL ID

MT ID

REGRESSION EQUATION

RUTWI = 0.0101 + 0.970 RUTIA

RUTWI = 0.0291 + 1.04 RUTAL

RUTwi = 0.0872 + 0.953 RUTMT

RUTWI = 0.0935 + 0.92 RUTID

RUT1L = 0.0004 + 0.991 RUTIA

RUTIL = 0.0199 + 1.06 RUTAL

RUT1L = 0.0866 + 0.946 RUTMT

RUTIL = 0.0916 + 0.918 RUTID

RUTIA = 0.0204 + 1.07 RUTAL

RUTIA = 0.103 + 0.892 RUTMT

RUTIA = 0.104 + 0.881 RUT10

RUTAL = 0.0769 + 0.839 RUTMT

RUT AL = 0.0779 + 0.827 RUTID

RUT MT = 0.0072 + 0.963 RUTID

E' 520.00

Q:: w _. c.:: 0 Q:: a.. 15.00

~ Q::

~ ~ 10.00 <(

:::!: 0 Q:: u..

~ 5.00 w 0

t:::> Q::


R-SQUA

91.7

92.2

99.9

99.6

95.7

96.2

99.2

99.0

100.0

89.8

93.7

90.2

93.9

99.4

0. 00 -+r...,...,..,r-T"T""l"'T"T"T"TT'T"T'"TTTT"T"rl"'T"...,...,..,r-T"T""I.....,..,.,..,.........,..,.~...,...,..,~ 0. 00 -+-r..-.-.r-r-r"T"T"ln-r,..,.,"T"T"T"T"T"T"T""m-rrm-rrm"T"T""rTT"I

0.00 1 .00 2.00 3.00 4.00 5.00 IRI FROM AU\BAMA ROAD PROFILER

FIGURE 2 IRI correlation between Idaho and Alabama road profilers (R2 = 99.9 percent)

0.00 5.00 10.00 15.00 20.00 RUT DEPTH FROM IOWA ROAD PROFILER (mm)

FIGURE 3 Rut depth correlation between Iowa and Alabama road profilers (R2 = 100.0 percent).

nesota. Eight pavement test sections were included in the experiment to reflect the various pavement types encountered in Minnesota. Each road profiler was run three times on all test sections. The collected data were then reduced, tabulated, and analyzed statistically. This analysis leads to the following conclusions:

almost all relationships. These relationships indicate that the systems do correlate among each other.

4. There is no conflict in the findings stated in Items 1 and 2. They simply reflect the fact that road profilers should be calibrated before any comparisons are conducted. Calibration will ensure the validity of the comparison.

1. Roughness and rut depth measurements obtained with any single system seem to be repeatable.

2. The t-test results indicate that roughness measurements obtained with all systems were statistically different in 64.5 percent of the cases. On the other hand, rut depth measurements were statistically different in 74.3 percent of the cases.

3. The regression analysis yielded very strong linear rela.· tionships among systems. R-squares were in the upper 90s for

5. The data collected were not adequate to determine whether pavement type influenced the repeatability of measurements of road profilers.

Finally, the urgency for establishing calibration procedures for South Dakota-type road profilers cannot be overemphasized. Highway agencies invest a huge amount of resources in collecting roughness data every year. Roughness data from all states are used by FHW A to determine the level of de-

Asnani et al.

terioration for the pavement network nationwide. For the FHW A and different states to use roughness data effectively, all states using the South Dakota-type road profiler should calibrate their devices to ensure data consistency. Calibration could be done by establishing regional calibration sites that could be used in establishing calibration factors that would ensure that roughness devices operating across the United States produce comparable results.

ACKNOWLEDGMENTS

This cooperative study was funded by the U.S. Department of Transportation's University Transportation Program through the Mountain-Plains Consortium, the Wyoming Transportation Department, and the University of Wyoming.

51

REFERENCES

1. Huft, D. L. Analysis of Errors for the South Dakota Profilometer. In Transportation Research Record 1000, TRB, National Research Council, Washington, D.C., 1984.

2. South Dakota Road Profiler User's Group. Report of Meeting. South Dakota Department of Transportation, Pierre, Nov. 14-16, 1989.

The authors are solely responsible for the contents of this paper, and the views expressed do not necessarily reflect the views of the research sponsors.



APP ARE: Personal Computer Software for Automated Pavement Profile Analysis and Roughness Evaluation

J. JIM ZHU AND RAJEEV NAYAR

Profilographs are widely used for characterization, specification, and quality control of initial pavement roughness during highway construction. Pavement roughness is indicated by a profile index (PI). The PI is usually evaluated manually from the profilogram, which is a strip chart of profile traces, using a blanking band profile index (BBPI) algorithm. The manual BBPI algorithm is laborious, subjective, and prone to operator errors; thus results are highly unreliable and unrepeatable. A new personal computer software package, APPARE (automated pavement profile analysis and roughness evaluation), currently being developed at the Louisiana State University using the profilogram and other types of digitized pavement profile data, is reported. APP ARE has an interactive graphical user interface and an image-processing. engine capable of digitizing profilograms using commercially available, low-cost desktop scanners; it evaluates the PI and other widely used roughness indexes, such as international roughness index, using digitized pavement profile data from any profiling and roughness measuring instrument. In particular, a finetuned computer BBPI algorithm is developed and successfully implemented.

A smooth road not only provides a comfortable ride for its users but also reduces vehicle wear and tear, improves fuel efficiency and safety of travel, and prolongs the life span of the pavement. Two main aspects of pavement smoothness that concern highway engineers and management authorities are (a) smoothness of newly constructed pavements and (b) performance of the entire highway system (1). Controlling the initial roughness during construction can significantly improve the life cycle of the road, consequently greatly reducing the cost of maintenance (J). The 1986 AASHTO Guide for Design of Pavement Structures emphasizes the need for initial pavement smoothness as an important design consideration (2).

Profilographs are widely used instruments for characterization, specification, and quality control of initial pavement roughness during highway construction. Commonly used profilographs (e.g., the Rainhart and California type) generate strip charts called profilograms. Pavement roughness is usually evaluated manually from the profilograms using a blanking band profile index (BBPI) algorithm to derive a profile index (PI) (3). This process is known as trace reduction. Consensus appears to be that the manual BBPI algorithm is laborious, subjective, and prone to operator errors (1,4-6). Consequently, the results are highly unreliable and unrepeatable. For instance, it has been reported that manual calculation of the PI using the BBPI could vary as much as 65

Remote Sensing and Image Processing Laboratory, College of Engineering, Louisiana State University, Baton Rouge, La. 70803.

m/km (4 in./mi) from one operator to another (4-7). In particular, it was found previously (7) that the PI of a sample of 19 profilograms evaluated by 23 different operators, 8 were considered experienced operators, and 2 computer algorithms had an average standard deviation some 52 percent of the mean PI; that is, there is a 32 percent chance that an average operator/computer will produce a PI that deviates from the "true" PI by more than ± 50 percent.

To improve objectivity and repeatability, attempts have been made to computerize the profilograph data acquisition and trace reduction process ( 4-8). However, those previous attempts reportedly have been prone to either under- or overestimating the PI (5-8). In one study the problem was attributed to the difficulty in selecting an appropriate (linear) filtering algorithm in profile trace reduction (8).

In addition to the difficulties in PI evaluation, it is well known that the profilographs severely distort the "frequency components" in the pavement profile (4,9-12), and the PI correlates poorly with other widely used roughness indexes, such as the international roughness index (IRI) (4,10,13). Consequently, TRB has recommended further study and evaluation of the profile trace (profilogram) produced by the profilograph (1).

The present paper together with two other accompanying papers (14,15) constitute an ongoing research effort on further study of profilographs sponsored by the Louisiana Department of Transportation and Development/Louisiana Transportation Research Center (LDOTOLLTRC), in cooperation with FHWA, U.S. Department of Transportation. The problem of and remedies for poor correlation of the PI with IRI and other widely used roughness indexes are addressed in the accompanying papers, so in this paper a new personal computer software package, APP ARE (for automated pavement profile analysis and roughness evaluation), currently being developed at the Louisiana State University, is discussed.

APP ARE has an interactive graphical user interface and an image-processing engine capable of digitizing profilograms using commercially available, low-cost desktop scanners, displaying, editing, and vectorizing raster images and evaluating various roughness indexes using digitized pavement profile data from a profilograph and other profiling instruments. Corrective filtering algorithms are implemented for statistically recovering the pavement profile from distorted profilogram data or data of other profiling devices to ensure the validity and repeatability of the PI. A fine-tuned computer BBPI algorithm was developed and successfully implemented. On test data from 27 road sections, a highly linear correlation

Zhu and Nayar

was obtained with an R2 of 0.991 between the Pis evaluated by APP ARE and by manual evaluation with strictly controlled precision. In addition, the evaluation of IRI and other widely used roughness indexes are facilitated for easy correlation analysis of the pavement roughness. Future development of this software includes the following: capabilities for power spectral analysis of pavement profiles, quarter-car simulation, and other tools for pavement surface and rideability analysis.

DEVELOPMENT AND FUNCTIONALITY OF APP ARE

APP ARE is a self-contained computer program developed to analyze pavement roughness from road elevation profiles. It contains three functional modules, as shown in Figure 1, to automate the process of (a) digitizing profile measurements ~f necessary, (b) statistically compensating for pavement profile distortions caused by profile measuring equipment, and ( c) calculating and mathematically correlating (converting) the various commonly used roughness indexes. These functionalities are discussed in detail in the following subsections.

· Profilogram Digitizer

Because of some reported instances of unsuccessful implementation of computerized profilographs (5 ,6,8), profilographs using strip chart recorders are still widely in use in many states, including Louisiana. To automate the profile trace reduction procedure, the first step is to convert the graphical profilogram (a strip chart) into a numerical format for computer processing. In APPARE this digitizing process is achieved through a three-step procedure: (a) scanning the profilogram, (b) editing the scanned image if necessary, and (c) extracting the digitized profile trace from the scanned image. Salient features of this procedure include the use of

FIGURE 1 Functional flow chart of APPARE.

53

commercially available, inexpensive desktop scanners, an interactive graphical editor, and a nonlinear midpoint profile extraction filter with a moving slope threshold for noise reduction.

Scanning Profilogram

A profilogram is usually a multipage fan-fold continuous form [216 x 280 mm (8.5 x 11 in.)]. For California-type profilograph using a 1:1 vertical scale and a 1:300 (1 in. to 25 ft) longitudinal scale, a 0.322-km (0.2-mi) profilogram consists of at least 4, sometimes 5, pages. These pages can be scanned using a desktop digital scanner either continuously through one pass or page by page. The scanned image is called a raster image, which consists of an array of white and black dots known as pixels. At 12 dots/mm (300 dots/in.) scanning resolution and using the above profilogram scaling ratio, the scanned profile trace has a vertical resolution of 1112 mm (1/ 300 in.) and a longitudinal resolution of 25.4 mm (1 in.). The scanned raster image can then be stored on a computer disk in some standard raster graphical image file format such as PCX format, TIFF format, and so on. The scanning and processing routines in APP ARE can handle both single-page continuous-form and multipage scanning and can support the PCX format now and the TIFF format in the near future.

Shown in Figure 2 is a screen snapshot of the scanned profile trace displayed by the Microsoft (MS) DOS version of APP ARE. Figure 3 shows the screen display of the MS Windows version of APP ARE.

Graphical Editor

Although the scanned images are usually of very high quality, at times they need some touch-up to remove spurious dots and streaks that do not belong to the profile trace. Moreover, sometimes it is difficult for the computer to determine where the profile trace actually starts and ends, and sometimes it may be desirable to select a section of the scanned profile trace to process; in such cases, manually selecting the start

FIGURE 2 Screen display of DOS version of APPARE: scanned data (black) overlaid with extracted trace (white).

54

FIGURE 3 Screen display of MS Windows version of APPARE: scanned data (black) overlaid with extracted trace (white).

and end points is necessary. To facilitate such operations, an interactive graphical editor is implemented in APPARE. The editor displays the scanned profile image and provides the user with a cross-hair cursor for selecting the start and end points, an eraser for removing spurious data points, and annotating tools for writing notes on the margins of a scanned profile image.

All changes made to a scanned profilogram are saved in a hidden file, which is attached to the graphical data file for the profilogram. This feature enables the operator to recover from editing mistakes or the supervisor, who is provided with a privileged password key to the hidden file, to reexamine the changes.

Profile Trace Extraction

The profile trace produced by the profilograph strip chart recorder is not an ideal thin curve because of pavement textures, dirt and rocks on the road, mechanical vibrations, and simply the thickness of the recording pen. Thus, for manual profile trace reduction, the profile trace needs to be "averaged by drawing a line through the vertical midpoint of the trace created by the profilograph using a pen of contrasting color" (3). This procedure apparently introduces some arbitrariness into the manual trace reduction procedure. On computerized profilographs, this appears to be one of the causes for a linear digital filter to produce unsatisfactory profile reduction results (6,8).

In APPARE, this "midpoint extraction" procedure is performed on the scanned raster image to obtain the digitized vector image, that is, a single-valued function described by x-y coordinates, of the profile trace. In addition to performing the midpoint extraction, the program has to deal with imperfect scanned images such as (a) spurious noisy dots that

· are very close to the profile trace and (b) missing data pointsthat is, a gap in the scanned profile trace-that might be as small as 1/12 mm (11300 in.) but large enough to confuse and halt the entire midpoint extraction process.


The midpoint extraction filter handles the first problem by using a moving slope threshold, based on the assumption that the slope of the road from one data point to the next cannot exceed a certain bound. The default value of this threshold is set to ± 45 degrees in APP ARE, which yields satisfactory results; otherwise it can be set by the user. Once the threshold is determined, the midpoint at the current location is found within the upper and lower slope limits projected from the previous data points, as illustrated in Figure 4. One exception is at the start point, where no previous data point is available, the slope limits are projected from a fictitious previous point at the same elevation of the start point.

To cope with the problem of missing data points within the slope threshold, a cubic polynomial extrapolation algorithm is used to estimate the missing data point using the previous four data points.

The extracted profile trace is then saved into a disk file for further process. Screen snapshots of extracted profile traces overlaid on the scanned profile traces are shown in Figures 2 and 3.

Power Spectral Compensation

It is well known that the profilograph severely distorts frequency contents of the pavement elevation profile because of its periodic, bandpass frequency response and infinitely many transmission zeros ( 4 ,11 ,12 ,15). As a consequence, profilograms cannot be used to reconstruct the original pavement elevation profile (15). However, in an earlier paper (14) a mathematical model for the pavement roughness is proposed in which the roughness profile is described by an ergodic, Gaussian, and white stochastic signal (sequence), possibly superimposed with certain deterministic features. As indicated by the kinematic model for the profilograph (4,11,12,15), the profilogram consists of a linear combination (a weighted

Elevation P( x) Spik~

• • • • • • •

• • •

• • •

Noise

Missing Data / Extrapolation cef

• •

Moving Slope Threshold

Longitudinal Distance

FIGURE 4 Trace extraction from scanned profilogram.

x

Zhu and Nayar

sum) of shifted input. Because the power spectral density (PSD) function of a white stochastic sequence is a constant, the profilograph can be used to characterize some stochastic features of the pavement roughness. It is, in fact, this latter property that makes the profilograph a valid profiling device (15).

To obtain a faithful statistical characterization of the road roughness, it is vital that the PSD of the profile be retained. Although it is known that the kinematics of the profilograph alters the PSD of the pavement roughness profile according to its spatial frequency response-provided that the PSD is white (15), all post data recording and processing devices, such as the pen plotter on a conventional profilograph, digital counter and filters on an automated profilograph, and even the operation speed of the profilograph, also alter the profile PSD to a certain extent (15). These problems can be compensated by a linear shift-invariant filter.

Let p(x) and q(x) be, respectively, the pavement profile and the measured profile through some (linear shift-invariant) profiling device having a frequency response Tiw). Then the power spectral density functions (PSDF) of p(x) and q(x) are related by

(1)

Clearly, the measured profile is distorted by the profiling device unless ITP(w)I = 1, which is unrealistic. Now suppose that a compensating filter having a frequency response Tc( w) can be designed such that

(2)

within the frequency band of interest w, ::; w ::; wh. Connect this compensating filter in series with the profiling device as shown in Figure 5 and denote by p(x) and '11.v.iw) the output and its PSDF, respectively. Then the PSDF of original profile can be recovered within the frequency band w1 ::; w ::; wh

(3)

Note that Equation 3 does not imply p(x) = p(x) at every longitudinal position x because p(x) and p(x) are two random spatial signals, and Equation 2 cannot be realized for all frequencies. However, Equation 3 ensures that the two random signals, p(x) and p(x), have the same statistics within the frequency range w1 ::; w::; wh, and these statistics can be used to characterize the roughness of the original and the recovered profile.

In addition to the profile distortions caused by the kinematics and dynamics of data recording and processing devices,

.~~ p(x)~ .p(x)

Profiling Device

p(x) =Pavement profile

Compensating Filter

q(x) =Measured profile with distortion

p( x) = Corrected profile

FIGURE 5 Block diagram for profile compensating filter.

55

extraneous signals may be introduced during the profiling process to corrupt the profile PSD. It is found that the eccentricity of the profiling wheel of a profilograph introduces a power concentration as high as 20 db at the frequency corresponding to the circumference of the wheel (15), and·misalignment in the scanning process causes excessive power concentration in the low-frequency spectrum. In APP ARE, the former problem is effectively coped with by using a notch filter, and the latter is corrected by using a linear regression filter.

Detailed analysis and discussions for power spectral compensation are presented in other work (14,15). The profilograph compensating filter developed by Zhu et al. (15) is implemented in APP ARE and applied to the extracted profile trace to improve validity of the Pl. The principle of using a spectral compensation can be applied to any profiling instrument with linear shift-invariant dynamics. Effective design algorithms exist for designing such digital compensating filters, which are readily implemented into APP ARE to pr0<;ess digitized profile data from computerized profilographs and other types of profiling instrument.

Roughness Index Evaluation and Conversion

The last functional component of APP ARE is for roughness index evaluation and conversion. In particular, an automated BBPI algorithm is developed and implemented with satisfactory results. The IRI algorithm is also implemented. Other widely used roughness indexes are either being implemented or will be implemented using empirical or mathematical correlations.

The implementation of the BBPI algorithm follows closely the procedures adopted by the LDOTD/L TRC, as specified previously (3) for manual PI evaluation for the Californiatype profilograph. Because the goal is to obtain a computer implementation of the BBPI procedure, no PSD compensation is used here. The algorithm accumulates the counts of excursions beyond a ± 2.5-mm (0.1-in.) blanking band weighted by their amplitude in increments of 1.25 mm (0.05 in.), then divides the total weighted counts by the total distance over which the profile is taken. This algorithm, although conceptually simple, turned out to be a nontrivial task. The main difficulty was to distinguish a deviation from one of the following situations according to LTRC (3):

1. "Spiked projections caused by the profilograph rolling over rocks or dirt on the pavement" are to be discounted;

2. "Small portions of the average profile trace ... visible outside the opaque blanking band, ... unless these projections extend 0.75 mm (0.03 inch) or more vertically and 2 mm (0.08 inch) or more longitudinally on the profile trace ... " are to be discounted;

3. "Only the highest peak of a double-peaked scallop is to be counted;" and

4. "Scallops that fall at the end of one position of the blanking band in such a manner that they fall into two positionings of the blanking band are to be counted only once."

To accommodate these ambiguous cases, a significance test is set up in the program. Only those excursions in which the

56

area enclosed by the profile trace and a blanking band exceeds a certain threshold value are counted for the Pl calculation. By fine tuning this threshold value to 130 mm2 (0.2 in. 2

), an excellent correlation between the automated and manually evaluated PI was obtained on data from 27 test sites with an R2 of 0.991. The test data are shown in Table 1 and Figure 6. The following comments are appropriate:

1. The poor reliability and repeatability of the manual BBPI procedure because of operator error can be attributed to two main factors: (a) operator negligence and (b) illusion in human visual perception. The former can be greatly reduced by carefully controlling the PI evaluation procedure, which was the case during these tests. The discrepancy shown in Figure 6 between the manually evaluated PI and APP ARE-evaluated PI is mainly attributable to illusion. It is clearly seen in Figure


6 that the manually evaluated Pis for smooth roads where PI::::; 78 mm/km (5 in./mi) tend to be overestimated, whereas those for rough roads tend to be underestimated. This discrepancy is because human perception relies heavily on comparison with the environment, which affects the operator's judgment on the magnitude of excursions in the ambiguous cases. To appreciate this conclusion, interested readers are invited to first evaluate the Pis for the two (fictitious) profilograms shown in Figure 7 and then use a ruler to draw the 1.25-mm (0.05 in.) elevation lines between the given 2.5 mm (0.1 in.) elevation lines, reevaluate the Pis, and compare with the first evaluation.

2. Ambiguities in the BBPI evaluation other than the aforementioned four cases also exist and are considered by various highway construction agencies. For instance, the Central Federal Lands Highway Division of FHWA, Denver, also con-

TABLE 1 Computation and Correlation Results for Profile Index

LOCATION SITE.# PI CPI (inch/mile) (inch/mile)

LA22 SB-OL Hm 1 1.00 0.5 LA22 SB-OL Hm 2 4.25 3.59 LA22 SB-OL Hm 3 1.75 1.53 LA@@ NB-OL Hm 4 0.50 0.51 LA@@ NB-OL Hm 5 2.50 2.56 LA@@ NB-OL Hm 6 1.25 1.02 LA44 NB-OL JCP 7 19.55 19.07 LA44 NB-OLJCP 8 14.75 15.04 LA44 SB-OL JCP 9 31.00 30.21 LA44 SB-OL JCP 10 25.20 25.49 Burbank SB-OL JCP 11 6.60 7.25 Burbank NB-OL JCP 12 10.31 10.88 I-49 NB-OL JCP 13 0.00 0 I-49 NB-OL JCP 14 2.25 1.53 I-49 NB-OLJCP 15 3.25 2.04 I-49 NB-OLJCP 16 2.75 1.54 I-49 NB-OL JCP 17 5.00 3.32 I-49 NB-OL JCP 18 2.75 2.3 I-49 NB-OLJCP 19 1.00 1.03 I-49 NB-OLJCP 20 2.00 1.02 I-49 NB-OL JCP 21 5.25 5.97 I-49 NB-OLJCP 22 2.00 1.52 I-49 NB-OL JCP 23 0.25 0.25 I-49 NB-OL JCP 24 0.25 0.25 I-49 NB-OL JCP 25 9.14 8.89 I-49 NB-OL JCP 26 8.10 6.01 I-49 NB-OL JCP 27 5.07 2.94

Linear Correlation of the PI and CPI:

Regression Parameters: Regression Results: Constant Std Err of Y Est R Squared No. of Observations Degrees of Freedom X Coefficient(s) Std Err of Coef.

PI Manually calculated Profile Index CPI Computer calculated Profile Index Profile Index unit conversion: 1 inch/mile = 15.6 mm/km

-0.454 0.770 0.991

27.000 25.000

1.005 0.019

Note

1 lar~e Bump

1 large Bump

Zhu and Nayar

35

30 ~

~ 0 25 c:

:::,.

a: i 20 -as "'5 0 15 ~

~ a. 10 E 8

5

0

lh._

v v

v /

/. ~ 0 5 10 15 20 25 30

Manually calculated Pl (inch/mile)

FIGURE 6 Correlation of computer calculated PI and manually calculated PI (see Table 1 for correlation results).

35

siders the situation in which "a scallop is incomplete at the end of a blanking band position but does not continue into the subsequent position." Thus, the definition and judgment of such ambiguities constitute another major factor for the poor reliability and repeatability of the BBPI procedure. Clearly, this problem cannot be resolved by computerized BBPI algorithms. In this regard, a new statistical profile index algorithm described in a revised paper of Zhu (14), which computes the mean deviation (MD) of the profile trace from the blanking band and is therefore called MDPI, appears to be a superior alternative.

In a recent study (7) on the accuracy and variability of the profilogram trace reduction, the PI evaluated by APP ARE

0.3

o.o

-0.3

FIGURE 7 Demonstration of illusion in manual Pl evaluation.

57

compared favorably with the PI evaluated by 23 operators, among whom 8 were considered experienced operators, and by two commercially available computer programs.


APP ARE is a newly developed software package for automated pavement profile analysis and roughness evaluation. It has a user-friendly graphical user interface and self-contained profile analysis functionalities. Salient features of the software include (a) a desktop scanner-based, midpoint profile trace extraction profilogram digitizer; (b) power spectral compensating filter for profile corrections; and ( c) successful computer implementation of the BBPI algorithm. The standard IRI algorithm is also implemented in APPARE, along with other widely used roughness indexes.

The basic functionalities of APP ARE are now fully operational. In the future, facilities to handle the power spectral analysis of pavement profiles, rideability analysis, and simulations and other commonly used roughness index algorithms will be developed and implemented in the software.

ACKNOWLEDGMENTS

The work described in this paper was funded by the Highway Planning and Research Program of FHWA, U.S. Department of Transportation, and LDOTD/LTRC, under an LDOTD research contract task order. The authors sincerely appreciate this support. The profilograph data used in this research were provided by L TRC. In particular, the authors gratefully acknowledge Steven L. Cumbaa of LTRC for his inspiration, close cooperation, and expeditious assistance. The authors sincerely thank the reviewers of this paper for their valuable comments. Special thanks are due to Jerry Budwig and John P. Penzien of FHWA, Denver, and Larry A. Scofield of Arizona Transportation Research Center for their enthusiastic interest and support of this research project and their many constructive suggestions for improving APPARE.

REFERENCES

1. Woodstrom, J. H. NCHRP Report 167: Measurements, Specifications, and Achievement of Smoothness for Pavement Construction. TRB, National Research Council, Washington, D.C., Nov. 1990.

2. Guide for Design of Pavement Structures. AASHTO, Washington, D.C., 1986.

3. Use and Care of the California Profilograph. Employee Training Manual L0201A. Louisiana Transportation Research Center, Baton Rouge.

4. Kulakowski, B. T., and J.C. Wambold. Development of Procedures for the Calibration of Profilographs. Report FHWA-RD-89-110. Pennsylvania Transportation Institute, University Park, Aug. 1989.

5. Cumbaa, S. L. Road Profile Study. Report FHWA/LA-86/185. Louisiana Transportation Research Center, Baton Rouge, Feb. 1986.

6. Scofield, L. A., S. A. Kalavela, and M. R. Anderson. Evaluation of the California Profilograph. In Transportation Research Record 1348, TRB, National Research Council, Washington, D.C., 1992.

7. Budwig, J. L. Profilograph Trace Reduction Study. Presented at 72nd Annual Meeting of the Transportation Research Board, Washington, D.C., Jan. 1993.

58

8. Huft, D. L. Analysis and Recommendations Concerning Profilograph Measurements on F0081(50)107 Kingsbury County. In Transportation Research Record 1348, TRB, National Research Council, Washington, D.C. 1992.

9. Woodstrom, J. H. The California Profilograph. Presented at ASTM Symposium on Measurement, Control, and Correction of Pavement Roughness in Construction, Phoenix, Ariz., Dec. 1982.

10. Walker, R. S., and H. T. Lin. Profilograph Correlation Study with Present Serviceability Index (PSI). Report 569-lF, FHWADP-88-072-002. Texas State Department of Highways and Public Transportation, Austin, March 1988.

11. Hankins, K. D. Construction Control Profilograph Principles. Research Report 49-1. Texas Highway Department, Austin, June 1967.

12. Kulakowski, B. T. and C. Lin. Effect of Design Parameters on Performance of Road Profilographs. In Transportation Research Record 1311, TRB, National Research Council, Washington, D.C., 1991, pp. 9-14.


13. Sayers, M. W. Profiles of Roughness. In Transportation Research Record 1260, TRB, National Research Council, Washington, D.C., 1990, pp. 106-111.

14. Zhu, J. J. Mathematical Characterization of Pavement Roughness. Preprint. Presented at 72nd Annual Meeting of the Transportation Research Board, Washington, D.C., 1993.

15. Zhu, J. J., R. Nayar, S. Rangarajan, and J.-H. Lan. Characterization and Improvement of Profilograph Using a Spatial Signal Processing Approach. Final Report, Contract 736-17-0101. Louisiana Department of Transportation/Louisiana Transportation Research Center, Baton Rouge; FHWA, U.S. Department of Transportation, Oct. 1993.

Publication of this paper sponsored by Committee on Surface PropertiesVehicle Interaction.

Factors Affecting Repeatability of Pavement Longitudinal Profile Measurements

l<HALED KSAIBATI, SANJAY ASNANI, AND THOMAS M. ADKINS

When looking at the accuracy of profilometers, most agencies are mainly concerned with hardware precision rather than the errors caused by the human operators or environmental factors. The Wyoming Transportation Department and the University of Wyoming conducted a joint research project to determine the effect of these two factors on the accuracy and repeatability of roughness and rut depth measurements. The Wyoming Transportation Department's road profiler, which is a duplicate of the South Dakota road profiler, was used in this study. A total of 36 test sections were tested by three different operators to determine the effect of human factors on measurement repeatability. In addition, a concrete test section was monitored and tested several times in the 1991 testing season to examine the effect of various combinations of environmental factors on the measured roughness. The data collected were then tabulated and statistically analyzed. The design of the experiment is summarized, the data that were collected are described, and specific conclusions with regard to the effect of human and environmental factors on the accuracy of roughness and rut depth measurements are discussed.

One of the primary operating characteristics of a road, whether paved or unpaved, is the level of service that it provides to its users. In turn, the variation of this level of service or serviceability with time provides one measure of the road's performance. This performance can be quantified by calculating pavement serviceability index (PSI) on the basis of roughness measurements.

Surface roughness of any pavement can be defined simply as the vertical surface undulations that affect the vehicle operating costs and the riding quality of that pavement as perceived by the user. Immediately after pavements are laid, deterioration starts as a result of continuous dynamic traffic loads and several environmental factors. Road surfaces start developing cracks, potholes, ruts, and so on. As road surfaces become rougher and if maintenance is not performed in a timely manner, roads will become uncomfortable to their users.

In the past few decades, roughness response devices were the primary instruments for estimating the roughness of a roadway section. However, several drawbacks involved in the use of such instruments made them unpopular, and the need was felt to develop a more effective way to measure roughness.

Profilometers were designed to measure the actual pavement profile instead of a vehicle's response to the profile.

K. Ksaibati and S. Asnani, Department of Civil Engineering, University of Wyoming, P.O. Box 3295, University Station, Laramie, Wyo. 82071. T.M. Adkins, Wyoming Department of Transportation, P.O. Box 1708, Laramie, Wyo. 82002.

Measurements obtained with profilometers are essentially independent of the test vehicle's suspension characteristics. Approximately 15 different types of road profilometers are in existence throughout the world. The first modern profilometer was developed in the early 1960s at the General Motors Corporation Research Laboratories (GMR) (J). The GMR profilometer, a contact-type device, used a high-quality potentiometer with several accelerometers to measure the road profiles. Since then several noncontact sensors were introduced to the market. K.J. Law Engineers, Inc., utilized the noncontact light beam measuring system in the 690 digital noncontact profilometers (2 ,3). In England, the Transportation Road Research Laboratory developed a high-speed, laser-based profilometer in the late 1970s. The South Dakota Department of Transportation (SDDOT) developed a profilometer that utilized ultrasonic (acoustic) sensors (2-4). This equipment, referred to as a road profiler, operates at highway speeds and measures pavement profiles only in the left wheelpath. SDDOT shared the road profiler technology with several other highway agencies. The demand for road profilers has become so great that they are now manufactured commercially. Today, 8 states have duplicated the road profiler, and about 25 others have bought commercially built systems. Two factors encouraged the fast spread of this technology:

1. FHW A requires that pavement roughness be reported in international roughness index (IRI) units.

2. The road profiler is relatively low cost compared with other available technologies.

Although quantifying roughness from pavement profiles proved to be much more accurate and reliable than depending on the point response of a vehicle, certain factors must be addressed when dealing with the measurements of pavement longitudinal profiles:

1. Effect of human operators on accuracy and repeatability of road profiler measurements;

2. Effect of environmental variations on pavement profiles; and

3. Importance of road profiler calibration.

The Wyoming Transportation Department and the University of Wyoming conducted a joint research project to examine the effect of these factors. The findings from the first two factors are discussed in this paper. The importance of calibration is discussed in detail by Asnani et al. in another paper in this Record.

60

BACKGROUND

When an agency is considering the purchase of a profilometer, factors related to the hardware accuracy normally are considered. Other important factors such as the effect of human operator on measurement repeatability or the effect of fluctuations in environmental factors on changing road profiles seldom are taken into account.

Operators' ability and experience can be one of the major factors contributing to the inaccuracy of the collected roughness data. The human profilometer operator has a limited ability to concentrate on the job of profiling. The ability to concentrate is somewhat time dependent. The operator will probably do a better job testing short control sections, where the required attention span is short, than longer inventory sections. Also, every operator has a particular style of driving and reaction to particular situations. For example, if an operator is familiar with the profile of the section being tested, he or she may tend to avoid driving over rough spots by deliberately swerving to the left or the right. This type of behavior will result in inaccuracies in measuring longitudinal road profiles. Thus, the very fact that a human is required to operate the profiling equipment may limit the accuracy and repeatability of the profilometer data.

Variations in environmental conditions can also have a significant impact on pavement longitudinal profiles. Road profile characteristics can change significantly as a result of the daily cycle of heating and cooling, seasonal cycles of heating and cooling, and wetting and drying. As an example, excess rainfall will change the moisture conditions in the subgrade and the pavement layers. Variation in water content may cause shrinkage or swelling of subgrade soils, contributing to change in the profile pattern of a pavement. Also, wide variations in temperature may cause the profile of a concrete pavement to change. During the day, the top of the pavement slab heats under the sunlight while the bottom of the slab remains relatively cooler. The maximum difference in temperature between the top and bottom of the pavement slab may occur sometime after noon. This may cause the slab to warp or bend downward, developing stresses (See Figure 1, top). Late in the evening, there may be reversal of warping stresses because of the heat transfer from top to bottom, ma~ing the top surface colder than the bottom surface (See


Figure 1, bottom). Seasonal variation in temperature may also contribute to the change in road profile of concrete sections. During summer, as the mean temperature of the slab increases, the concrete pavement expands. As the slab tends to expand, compressive stress is developed at its bottom. Similarly, during winter the slab contracts, causing tensile stresses at the bottom (5 ,6). If the profile of a road changes from day to day and season to season, it raises the question about the value of acquiring highly accurate and repeatable profilometer data.

Lack of calibration among presently existing profiling systems may also lead to noncomparable data collected by various states across the United States. Research was recently completed by Asnani et al. (and is reported in this Record) to investigate the effect of lack of calibration among some of the existing systems. Eleven road profilers participated in that experiment in which IRI and rut depth data were collected on eight pavement test sections. The major findings of that experiment were as follows:

1. Roughness and rut depth measurements obtained with any single system are repeatable.

2. Most roughness and rut depth measurements with all the systems are statistically different, but there exist strong relationships among the systems. This indicates the need for calibrating road profilers against each other.

DESIGN OF EXPERIMENT

A detailed plan was prepared to determine the effect of human factors and the environmental variations on the accuracy of pavement longitudinal profile measurements. This testing plan involved the creation of two data bases. The first data set was used to examine the effect of human operators on the accuracy of profile measurements, whereas the second data set was used to determine the magnitude of changes in pavement profiles (roughness) catised by changes in environmental factors. Figure 2 shows the data collection and analysis strategies for this experiment. The road profiler of the Wyoming Transportation Department was used to measure the roughness of all test sections included in the experiment.

Surface Temperature Rising

Surface Temperature Falling

FIGURE 1 Temperature effects on concrete slabs: top, surface temperature is higher than temperature at bottom of slab; bottom, surface temperature is lower than temperature at bottom of slab.

Ksaibati et al.

ROAD PROFILERS CALIBRATION

FACTORS AFFECTING REPEATABILITY OF ROAD PROFILE MEASUREMENTS

EFFECT OF HUMAN OPERATORS

DATA BASE PREPARATION

STATISTICAL ANALYSIS

SELECTING 3 DRIVERS

SELECTING 36 TEST SECTIONS


FIGURE 2 Data collection and analysis strategies.

To examine the effect of human operator on the repeatability of road profiler measurements, 36 sections were included in the experiment. A total of 27 pavements were flexible and 9 were rigid. The sections were selected to represent all possible ranges of roughness and rut depth values. These ranges were as follows:

•Low IRI: 0 :s IRI :s 2.0 mmlm •Medium IRI: 2.0 mm/m < IRI :s 3.0 mm/m •High IRI: 3.0 mm/m < IRI

• Low rut depth: 0 :s rut depth :s 2.54 mm • Medium rut depth: 2.54 < rut depth :s 6.35 mm •High rut depth: 6.35 mm < rut depth

61

The test sections were located on I-25, SR-96, and SR-211 in the southeastern corner of Wyoming. Table 1 presents the testing matrix for this experiment. Three operators were selected to operate the road profiler. The regular operator who normally conducts the routine inventory testing for the Wyoming Transportation Department was included in this study. The other two operators had no prior experience in driving the road profiler. Each operator drove the road profiler three times on each test section. The operators were not told the exact locations of test sections. Instead, they were asked to cover long test segments on different highways. This was done to simulate regular field operating conditions when the operators are collecting routine data for inventory purposes. After the data on all the sections were collected, IRI and rut depth measurements for test sections 0.2 mi (0.12 km) long were extracted from the long segments. The means, standard deviations, and coefficients of variations of IRI and rut depth observations were then calculated. Tables 2, 3, and 4 summarize these values for the flexible and rigid test sections.

To examine the effect of environmental factors on pavement longitudinal profiles, one test section was monitored in 1991 for 3 consecutive months. This test section was located on a stretch of I-25 4 mi (2.5 km) long between Mileposts 13.8 and 16.2. The wearing surface of the test section consisted of a 9-in. (23-cm) jointed unreinforced portland cement concrete underlain by 6 in. (15.2 cm) of crushed gravel. Roughness data were collected on the test section under various combinations of environmental conditions, such as

1. 24-hr rainfall, in millimeters; 2. 72-hr rainfall, in millimeters; 3. Ground temperature at bottom of the slab, in degrees

Celsius;

TABLE 1 Locations of Test Sections Used to Evaluate Operators' Effect on Roughness Measurement Accuracy

PAVEMENT TYPE

FLEXIBLE RIGID

PERFORMANCE INDEX PERFORMANCE INDEX

ROAD PROFILER PROJECT IRI IRI

LOW MEDIUM HIGH

RUT RUT RUT L M H

L M H L M H L M H

25N 25N 25N 25N 25N 25N 211 25N 96W 25N 25N 25S 1 MP MP MP MP MP MP MP MP MP MP MP MP

5.0 6.2 18.0 9.7 14.9 14.3 39.0 15.1 2.1 11.3 11.0 9.5

25N 25N 25N 25N 25N 25N 211 25N 96W 25N 25N 25S

SECTIONS 2 MP MP MP MP MP MP MP MP MP MP MP MP 4.1 6.5 18.3 9.2 16.3 14.6 40.2 15.9 1.7 11.5 12.4 10.3

25N 25N 25N 25N 25N W96 211 25N 96W 25N 25N 25S 3 MP MP MP MP MP MP MP MP MP MP MP MP

9.4 7.0 18.6 4.0 28.3 1.0 40.9 16.1 0.1. 11.7 12.9 12.1


TABLE 2 Means and Standard Deviations of IRI Values for Flexible Test Sections

LOW IRI RUT

LOW MEDIUM HIGH SECTIONS

1 2 3 1 2 3 1 2 3

AVG. 2.27 1.52 1.39 1.50 1.63 1.69 1.97 2.09 1.51 1 S.D. 0.27 0.11 0.13 0.17 0.14 0.13 0.05 0.15 0.09

c.v. 11.89 7.24 9.35 11.33 8.59 7.69 2.54 7.18 5.96

AVG. 2.08 1.44 1. 74 1. 51 1.61 1.54 1.68 2.08 1.21 DRIVERS 2 S.D. 0.16 0.07 0.30 0.05 0.16 0.14 0.06 0.11 0.05

c.v. 7.69 4.86 17.24 3.31 9.94 9.09 3.57 5.29 4.13

AVG. 2.24 1.45 1.53 1.69 1.87 1. 74 1.84 2.05 1.36 3 S.D. 0.12 0.10 0.18 0.08 0.15 0.12 0.11 0.15 0.06

c.v. 5.36 6.90 11. 76 4.73 8.02 6.90 5.98 7.32 4.41

TABLE 3 Means and Standard Deviations of IRI Values for Rigid Test Sections

LOW IRI SECTION

1 2 3

AVG. 2.12 1.86 1.81

1 S.D. 0.09 0.08 0.11 c.v. 4.25 4.30 6.08

AVG. 2.31 2.06 1.99

DRIVERS 2 S.D. 0.10 0.28 0.21 c.v. 4.33 13.59 10.55

AVG. 2.26 1.90 1. 79

3 S.D. 0.17 0.06 0.03 c.v. 7.52 3.16 1.68

4. Average daily air temperature, in degrees Celsius; and 5. Change in 24-hr air temperature, in degrees Celsius.

Table 5 summarizes all roughness and environmental data collected on the test section.

DATA ANALYSIS

All collected data were reduced and compiled in computer files. Data analysis was later conducted by using regular statistical tools. The main objectives of the analysis were to

MEDIUM IRI HIGH IRI

SECTION SECTION

1 2 3 1 2 3

2.08 2.34 2.42 2.33 2.80 3.10

0.12 0.13 0.1.6 0.04 0.21 0.01

5.77 5.56 6.61 1. 72 7.50 0.32

2.24 2.37 2.50 2.44 2.83 3.30

0.10 0.07 0.06 0.17 0.16 0.06

4.46 2.95 2.40 6.97 5.65 1.82

2.06 2.20 2.55 2.26 3.00 3.26

0.10 0.08 0.09 0.10 0.43 0.11

4.85 3.64 3.53 4.42 14.33 3.37

investigate the repeatability of roughness and rut depth measurements obtained by each operator, compare the results obtained from three operators, and study the effect of environmental factors on pavement roughness.

Repeatability of Roughness Measurements by Each Operator

Each operator drove the road profiler 3 times on all 36 test sections. The averages, standard deviations, and coefficients of variation were then calculated for IRI and rut depth data

Ksaibati et al. 63

TABLE 4 Means and Standard Deviations of Rut Depth Values for Flexible Test Sections

LOW IRI RUT

LOW MEDIUM HIGH SECTIONS

1 2 3 1 2 3 1 2 3 AVG. 0.10 0.10 0.08 0.15 0.13 0.10 0.33 0.46 0.52

1 S.D. 0.04 0.01 0.01 0.06 0.04 0.06 0.11 0.02 0.03 c.v. 40.00 10.00 12.50 40.00 30.77 60.00 33.33 4.35 5.77

AVG. 0.13 0.11 0.09 0.11 0.13 0.07 0.40 0.47 0.49 DRIVERS 2 S.D. 0.05 0.01 o.oo 0.02 0.05 0.04 0.06 0.02 0.07

·c.v. 38.46 9.09 o.oo 18.18 38.46 57.14 15.00 4.26 14.29

AVG. 0.09 0.07 0.09 0.19 0.21 0.17 0.39 0.45 0.52 3 S.D. 0.07 0.02 o.oo 0.01 0.02 0.02 0.06 0.02 0.02

c.v. 77.78 28.57 0.00 5.26 9.52 11. 76 15.38 4.44 3.85

TABLE 5 Data Collected for IRI and Other Environmental Factors

TEST IRI GROUND AVERAGE CHANGE IN TOTAL 24- TOTAL 72- CHANGE IN NO. (mm/m) TEMPERATURE. DAILY AIR 24-HOUR AIR HOUR RAIN HOUR RAIN AIR v GROUND

(°C) TEMPERATURE TEMPERATURE (mm) (mm) TEMPERATURE (°C)

1 2.76 13 13

2 2.75 12 13

3 3.04 15 11

4 2.76 14 15

5 2.72 20 19

6 2.70 21 22

7 2.75 22 21

8 2.68 21 16.

9 2.71 22 21

10 2.81 24 22

11 2.69 16 21

12 2.77 20 22

on all test sections (see Tables 2 through 4). The coefficient of variation, the ratio of standard deviation to the mean expressed as a percent, is normally used to measure the relative variability of any factor. In this analysis, the coefficient of variation for IRI ranged from 0.32 to 14.33 on concrete sections and from 0.92 to 17.74 on bituminous sections. These coefficients of variation indicate acceptable variability of IRI measurements. In other words, IRI measurements obtained by any operator were repeatable. On the other hand, the coefficients of variation for rut depth measurements ranged from 0 to 77. 78, indicating high relative variability for rut depth measurements.

(°C) (oC)

+2.8 0.00 2.03 0.6

0 o.oo 3.64 1.1

-2.2 18.54 73.15 3.9

-1. 7 0.00 5.33 0.6

+1.1 0.00 0.51 1.1

+3.9 o.oo 1. 27 1.1

+0.6 2.30 15.55 1.1

-1.1 0.00 o.oo 5.0

-0.6 3.30 3.81 0.6

+1. 7 0.00 4.06 2.2

+2.2 0.00 4.06 5.0

+1.1 0.00 o.oo 1. 7

Comparison Among Three Operators

Pavement longitudinal profiles obtained by the three drivers were first plotted and compared visually. Figure 3 shows some of these profiles on a selected test section. Because no definite conclusions could be obtained by the visual comparison, IRI and rut depth measurements were calculated and averaged on each test section. The two-sample t-test was then used to conduct paired comparisons between the means. Basically, average measurements from any two operators were compared to determine whether they were statistically different at 95 percent confidence level.


Cl) .

.L:. ()

c 100 ft. ~~ H =~~'Q~a~~~-vb~ I

6. oo+ ~o 6. oo+o. 40

] k: f t .

~~ 0

6.00+0.20 6.00+0.40

FIGURE 3 Profile of test section between Mileposts 6.2 and 6.4 for Run 3 by Drivers 2 (top) and 3 (bottom).

The t-statistic used in the analysis was calculated with the following equation:

where

sample means, sample sizes (three in this case), estimate of common variance computed with the following equation:

(n1 - l)Si + (n2 1)~ s~ = ....:........;;..._~-'--"°--~:.......::.~---'-=

n 1 + n2 - 2

and

Si, ~ = two individual sample variances.

The calculated t-value was compared with ta12 ,n1 + nz -2 = 2.776 (for ex = 0.05 and 4 degrees of freedom). If ABS (t) > ta12 ,n

1 +n

2_ 2 , it would be concluded that the two means are

statistically different. Using this two-sample t-test, a large number of paired com

parisons were conducted on IRI and rut depth data. Measurements obtained with each operator were compared with measurements from the other two operators on all 36 te~t sections. The results of the statistical analysis are summarized in Tables 6 and 7 for IRI and rut depth data, respectively. Table 6 indicates that the IRI measurements obtained with the three operators were equal in all cases except five. It is interesting that three of the five cases were on flexible sections with low roughness levels. On the other hand, Table 7 shows how the disagreement among operators was much higher when dealing with rut depth measurements. In this case, more differences were detected on sections with high roughness level.

Effect of Environmental Factors on Pavement Roughness

The environmental data collected on the concrete test section were analyzed statistically. The main objectives of the analysis were first to determine which environmental factors cause changes in pavement profiles and second to develop a regres-

sion relationship that can predict IRI on the basis of these important factors. The following regression model was initially used:

where

Y; = value of response variable IRI; X1 , X 2 , X 3 = independent variables (environmental fac

tors such as temperature and rain); and B0 , B 1 , B3 = regression constants.

On the basis of the regression model, relationships were established by using the MINITAB software package. All factors were linearly correlated with IRI, and the resulting R-squares were examined. None of the linear models seemed to fit adequately. Graphs were then drawn to determine the general shape of the relationship between each environmental factor and IRI. The relationship between IRI and the variation in air temperature during 24 hr is shown in Figure 4. It is clear from this figure that a nonlinear rather than a linear relationship should be established between these two factors. After considering this fact, the following regression model was obtained with R2 = 0.849:

IRI = 2.72 + O.ll7A + 0.00357B - 0.00065B2

where

IRI international roughness index; A 72-:-hr rainfall before testing; and B change in 24-hr air temperature.

This relationship indicates clearly that IRI is influenced by environmental factors. Specifically, the higher the amount of rain falling on the section within 72 hr before testing the higher the measured IRI value. Also, the roughness (IRI) of a concrete section will vary depending on air temperature fluctuation before testing. Some other relationships were developed with the factor 24-hr rainfall. However, these relationships produced a lower R 2

•

CONCLUSIONS

In this research, an attempt was made to identify the effect of human and environmental factors on the accuracy of pave-

TABLE 6 Results from !RI-Paired Comparisons

PAVEMENT TYPE

FLEXIBLE RIGID

PERFORMANCE INDEX PERFORMANCE INDEX

IRI

ROAD PROFILER PROJECT LOW MEDIUM

RUT RUT

L M H L M H L

1 E" E E E E E

DRIVERS SECTIONS 2 E E E E E E E (1) AND (2) .. 3 E E E E E E

1 E E E E E E E

DRIVERS SECTIONS 2 E E E E E E E

(1) AND (3) 3 E E E Bl E E E • 1 E E E E E E

DRIVERS SECTIONS 2 E E E E E E E

(2) AND (3) 3 E E E E E E E

• E: IRI DATA OBTAINED WITH RESPECTIVE DRIVERS ARE STATISTICALLY EQUAL. ••NE: IRI DATA OBTAINED WITH RESPECTIVE DRIVERS ARE STATISTICALLY DIFFERENT.

TABLE 7 Results from Rut Depth-Paired Comparisons

PAVEMENT TYPE

FLEXIBLE

IRI

HIGH L M

RUT

M H

E E E E

E E E E

E E E E

E E E E

E E E E

E E E E

E E E E

E E E E

E E E E

PERFORMANCE INDEX

IRI

ROAD PROFILER PROJECT LOW MEDIUM HIGH

RUT RUT RUT

L M H L M H L M H

1 E' E E E E E Iii E E

DRIVERS SECTIONS 2 E E E E E E E E E (1) AND (2)

:-:N~::::: 3 E E E E E E E E

1 E E E E E E ~ E .ltl2 DRIVERS SECTIONS 2 E ;:~~!/ E E ~ E

E II E (1) AND (3)

3 E E E E E E E E

1 E \"~ij/: E E E E E E El .. ·. ·=':':

::::.:~~-:.:11 DRIVERS SECTIONS 2 ·::.:::: :.-:·· E E E E E E E

(2) AND (3) :• .. ·.· ... · - ::

3 E /NE). E E E > ·.·. ~:::="=::" E

• E: RUT DEPTH DATA OBTAINED WITH RESPECTIVE DRIVERS ARE STATISTICALLY EQUAL. ••NE: RUT DEPTH DATA OBTAINED WITH RESPECTIVE DRIVERS ARE STATISTICALLY DIFFERENT.

H

E

E

Ill E

E

E

E

E

E

66

3.10

~2.90 E

.§.

~

2.70

2.50 TT"T" -2.50 -0.50 1.50 3.50

CHANGE IN 24-HOUR AIR TEMPERATURE (C)

FIGURE 4 IRI versus change in 24-hr air temperature.

ment roughness and rut depth measurements. An extensive testing program was performed. The collected data were then reduced, tabulated, and analyzed statistically. This analysis leads to the following conclusions:

1. When considering measurements obtained by any single road profiler operator, the coefficient of variation of rut measurements is much higher than the coefficient of variation of roughness measurements. In other words, the roughness measuring capability of the road profiler is much better than its rut depth-measuring capability.

2. The t-test results indicate that roughness measurements obtained by the three operators were statistically equal in all but five cases. Three of these five cases were on sections with low roughness level. These results indicate· that road profiler operators should give more attention when measuring roughness of smooth pavements. On the other hand, rut depth measurements obtained by different operators were statistically different in 20 percent of the cases. More differences were detected on sections with a high roughness level where it is harder for the operator to drive in the wheelpaths.


3. The regression analysis yielded a good nonlinear relationship between IRI and two environmental factors. R 2 for this relationship was almost 85 percent, which indicates that pavement roughness does fluctuate as a result of changes in environmental conditions.

ACKNOWLEDGMENTS

This cooperative study was funded by the U.S. Department of Transportation's University Transportation Program through the Mountain-Plains Consortium, the Wyoming Transportation Department, and the University of Wyoming.

REFERENCES

1. Spangler, E. B. Inertial Profilometer Uses in the Pavement Management Process. In Transportation Research Record 893, TRB, National Research Council, Washington, D.C., 1982.

2. Carmicheal, R. W. State-of-the-Practice of Roughness and Profile Measuring Technology. Proc., 2nd North American Conference on Managing Pavements, Vol. 3, 1987.

3. Carmicheal, R. W. Automated Pavement Data Collection Equipment, Roughness and Profile Measurement. Demonstration Project 72, FHWA-DP-72-1. FHWA, U.S. Department of Transportation, Sept. 1986.

4. Huft, D. L. Analysis of Errors for the South Dakota Profilometer. In Transportation Research Record 1000, TRB, National Research Council, Washington, D.C., 1984.

5. Khanna, S. K., and C. E.G. Justo. Highway Engineering, 7th ed. Nern Chand and Brothers, Roorkee (U.P.), India, 1991.

6. Yoder, E. J., and M. W. Witczak. Principals of Pavement Design, 2nd ed. John Wiley and Sons, Inc., New York, 1975.

The authors are solely responsible for the contents of this paper, and the views expressed do not necessarily reflect the views of the research sponsors.



Automated Versus Manual Profilograph Correlation

CARL B. BERTRAND

The Texas Department of Transportation (TxDOT) has _attempted to correlate the outputs of the _automated Cox profilograph, the automated McCracken profilogr~ph and a TxDOT manual McCracken profilograph. The evaluation process was precipitated by calls from construction engineers ~ithin _th~ TxDOT highway agency and paving contractors workmg w1t~m Texas. Both the state and contractor personnel were requestmg the use of the automated profilograph. The results of the evaluation process were as follows. The Cox automated profilograph used a filter setting number of 5, which represents the attenuation of 2.2 ft (0.067 m) and less, whereas the McCracken mo~el used a data filter cutoff frequency of 2.5 ft (0.76 m). The profilogra~s from the TxDOT manual profilograph were reduced by two ~1fferent interpreters. Both of the automated ve_rsions of t~e profilograph were slightly more repeatable than the mterpretatlon of the manual profilograph. The automated profilographs showed very ~lose correlation with the manual profilograph on the smooth, medmm, and rough sections of asphalt concrete pavement and on the rough sections of continuously reinforced concrete (CRC) paveme~t. The automated profilographs deviated from the manual profilograph output on the smooth-section CRC pavement. This deviation was from 0.5 to 2.0 (0.789 to 3.16 cm) PI counts smoother (lower) than the output of the manual profilograph

During the past few years several state highway authorities have reported problems when an automated California type of profilograph was used to determine contractor payment on pavement construction projects (J). The state highway authorities owned a manual California-type profilograph, whereas the contractor was allowed to use an automated (computerized) California-type profilograph. The contractor's automated profilograph was used daily to determine the bonus or penalty payments for the paving operation. When the state authority came back to verify the results with their manual profilograph, the resulting profile index (PI) was significantly different. The difference was always in the contractor's favor-for example, lower PI values. This situation led FHWA to restrict the use of automated California-type profilographs for use in determining contractor payment on federally funded paving projects in FHWA Region 6 (2). Automated profilographs can be used only after it has been demonstrated and documented that they yield Pis within 0.5 in./mi (0.789 cm/ km) of a standard manual profilograph.

This paper details an attempt by TxDOT to correlate the outputs of the automated Cox profilograph, the automated McCracken profilograph, and a TxDOT manual McCracken profilograph. The evaluation process was precipitated by calls from construction engineers within TxDOT and paving con-

Pavements Section, Pavement Design Division, Texas Department of Transportation, 125 East 11th Street, Austin, Tex. 78701.

tractors working in Texas. Both the state and contractor personnel were requesting the use of the automated profilograph. ·

This correlation effort was not an attempt to evaluate the various filter settings on the automated versions of the profilographs. Some of the problems with the software filter settings (J,3) as well as the frequency response of the 12-wheel profilographs ( 4) have been studied and well documented. The manufacturers, Cox and Sons and McCracken Pipe Co., were asked to provide instruments and software with the suggested filter settings. The filter settings were supposed to provide the best correlation with the manual interpretation of a profilogram. These manufacturers suggested filter settings that were maintained throughout the testing process. Also, each automated profilograph manufacturer provided a representative who was present during the testing process and operated their individual profilographs.

SCOPE

A series of comparative tests was performed using an automated Cox, an automated McCracken, and a TxDOT manual McCracken California-type profilograph. The goal of this correlation process was to determine whether the manufacturers' suggested filter settings used for the reduction of automated profilograph data yields the same results as the filter settings on the manual profilograms collected on the same pavement sections. Recently several states have reported that the use of the automated version of the profilograph has given significantly lower PI values than did the manual version of the profilograph data.

The testing procedures used for this correlation effort are presented along with descriptions of the selected test sites. The correlations presented in this paper represent only the resulting PI values from the test sections. The bump responses and bump locations from the various instruments on the test sections were not compared.

A brief description of the automated profilograph software settings is presented. The manual profilograph data reduction procedures followed by TxDOT personnel are .describe~. A description of the correlation analysis used for this companson testing is presented. Finally, a set of conclusions with the associated recommendations is presented.

TESTING PROCEDURE

The following testing procedure was used to determine the correlation between the Cox and McCracken automated pro-

68

filographs and the TxDOT standard, the McCracken manual profilograph. This testing procedure was modeled after the procedures of FHW A's Highway Performance Monitoring System Field Manual (5). These procedures were intended to obtain correlations between high-resolution profiling devices (Class 1 or 2 instruments) and response-type road roughness measuring systems (Class 3 instruments).

The two· automated profilograph manufacturers provided TxDOT personnel with their suggested filter values. Representatives of each manufacturer were present during the comparison testing and operated their individual profilographs. The representatives were asked to verify that their machines were in horizontal and vertical calibration before testing. Personnel from TxDOT Maintenance and Operations Division, Pavement Management Section (D-18PM), measured a 500-ft (0.153-km) horizontal calibration site that was used by all three instruments to check and adjust, if necessary, the distance calibration. A set of gauge blocks shipped with the automated Cox profilograph were used in the vertical calibration determination. The profiling tire inflation pressure was checked daily and adjusted if necessary. The horizontal calibration was performed once before the collection of any data. Each instrument was disassembled and taken by trailer to each of the three pavement-type locations. The vertical calibration was checked before data collection-·after the instruments were transported to a new test location. D-18PM personnel verified that all three instruments were in calibration.

The TxDOT manual profilograph was operated by several different TxDOT D-18PM personnel during the testing sequence. All operators were experienced in the proper use of the profilograph before the testing. A single operator pushed the manual profilograph on all three runs of an individual test site. All testing was accomplished in the span of 2 days. The profilographs were pushed one after the other on all sections. Each profilograph completed the required three runs on. each test section before moving to the next section. Variations as a result of multiple operators of the manual profilographs were not statistically considered in this effort. These variations have been documented in other studies (3 ,6). At the end of each day's testing, the profilograms from all three instruments were collected by D-18PM personnel. The data reduction and comparisons were accomplished after all of the testing was completed.

TEST SITE DESCRIPTION

Before the arrival of the automated profilographs, D-18PM personnel located several test sections that were used in the correlation effort. All selected test sites were on in-service pavements. An attempt was made to locate both continuous reinforced concrete (CRC) pavement and jointed concrete pavement (JCP) sections within close proximity to D-18PM headquarters in Austin, Texas. Unfortunately, the locations of the JCP sites were all city streets and exhibited very large PI values. Therefore, only one JCP section, Harris Branch Parkway, was used in this correlation effort. Four CRC pavement test sections were located on the State Highway 71 bypass built around thd city of La Grange, Texas. Six asphaltic


concrete pavement (ACP) sections were located on Southwest Parkway within the Austin, Texas, city limits.

Each test site was 0.1 mi (0.1609 km) long. Only one wheelpath in each travel lane was profiled. The outside wheel path of the outside travel lane was used on the ACP and· JCP sections, whereas the inside most wheelpath of the inside travel lane was used on the CRC pavement sections. The decision about which wheelpath and travel lane to use on each group of test sections was driven by safety concerns, because sections were on in-service pavements. The beginning and end and each section's identification were marked with traffic paint. A series of painted dots along each selected wheelpath was used as a guide for each instrument operator to follow. Each instrument was to make three runs on each test site.

TxDOT's surface dynamics (SD) profilometer and a profilograph simulation program developed by Roger Walker of the University of Texas at Arlington (7) were used to determine candidate sections with the appropriate ranges of roughness. Three roughness levels in PI terms were targeted for the correlations. Smooth sections ranged from Pis of 0 to 7 in./mi (0 to 11.06 cm/km); medium sections ranged from Pis of greater than 7 to 15 in./mi (11.06 to 23.7 cm/km); and rough sections had Pis greater than 15 in./mi (>23.7 cm/km). There were four smooth, three medium, and three rough sections in the test matrix. The JCP section exhibited an estimated PI of over 80 in./mi (126.38 cm/km).

Each test site was given a unique designation code to prevent mistakes during data reduction and site misidentification by the operators during data collection. The designation code for each test site was painted on the pavement surface at the beginning of each section. The ACP sections on Southwest Parkway were designated SWPOl through SWP06. The CRC pavement sections on State Highway 71 outside of La Grange were designated LAGOl through LAG04, and the JCP site on Harris Branch Parkway was designated HBROl. The operators of each profilograph were instructed to identify each profilogram, in the case of the manual profilograph, and each header name, in the case of the automated profilograph, with the appropriate site designation followed by a 01, 02, or 03. These numerics designate the run number for each of the three required runs for each section. The test date and the test operator's name were also recorded on each run.

AUTOMATED PROFILOGRAPH SOFTWARE DESCRIPTION

The software in both the automated profilographs was set to use a 0.2-in. (0.508-cm) blanking band, a 0.3-in. (0.76-cm) bump height, and a 25.0-ft (7.62-m) bump width. These parameters were used to reduce the manual profilograph data and are specified in the Texas Test Method Tex-1000-S (8) procedures for reducing profilograms. The automated Cox profilograph used Filter Setting 5, which represents the attenuation of wavelengths of up to 2.2 ft (0.67 m). The automated McCracken profilograph used a data filter cutoff frequency setting of2.5 ft (0.76 m). Both of these filter settings were used throughout the comparison testing sequence. These were the filter settings that the manufacturers suggested using for the correlations. Both manufacturers use a third-order

Bertrand

Butterworth filter. In addition, a null band filter setting on the automated Cox profilograph is used to help reduce data through short radius curves and super elevations. The null band filter can be used only by setting the null band filter switch to the filter position. The fixed-distance position eliminates the use of the null band filter in the data reduction. The fixed distance position was used throughout the testing sequence.

MANUAL DATA REDUCTION

The data from the TxDOT manual McCracken profilograph were reduced by two experienced interpreters. The profilograms were separated into two rolls. The three runs on each of the six ACP sections were reduced by Interpreter 1, whereas the runs from the four CRC pavement sections and the one JCP section were reduced by Interpreter 2. After interpreters completed their respective sections and calculated the resulting Pis, they switched rolls. Both interpreters were told to disregard the other person's markings on the profilograms, independently align the blanking band scale, and calculate the resulting PI values.

RESULTS

All the raw data from both interpreters of the manual profilographs and the two automated profilographs were entered into a spread sheet. Table 1 presents the spread sheet that contains the raw data along with a few preliminary statistics. The run number column is used to identify individual runs (1-3) on each test section for the manual interpreters and each automated profilograph. The post number column is used as the x-axis on several subsequent graphs and is continuously numbered 1 through 33. This represents the total number of runs made for the comparison testing. The section identification (ID) column indicates the location of the test section, as well as the individual test sections within each location. The four CRC pavement sections are designated as LAGOl through LAG04 and correspond to Sections 1 through 4, respectively. The six ACP sections are designated as SWPOl through SWP06 and correspond to Sections 5 through 10, respectively. The JCP section is designated HBROl and corresponds to Section 11.

Repeatability

The repeatability of the individual instruments was determined using the standard deviations of the three runs on each test section. Figure 1 illustrates the standard deviations by test section for both manual interpreters and for both of the automated profilographs. Figure 1 also indicates the sections by pavement type. Figures 2 through 5 present four graphs of the standard deviations for the two interpreters and the two automated instruments. From the standard deviations it can be generally stated that both automated profilographs are more repeatable than the manual interpretation of the same profilogram. This might have been expected because one of

69

the advantages of using the automated profilograph is the elimination of the subjectivity in the data reduction process.

General Observations

Figure 6 shows the plots of the PI values from each run on each test section for all instruments. The large Pl values obtained on the JCP section cause the scale of the y-axis to be rather large. The range of the y-axis scale causes the automated and manual profilographs to appear to correlate well. Because the scale used in Texas for the bonus and penalty payments of newly constructed pavements sets Pis greater than 15.0 in./mi (23. 7 cm/km) as the cutoff for accepting a pavement, it was decided to eliminate the JCP section from most of the following statistical and graphical comparisons. Another reason to eliminate the JCP section is that the test matrix has no data points between Pis of approximately 25 in./mi (39.49 cm/km) and the JCP Pis of 80 to 100 in./mi (126.38 to 157.98 cm/km). It would not be acceptable to assume that the response of the instruments or their data filters are linear through this region. Looking at the automated McCracken data for the JCP section it would appear that its data filter may not be linear through this range of roughness. This can be observed by looking at the close agreement on the JCP section between both interpreters and the automated Cox. The automated McCracken yields PI values that are 10 in./mi (15.8 cm/km) less than the other instruments.

Figures 7 through 9 provide three bar charts that represent the raw data differences between the manual interpreters and the automated profilographs. The Pl values from the automated instruments were always subtracted from those from the manual instruments. This process indicates that for the majority of the test runs, the manual interpreters were higher than the automated profilographs, hence the positive differences. Figure 7 shows the differences between the manual interpreters and indicates that Interpreter 1 is consistently higher than Interpreter 2. In general, it can be seen that Interpreter 2 is closer to the automated profilograph PI values.

Reduced Y-Axis Scale Observations

Figure 10 demonstrates a plot for all the PI data from the instruments with the JCP section data eliminated from the data set. This plot yields a more accurate representation of the correlation between the interpreters and the automated instruments over the region of interest. As can be observed from Figure 10, the automated profilographs correlate with the interpreters at least as well as the interpreters correlate with themselves. This observation is generally true except on the three smoothest sections of the CRC pavement represented by Posts 1through6 and 10 through 12. The differences between the automated profilographs and the manual interpreters on these sections vary with the interpreter. It can be seen that the automated profilographs are reading between 0.5 (0.789 cm/km) and 2 (3.16 cm/km) PI values less than either interpreter. The automated McCracken profilograph did have a negative difference from both of the interpreters on the first two runs of section LAGOl. This difference was

TABLE 1 Summary of Profilograph Correlation Data

SECTION INFORMATION MANUAL DIFF %DIFF

ISECTID RUN# POST# PVT TYPE NTER #1 INTER #2 (#1-#2) DIFF/#1

l..AG01 1 1 CRC p,75 (9.08) 4.50 (7.11) 1.25 21.74

2 2 p.25 (8.29) 4.50 (7.11) 0.75 14.29

3 3 p.50 (10.27) 5.75 (9.08) 0.75 11.54

AVE p.83 (9.22) 4.92 (7.77) 0.92 15.85

STD DEV b.51 059

l..AG02 1 4 CRC 11.00 (11.06) 7.00 (11.06) 0.00 0.00

2 5 ~.00 (14.22) 8.25 (13.03) 0.75 8.33

3 6 7.75 (12.24) 7.75 (12.24) 0.00 0.00

AVE 17.92 (12.51) 7.67 (12.11) 0.25 3.16

STD DEV b.82 0.51

~G03 1 7 CRC 23.75 (37.52) 21.25 (33.57) 2.50 10.53

2 8 "9.75 (31.20) 21 .00 (33.18) -1.25 -6.33

3 9 26.45 (41.79) 22.25 (35.15) 4.20 15.88

AVE 23.32 (36.84) 21.50 (33.97) 1.82 7.79

STD DEV 2.75 0.54

l..AG04 1 10 CRC 13. 75 (13.82) 8.25 (13.03) 0.50 5.71

2 11 ~.00 (14.22) 8. 75 (13.82) 0.25 2.78

3 12 ~ 1.25 (17.77) 7.50 (11.85) 3.75 33.33

AVE ~.67 (15.27) 8.17 (12.90) 1.50 15.52

STD DEV ~ .12 0.51

ISWP01 1 13 ACP 23.50 (37.12) 24.50 (38. 70) -1.00 -4.26

2 14 '1.50 (33.97) 26.25 (41.47) -4.75 -22.09

3 15 23.50 (37.12) 25.25 (39.89) -1.75 -7.45

AVE 22.83 (36.07) 25.33 (40.02) -2.50 -10.95

STD DEV b.94 0.72

SWP02 1 16 ACP 3.00 (4.74) 3.25 (5.13) -0.25 -8.33

2 17 3.00 (4.74) 2.50 (3.95) 0.50 16.67

3 18 2.75 (4.34) 2.75 (4.34) 0.00 0.00

AVE 2.92 (4.61) 2.83 (4.48) 0.08 2.86

STD DEV D.12 0.31

5WP03 1 19 ACP o.25 (8.29) 4.50 (7.11) 0.75 14.29

2 20 6.00 (9.48) 4.75 (7.50) 1.25 20.83

3 21 5.25 (8.29) 6. 75 (10.66) -1.50 -28.57

AVE o.50 (8.69) 5.33 (8.43) 0.17 3.03

STD DEV D.35 1.01

SWP04 1 22 ACP 10. 75 (16.98) 11.75 (18.56) -1.00 -9.30

2 23 13.00 (20.54) 14.00 (22.12) -1.00 -7.69

3 24 14.00 (22.12) 12.75 (20.14) 1.25 8.93

AVE 12.58 (19.88) 12.83 (20.27) -0.25 -1.99

STD DEV 1.36 0.92

SWP05 1 25 ACP 2.25 (3.55) 1.50 (2.37) 0.75 33.33

2 26 1.50 (2.37) 1.25 (1.97) 0.25 16.67

3 27 2.50 (3.95) 2.00 (3.16) 0.50 20.00

AVE ~.08 (3.29) 1.58 (2.50) 0.50 24.00

STD DEV D.42 0.31

SWPOO 1 28 ACP 10.00 (15.80) 9.00 (14.22) 1.00 10.00

2 29 10.50 (16.59) 10.25 (16.19) 0.25 2.38

3 30 11.25 (17.77) 10.50 (16.59) 0.75 6.67

AVE 10.58 (16.72) 9.92 (15.67) 0.67 6.30

STD DEV D.51 0.66

HBR01 1 31 JCP gg.oo (156.40) 91.50 (144.55) 7.50 7.58

2 32 ~4.50 (149.29) 96.00 (151.66) -1.50 -1.59

3 33 100.5 (158.77) 94.50 (149.29) 6.00 5.97

AVE ~8.00 (154.82) 94.00 (148.50) 4.00 4.08

STD DEV 2.55 1.87


TABLE 1 (continued)

SECTION INFORMATION AUTO DIFF %DIFF DIFF %DIFF ·

SECT ID RUN# POST# PVT TYPE MCCRACK (#1-MCC) DIFF/#1 (#2-MCC) DIFF/#2

LAG01 1 1 CRC M0(9.48) -0.25 -4.35 -1.50 -33.33

2 2 ~.50 (8.69) -0.25 -4.76 -1.00 -22.22

3 3 14.00 (6.32) 2.50 38.46 1.75 30.43

AVE 5.17 (8.16) 0.67 9.78 -0.25 -8.37

STD DEV J.85

LAG02 1 4 CRC ~.00 (6.32) 3.00 42.86 3.00 42.86

2 5 5.50 (8.69) 3.50 38.89 2.75 33.33

3 6 17.50 (11.85) 0.25 3.23 0.25 3.23

AVE 5.67 (8.95) 2.25 28.32 2.00 26.47

STD DEV n .43

LAG03 1 7 CRC ~0.00 (31.60) 3.75 15.79 1.25 5.88

2 8 ~.50 (32.39) -0.75 -3.80 0.50 2.38

3 9 0.00

AVE 20.25 (31.99) 1.50 6.00 0.88 2.75

STD DEV D.25

LAG04 1 10 CRC 7.00 (11 .06) 1.75 20.00 1.25 15.15

2 11 15.50 (10.27) 2.50 27.78 2.25 25.71

3 12 7.00 (11.06) 4.25 37.78 0.50 6.67

AVE 6.83 (10.80) 2.83 28.52 1.33 15.84

STD DEV D.24

SWP01 1 13 ACP 24.00 (37.91) -0.50 -2.13 0.50 2.04

2 14 24.00 (37.91) -2.50 -11.63 2.25 8.57

3 15 22.50 (35.55) 1.00 4.26 2.75 10.89

AVE 23.50 (37.12) -0.67 -3.17 1.83 7.17

STD DEV J.71

$WP02 1 16 ACP 2.50 (3.95) 0.50 16.67 0.75 23.08

2 17 1.50 (2.37) 1.50 50.00 1.00 40.00

3 18 2.50 (3.95) 0.25 9.09 0.25 9.09

AVE 2.17 (3.42) 0.75 25.25 0.67 24.06

STD DEV J.47

$WP03 1 19 ACP 6.50 (10.27) -1.25 -23.81· -2.00 -44.44

2 20 14.50 (7.11) 1.50 25.00 0.25 5.26

3 21 5.50 (10.27) -1.25 -23.81 0.25 3.70

AVE :>.83 (9.22) -0.33 -7.54 -0.50 -11.83

STD DEV P.94

SWP04 1 22 ACP ~2.50 (19.75) -1.75 -16.28 -0.75 -6.38

2 23 H2.50 (19.75) 0.50 3.85 1.50 10.71

3 24 11.50 (18.17) 2.50 17.86 1.25 9.80

AVE 12.17 (19.22) 0.42 1.81 0.67 4.71

STD DEV D.47

SWP05 1 25 ACP 1.50 (2.37) 0.75 33.33 0.00 0.00

2 26 2.00 (3.16) -0.50 -33.33 -0.75 -60.00

3 27 2.50 (3.95) 0.00 0.00 -0.50 -25.00

AVE 2.00 (3.16) 0.08 0.00 -0.42 -28.33

STD DEV D.41

SWP06 1 28 ACP 10.00 (15.80) 0.00 0.00 -1.00 -11.11

2 29 10.00 (15.80) 0.50 4.76 0.25 ·2.44

3 30 9.00 (14.22) 2.25 20.00 1.50 14.29

AVE 9.67 (15.27) 0.92 8.25 0.25 1.87

STD DEV J.47

HBR01 1 31 JCP go_50 (127.17) 18.50 18.69 11.00 12.02

2 32 g2.50 (130.33) 12.00 12.70 13.50 14.06

3 33 gJ,00 (131.12) 1.7.50 17.41 11.50 12.17

AVE g2.00 (129.54) 16.00 16.27 12.00 12.75

STD DEV H.08


TABLE 1 (continued)

SECTION INFORMATION f\UTO DIFF %DIFF DIFF %DIFF

:>ECTID RUN# POST# PVT TYPE ~ox (#1-COX) DIFF/#1 (#2-COX) DIFF/#2

LAG01 1 1 CRC 14.50 (7.11) 1.25 21.74 0.00 0.00

2 2 13.50 (5.53) 1.75 33.33 1.00 22.22

3 3 14.00 (6.32) 2.50 38.46 1.75 30.43

AVE 14.00 (6.32) 1.83 31.43 0.92 18.64

STD DEV b.41

LAG02 1 4 CRC 5.50 (8.69) 1.50 21.43 1.50 21.43

2 5 17.00 (11.06) 2.00 22.22 1.25 15.15

3 6 6.50 (10.27) 1.25 16.13 1.25 16.13

AVE 6.33 (10.01) 1.58 20.00 1.33 17.39

STD DEV b.62

LAG03 1 7 CRC 20.90 (33.02) 2.85 12.00 0.35 1.65

2 8 h 9.90 (31.44) -0.15 -0.76 1.10 5.24

3 9 h 8.90 (29.86) 7.55 28.54 3.35 15.06

AVE 19.90 (31 .44) 3.42 14.65 1.60 7.44

STD DEV ).82

LAG04 1 10 CRC 6.00 {9.48) 2.75 31.43 2.25 27.27

2 11 7.50 (11.85) 1.50 16.67 1.25 14.29

3 12 6.50 (10.27) 4.75 42.22 1.00 13.33

AVE 5.67 (10.53) 3.00 31.03 1.50 18.37

STD DEV ).62

SWP01 1 13 ACP 25.00 (39.49) -1.50 -6.38 -0.50 -2.04

2 14 =>5.00 (39.49) -3.50 -16.28 1.25 4.76

3 15 24.50 (38. 70) -1.00 -4.26 0.75 2.97

AVE 24.83 (39.23) -2.00 -8.76 0.50 1.97

STD DEV ).24

SWP02 1 16 ACP 2.00 (3.16) 1.00 33.33 1.25 38.46

2 17 2.00 (3.16) 1.00 33.33 0.50 20.00

3 18 2.00 (3.16) 0.75 27.27 0.75 27.27

AVE 2.00 (3.16) 0.92 31.43 0.83 29.41

STD DEV b.00

SWP03 1 19 ACP p.00 (7.90) 0.25 4.76 -0.50 -11.11

2 20 14.00 (6.32) 2.00 33.33 0.75 15.79

3 21 14.50 (7.11) 0.75 14.29 2.25 33.33

AVE 14.50 (7.11) 1.00 18.18 0.83 15.62

STD DEV b.41

SWP04 1 22 ACP ~2.00 (18.96) -1.25 -11.63 -0.25 -2.13

2 23 ~ 1.50 (18.17) 1.50 11.54 2.50 17.86

3 24 ~1.50 (18.17) 2.50 17.86 1.25 9.80

AVE h1.67(18.43) 0.92 7.28 1.17 9.09

STD DEV b.24

SWP05 1 25 ACP 2.50 (3.95) -0.25 -11.11 -1.00 -66.67

2 26 2.00 (3.16) -0.50 -33.33 -0.75 -60.00

3 27 2.00 (3.16) 0.50 20.00 0.00 0.00

AVE 2.17 (3.42) -0.08 -4.00 -0.58 -36.84

STD DEV b.24

$WP06 1 28 ACP 9.50 (15.01) 0.50 5.00 -0.50 -5.56

2 29 h1.50 (18.17) -1.00 -9.52 -1.25 -12.20

3 30 9.50 (15.01) 1.75 15.56 1.00 9.52

AVE 10.17 (16.06) 0.42 3.94 -0.25 -2.52

STD DEV b.94

HBR01 1 31 JCP 91 .80 (145.02) 7.20 7.27 -0.30 -0.33

2 32 00.80 (143.44) 3.70 3.92 5.20 5.42

3 33 92.30 (145.81) 8.20 8.16 2.20 2.33

AVE 91.63 (144.76) 6.37 6.50 2.37 2.52

STD DEV P.62

Note: Values 1n parentheses are centimetres/kilometer and conversion factor is 1 inch/mile* 1.58 =cm/km

Bertrand

3.0..,,.---;::======::::::;------;:::======:::::;------;======;i f RC SECTION s 1 - 41 §cP SECTIONS s - 1 o I I JCP SECTION 11 I

2.s-11---------l 1-----------------------1 1-------1

Qj'" e 2.o-11---------l 1-----------------------1 1-------1 c-= z 0 ~ ~ 1.5 0 0

~ z ~ 1.0 en

2 3 4 5 6 7 10 11 SECTION NUMBER

I CJ INTERPRETER #1 ~ INTERPRETER #2 IT§] AUTO McCRACKEN ~ AUTO cox

FIGURE 1 Standard deviations from all instruments on all test sites.

3.0 jcRC SECTION S 1 - 41 jAcP SECTIONS s - 1 o I I JCP SECTION 11

z 0

~

2.5

~ 1.5 0 0 a: ~ ~ 1.0 en

0.5

jAvERAGE STD DEV = 1.041

5 6 7 10 SECTION NUMBER

FIGURE 2 Standard deviations for Interpreter 1 on all test sites.

11

73

3.0 f RC SECTION S 1 - 41 bcP SECTIONS s -1 o I I JCP SECTION 11

!AVERAGE STD DEV= 0.721

z 0

~ ~ 1.5

0 0 a: <( 0 z 1.0 ~ (/)

0.5

4 . 5 6 7 10 11 SECTION NUMBER

FIGURE 3 Standard deviations for Interpreter 2 on all test sites.

3.0 f RC SECTION S 1 - 41 rep SECTIONS 5 - 10 I I JCP SECTION 11

2.5

AVERAGE STD DEV = 0.67 ~ ~ 2.0

= z 0 j:: <(

> 1.5 w 0 0 a: <( 0 z 1.0 ~ (/)

0.5

5 6 7 10 11

SECTION NUMBER

FIGURE 4 Standard deviations for automated McCracken on all test sites.

3.0

2.5

~ ~ 2.0

= z 0

~ Gi 1.5 c c a: ~ ~ 1.0 Cl)

0.5

jcRC SECTION S 1 - 41 f CP SECTIONS 5 - 10 I I JCP SECTION 11

AVERAGE STD DEV= 0.47

3 5 6 7 10 11 SECTION NUMBER

FIGURE 5 Standard deviations for automated Cox on all test sites.

110

100

90

80

~ 70

~ = x 60 w c ~ w 50 ...J u::: 0 a: a.. 40

30

20

10

0 0 10 15 20 25 30

POST NUMBER

1-E- INTEf3 #1 -+-INTER #2 ""°*""" AUTO McCRACKEN -B- AUTO COX

FIGURE 6 Plot of all instruments on all test sites.

35


-0.25 PI on both runs in the case of Interpreter 1 and - 1.5 and -1.0 PI in the case of Interpreter 2. Thereafter, the general trend of positive differences holds. The automated Cox profilograph also showed a 0.0 difference between Interpreter 2 on the first run of section LAGOl.

Regression and Residual Errors

Table 2 provides a series of linear regression equations when the actual data points are plotted, the resulting R2 values, and the sum of the residuals calculated from the resulting residual

~ ·e

20

15

;§. 10 w 0 z w a: w tt 5 i5 a:

1 2 3 4 5 6 7 8 9101112131415161718192021222324252627282930313233 POST NUMBER

INTER#1-INTER#2

FIGURE 7 Interpreter I-Interpreter 2 differences in PI index for all sites.

~ :§.

20

15

:§. 10 w 0 z w a: w tt 5 i5 a:

1 2 3 4 5 6 7 8 9 101112131415161718192021222324252627282930313233 POST NUMBER

FIGURE 8 Differences in PI between both interpreters and automated McCracken.

Bertrand

errors associated with each regression. These regressions were computed without the data points associated with the JCP test section. The firstregression equation, Interpreter 1 versus Interpreter 2, indicates that the fit between the two interpreters has very good correlation with they-intercept, which is slightly negative ( -0.47 in./mi), a slope of 1.015, R2 =

0.95, and that the sum of the residuals is close to 0 (-0.05). Interpreter 2 was generally lower than interpreter 1, which is

20

15

Ji'

l 10 w 0 z w a: w It 5 0 a:

77

verified by they-intercept. The R 2 value of 0.95 is less than that of the remainder of the regressions, whereas the slope is closer to a perfect 1.0. The greatest differences between the two interpreters occur as the PI count of the test section increases. The run numbers with the greatest residual errors generally occur on test sections with the greatest PI count. These observations can be expected because, as the PI count of a section goes up, the differences in the manual interpre-

1 2 3 4 5 6 7 8 9 101112131415161718192021222324252627282930313233 POST NUMBER

FIGURE 9 Differences in PI between both interpreters and automated Cox.

10 15 20 25 30 POST NUMBER

l-7S-- INTER #1 -+-INTER #2 """*-- AUTO McCRACKEN -El- AUTO COX

FIGURE 10 Plot of all instruments on all test sites except JCP site.

78

tation generally will increase because of the subjectivity in the manual interpretations.

The Table 2 regressions for the automated profilographs were performed using only the results from Interpreter 2. Interpreter 2 results generally were closer to the automated results and, therefore, represent the best correlations from the data set. Because the correlation between the two interpreters was so good, it appeared redundant to regress the results against both interpreters. Both sets of regression data and the resulting residual errors indicate that the automated profilographs are reading slightly lower than the manual interpretation, even though the y-intercept from the automated McCracken is positive. The correlation (R2 value) is higher between the automated profilographs and Interpreter 2 than that for the manual interpreters because the automated units have better standard deviations and therefore are more repeatable.

Breakdown by Pavement Type

Tables 3 and 4 contain the regression equations, R2 values, and the sum of the residual errors for the individual pavement types. Table 3 illustrates the results from only the CRC data points, whereas Table 4 shows the results from only the ACP data points. The results of these two tables can be summed up by the following statements:


1. The regression analysis results from the CRC pavement sections are worse than those from the ACP sections. This is the case even when comparing the two manual interpreters against each other.

2. The R2 values from each of the CRC regressions generally are less than those from either Table 2 or 4, and the slopes of the lines generally are further from 1.0.

3. The y-intercepts in Table 3 are all negative, indicating that the automated profilographs are reading lower PI values than both manual interpreters.

CONCLUSIONS

The following general conclusions can be drawn about the correlation between the automated profilographs and the manual interpretation of the profilograms. These conclusions are based on a very limited amount of data and cannot be considered conclusive evidence in quantitative terms with reference to the extent of correlation between automated and manual profilographs. The correlation results indicate that PI values derived by both automated profilographs are close to those derived manually.

1. There is good correlation between both automated models of the profilograph and the manual interpreters. This conclu-

TABLE 2 Regression Results from All Test Sections Except JCP Site

y VALUE X VALUE REGRESSION R2 SUM OF EQUATION VALUE RESIDUALS

INTERP. #2 INTERP. #1 Y=-0.47+1.015(X) 0.95 -0.05

AUTO INTERP. #2 Y=0.079+0.93(X) 0.97 -1. 33 McCRACKEN

AUTO COX INTERP. #2 Y=-0.46+0.97(X) 0.98 -0.743

AUTO AUTO COX Y=0.65+0.93(X) 0.98 0.896 McCRACKEN

TABLE 3 Regression Results from CRC Pavement Sites

Y VALUE X VALUE REGRESSION R2 SUM OF EQUATION VALUE RESIDUALS

INTERP. #2 INTERP. #1 Y=0.077+0.897(X) 0.96 1. 0

AUTO INTERP. #1 Y=-1.3+0.95(X) 0.91 -0.26 McCRACKEN

AUTO INTERP. #2 Y=-0. 71+0. 97 (X) 0.94 -0.05 McCRACKEN

AUTO cox INTERP. #1 Y=-0.84+0.86(X) 0.93 0.16


AUTO AUTO COX Y=0.55+0.95(X) 0.96 0.24 McCRACKEN

Bertrand 79

TABLE 4 Regression Results from ACP Sites

Y VALUE X VALUE REGRESSION R2 SUM OF EQUATION VALUE RESIDUALS

INTERP. #2 INTERP. #1 Y=-0. 87+1. 116 (X) 0.98 o.o AUTO INTERP. #1 Y=-0.38+1.019(X) 0.97 0.12 McCRACKEN

AUTO INTERP. #2 Y=0.44+0.91(X) 0.98 0.19 McCRACKEN


AUTO cox INTERP. ·#2 Y=-0.17+0.97(X) 0.98 o. 77

AUTO AUTO COX Y=0.66+0.93(X) 0.99 -0.26 McCRACKEN

sion must be qualified by the following statement. The individual manufacturers' specified filter settings (5 for the Cox and 2.5 ft for the McCracken) must be used.

2. The correlations between both the interpreters against each other, each interpreter against each automated unit, and both automated units against each other are worse on smooth CRC pavement than on ACP pavements and rough CRC pavements. The tining on CRC pavements may be the cause of the differences observed on the smooth CRC sections. The manual interpreter must subjectively draw a line through the small jagged deviations on the profilogram, thereby smoothing the trace and providing a reference line for obtaining the PI counts. The placement of this line and positioning of the blanking band can be critical in the PI value calculation on smooth pavement sections. The automated profilograph software essentially performs the task of smoothing the trace by applying a filter.

3. No valid conclusions can be drawn from this experiment regarding JCP. Another set of correlation data will need to be collected on some jointed pavement sections on the basis of the results of the smooth CRC correlation.

4. The PI results from the automated type of profilographs appear more repeatable than those from the manual interpretation of the profilogram.

5. The automated profilographs allow the PI values of a pavement section to be instantly available, whereas the manual profilographs req1,1ire that the values be brought from the field and subjectively reduced.

RECOMMENDATIONS

The data reduction filters used by the automated profilographs are software based and could be modified. The highway authority must devise a methodology for determining whether the specified filter is actually in use during pavement testing: This could be accomplished by spot checks with a state-owned profilograph, whether it is manual or automated.

Because the smooth CRC pavement sections appear to be read smoother by the automated profilographs, it could be

that different filters or a lower cutoff frequency of the same filter should be used on these sections. As an alternative to different filter settings, the highway authority could specify a universal filter setting and reduce the bonus scale for CRC pavements when automated profilographs are being used.

A set of procedures needs to be developed for the acceptance of new automated profilograph types that may exist in the future. These procedures need to specify a standard against which to measure and, on the basis of these results, include a spectrum of pavement types with a range of roughness.

REFERENCES

1. Huft, D. L., Analysis and Recommendations Concerning Profilograph Measurements on F0081(50)107 Kingsbury County. In Transportation Research Record 1348, TRB, National Research Council, Washington, D.C., Jan. 1992.

2. Jones, W. C., Use of Computerized Profilographs for Determining Pavement Smoothness. Letter HC-TX. FHWA, U.S. Department of Transportation, Austin; Tex., April 1991.

3. Scofield, L. A., S. A. Kalevela, and M. R. Anderson. Evaluation of the California Profilograph. In Transportation Research Record 1348, TRB, National Research Council, Washington, D.C., Jan. 1992.

4. Kulakowski, B. T., and J.C. Wambold. Development of Procedures for the Calibration of Profilographs. Report FHWA-RD-89-110. FHWA, U.S. Department of Transportation, Aug. 1989.

5. FHWA, Highway Performance Monitoring System Field Manual, Appendix J. FHW A Publication 5600. lA. U.S. Department of Transportation, Dec. 1987.

6. Harrison, R., and C. Bertrand. The Development of Smoothness Specifications for Rigid and Flexible Pavements in Texas. Report FHWA!fX-91+1167-1. FHWA, U.S. Department of Transportation, Jan. 1991.

7. Walker, R. S., and H. T. Lin. Profilograph Correlation Study with Present Serviceability Index. Report FHWA-DP-72-3. FHWA, U.S. Department of 'J'.ransportation, 1987.

8. Operation of Pavement Profilograph and Evaluation of Profiles. Texas Test Method Tex-1000-S. Division of Materials and Tests, Texas Department of Transportation, Austin, Jan. 1992.

Publication of this paper sponsored by Committee on Surface Properties-Vehicle Weight.


Video Cameras for Roadway Surveillance: Technology Review, Test Methods, and Results

CARL ARTHUR MAcCARLEY, DANIEL NEED, AND ROBERT L. NIEMAN

Effective implementation of advanced traffic management strategies depends on timely, reliable, and comprehensive information on traffic conditions. Closed circuit television surveillance of the roadway network is believed to be one of the best mechanisms for providing this information to a traffic operation center. This work supports the application of video surveillance technologies to roadway traffic monitoring. The current state of the art in surveillance camera technology is reviewed. Technical considerations relevant to the selection of video cameras for traffic surveillance applications are summarized. Applicable standards are identified, and evaluation criteria and test procedures are described. A total of 32 commercially available monochrome and color video cameras are evaluated with respect to these criteria .. General considerations and specific test results are reported.

Traffic surveillance is an important part of operational strategies to improve the management of roadways. Recent advances in closed-circuit television (CCTV) technology permit improved monitoring of traffic flow for data collection, traffic management, and incident detection. Closed-circuit video surveillance can serve as a valuable aid to traffic control personnel, extending their effectiveness considerably.

Until now field implementations of CCTV systems have been limited because of both technical limitations and institutional factors. The technical limitations include problems related to the collection and transmission of video images and equipment reliability and maintainability.

Improved technology has overcome many of these problems. Video camera technology has improved substantially in the past few years with the introduction of monolithic silicon photosensor arrays. These advances improve the feasibility of video surveillance as a real-time source of information for traffic operations center (TOC) personnel.

At the request of the California Department of Transportation (Caltrans), technical issues in the selection of video cameras for roadway surveillance were studied. A total of 32 monochrome (black and white) and color video cameras were selected for evaluation on the basis of manufacturers' recommendations of appropriate cameras for traffic surveillance applications. Evaluation criteria that emphasized factors of greatest relevance to roadway surveillance were established. Tests were designed to address these criteria, including laboratory video tests and field test procedures involving human observers.

Electronic and Electrical Engineering Department, California Polytechnic State University, San Luis Obispo, Calif. 93407.

The evaluation was limited to a "snapshot" of the available technology at a particular point in time, specifically, cameras available commercially in 1990. The evaluation considered only the video camera, which is one of many components that constitute a CCTV system. Other components of equal importance include the optics and electromechanical lens controls, video signal transmission network, video amplifiers, multiplexors or switchers, video signal compression equipment, and monitors.

VIDEO CAMERA CHARACTERISTICS AND FEATURES

Before the 1980s, electron tube imaging systems, best exemplified by the Vidicon system, were most common in surveillance caineras. A significant improvement in this technology occurred with the introduction of .charge coupled device (CCD) solid-state imaging integrated circuits (ICs or "chips"). Costs for solid-state cameras have decreased, and quality has improved significantly, such that solid-state or chip cameras have almost completely replaced "tube" cameras in surveillance applications.

Compared with tube cameras, solid-state cameras consume less power, dissipate less heat, can provide excellent resolution, have better geometric linearity and better resistance to flair and bloom (defined later), and are more reliable. Applications in which tube technology is still used are usually those requiring extreme sensitivity.

All 32 cameras tested were considered by their manufacturers as suitable for traffic surveillance, and all used solidstate technology. A range of cost and performance was represented for each product line. As a baseline comparison, one Vidicon camera, representative of the state of the art in approximately 1980, was also evaluated. (Data are not reported for this camera because it was not intended for traffic surveillance.)

Cameras in this class generally provide analog signals with video information content in the range of 0 to 0. 7 V, which equilibrates to 0 to 100 IRE (Institute of Radio Engineers) units.

The spectral response of most monochrome solid-state cameras extends well into the nonvisible infrared (IR) range. Some cameras are provided with removable IR filters to reduce the problems associated with IR sensitivity, such as reporting hot surfaces (such as vehicle tires and black roadway surfaces) as bright objects.

MacCarley et al.

Surveillance cameras generally are available with photographic standard Type C lens mounts, although smaller lens systems are also popular.

Most cameras contain both the imaging array and associated electronics in a common package. However, for discrete surveillance, some cameras incorporate two modules: the main imaging head and the electronics or power supply. Some cameras ·have enclosures that are sufficiently durable and weathertight, whereas others require separate environmental enclosures.

Video cameras designed for surveillance applications differ from those designed for general usage or broadcast. Surveillance cameras are designed for optimum imaging of a stationary field of view, containing a very wide range of light intensities. This requires higher-than-normal resolution and a wide dynamic range (light to dark range). Good sensitivity for best night vision also may be important. Surveillance cameras often are calibrated for· a nearly linear response (a proportional relationship between incident light and the corresponding video signal voltage). It is known that this type of calibration often produces images that are less aesthetically pleasing and somewhat "flat" in appearance. Some cameras use contrast enhancement circuits, which accentuate light-todark or dark-to-light transitions in the image .. This feature has advantages and disadvantages in traffic surveillance applications: vehicle outlines are more crisply defined in low light or fog conditions, but signs and license plates become washed out because of the overshoot.

Traditionally, surveillance-type cameras are monochrome rather than color. Monochrome cameras generally provide greater resolution and sensitivity than color cameras, although several high-resolution color video cameras, specifically designed for surveillance applications, have been introduced recently.

Some of the electronic features that distinguish different video cameras include the following:

Gamma

Most cameras provide either a continuous adjustment or switchselectable setting for gamma. This parameter affects the camera linearity in translating light levels to voltage levels.

White Balance (Color Cameras Only)

A feature that distinguishes various color video cameras is an adjustment for its ability to define the color white, which is an equal mix of all primary colors. Some cameras have automatic white balance capabilities, whereas some have none or only ·manual static adjustments.

Automatic Gain Control and Auto-Iris Control

Automatic gain control (AGC) electronically adjusts the overall camera sensitivity in response to the average light level. This feature has the effect of maintaining a reasonably constant brightness level in the picture. On some cameras, the AGC may be switched off for testing purposes or special applications.

81

The effect of the AGC is similar to that of another feature called an auto-iris, which controls the sensitivity by electromechanical adjustment of the aperture (iris) in response to the average light level. Auto-iris control produces a higherquality image than one controlled by the AGC. However, AGC can respond instantaneously to light level changes, whereas an auto-iris is relatively slow because of the response time of the mechanical components.

Imager Size

CCD cameras typically utilize imaging ICs with diagonally measured imaging surface dimensions of between % and 713 in., 1/z in. being typical. Generally, the larger the chip, the better the image resolution capability, although this also depends on the size of each CCD imaging cell or pixel. Resolution in CCD cameras is directly proportional to the number of pixels on the chip, typically between 200,000 and 400,000. Reducing the pixel size will have a positive effect on the price of the camera because the cost is directly related to the silicon surface area of the chip. Improvements are directly related to developments in IC process technology. The focal length of a lens must be mitched with the imaging chip size to yield the correct field of vi~w.

Shutter Speed

Unless specifically designed for high-speed (slow-motion) photography, mechanical shutters are not used in video cameras. Shuttering is accomplished electronically. Electronics Industry Association (EIA)/National ·Television Standards Committee (NTSC) cameras have an effective shutter speed of less than %0 sec, the rate at which complete video frames are produced (even though they are transmitted as two raster fields at V60 sec each). Some cameras are designed for faster shutter speeds; however, faster speeds reduce camera sensitivity because of reduced photon integration time.

A common use of fast shutter speeds is to avoid smearing when capturing fast-moving objects. For typical camera placements, the motion of roadway traffic in the field of view was not found to warrant faster shutter speeds.

Synchronization

When multiple cameras are integrated into a network, synchronization becomes an issue. If the cameras are not synchronized when switched successively onto the same monitor, picture roll occurs while the monitor is attempting to resynchronize with the frame rate of the new camera. Surveillance cameras are manufactured with one of the following three frame timing control options:

• Internal clock: camera frame rate is unsynchronized, timed independently from an internal clock.

• Phase lock: cameras use the alternating current (AC) line frequency from the power supply for frame synchronization. An initial phase adjustment is usually provided to compensate for phase shift over a large network.

82

•Line lock or external synchronization: an external synchronization generator provides a common-frame synchronization signal to all cameras in the network .

. Cameras using phase lock or external synchronization will switch smoothly without picture roll. Phase synchronization usually is considered only when all cameras are powered from a common AC source. This would be the case for .a surveillance system within one building or within one industrial installation using a common secondary power transformer. However, a surveillance network with cameras spread out over miles of freeway probably would not meet this requirement. The line-lock external synchronizati.on option is technically the superior approach but is more expensive to implement.

APPLICABLE STANDARDS

Several video display and signal formats are in use internationally. The basic frame rate and vertical resolution (number of scan lines) for video signals usually conforms with one of two international standards:

1. EIA of the United States specifies a standard frame rate of 30 full video image frames per second, each frame displayed as two interlaced fields (half resolution frames) at a rate of 60 fields per second. A total of 525 vertical lines of resolution are specified, each field consisting of 262.5 scan lines (1). The color image signal format based on the EIA basic display format is that established by NTSC of the United States. The EIA and NTSC standards are adhered to in the United States, Canada, Mexico, most of South America, and Japan.

2. The International Radio Consultive Committee (CCIR) operates under the auspices of the International Telecommunications Union based in Geneva, Switzerland. The recommendations of the CCIR (1966) permit a variety of color video signal formats, most notably the phase alteration line rate (PAL) standard used throughout most of Europe and the sequential color with memory (SECAM) Standards 1 through 3 used in France and most Eastern Block countries. The basic display format of CCIR-derived formats is 25 frames per second full frame rate, displayed as 50 interlaced fields per second and 625 vertical lines (312.5 per field). Video cameras manufactured for use in Europe generally conformed to CCIR display formats and PAL or SECAM color standards.

Commercial broadcast NTSC, PAL, and SECAM signals usually are allocated approximately a 6-MHz signal bandwidth, compatible with the channel separation of broadcast television in both the United States and Europe. For CCTV systems, this channel capacity limitation does not necessarily exist because the signal does not need to conform with commercial broadcast channel bandwidth restrictions.

Signal bandwidth equilibrates directly to horizontal display resolution expressed in lines, to be discussed later. Commercial broadcast color video signals usually are limited to 200 to 300 lines of horizontal resolution. By comparison, a highquality monochrome CCTV surveillance camera may provide 600 lines of horizontal resolution.


Our laboratory and field test apparatuses were equipped to handle both EIA/NTSC and CCIR/P AL video formats.

CAMERA PERFORMANCE REQUIREMENTS

The performance requirements for surveillance videocameras include consideration of the following criteria (D. Larkins, of the Ampex Corporation, Redwood City, California, provided assistance in compilation in 1990):

System-Level Considerations

1. Information requirements: images should contain sufficient data to support judgments pertaining to traffic control.

2. Surveillance density-images per mile: different monitoring requirements require different image densities. A sparse camera placement density would require greater information content in the image.

3. System cost: the contribution of the video camera to the overall deployed system expense, relative to the surveillance area.

4. Operating environment: a wide range of environmental factors must be considered.

5. System reliability, maintainability, and security: these considerations directly affect the service costs and usefulness of the system.

6. Technology life span-expandability, compatibility, and life: the life span in terms of obsolescence and future availability and maintainability should be considered.

Surveillance Objectives

Camera performance must be adequate to allow the CCTV system to acquire the following data:

1. Traffic flow metrics: vehicle speed, traffic volume, and density determined from visual analysis or computer image processing.

2. Vehicle classification: for roadway utilization data acquisition.

3. Roadway surface conditions-ice, snow, rain, flood, glare, and surface flaws: adverse road surface conditions affecting driver safety.

4. Visibility: roadway visibility as perceived by drivers. 5. Incident detection-collision or stalled vehicle: roadway

incidents such as collision, stalled vehicles, or other situations impeding normal traffic flow.

6. Hazardous or impaired drivers: nonconforming vehicle behavior suggestive of driver impairment.

7. Specific vehicle identification: identification of specific vehicles.

Camera Placement Considerations

1. Effective camera range and field of view: required effective camera range will vary depending on the detection criteria. Use of remote pan, tilt, or zoom may mitigate this requirement.

MacCarley et al.

2. Coverage, redundancy, and overlap requirements: the extent of roadway coverage by a single camera will be reduced on curved roadways and hilly terrain. Overlap or dedicated coverage may be required for isolated areas, for example, tunnels or interchanges.

3. Number of personnel in control room: the number of personnel in the TOC will limit the useful number of camera placements, assuming some maximum number of monitors assigned to each operator.

Environmental Considerations

1. Aesthetic requirements: for minimum public impact, the enclosure and mounting system should be as incongruous as possible.

2. Serviceability: serviceability represents a significant portion of the ongoing system costs. Tradeoffs include minimal maintenance at a higher installed cost versus difficult service at a minimal installed cost.

3. Rain survival and removal: the camera system must be capable of withstanding· rain from all angles and high humidity. Rain droplets that adhere to the foremost optical transmission element could significantly reduce the image. Possible rain removal methods include windshield wipers, spinning windows, forced air deflection, and rain-avoiding enclosures.

4. Snow and ice survival and removal: snowflakes that adhere to the foremost optical transmission element could significantly reduce the image quality from the camera. Ice could also present significant problems with the mechanical components, such as pan and tilt mechanism or zoom lens. Possible snow and ice removal methods include those mentioned for rain removal and the use of a heated front window.

5. High temperature survival: sustained operations at elevated temperatures may be required. Some mechanism for dissipation of external as well as internally generated heat may be necessary.

6. Dust and grime removal and survival: dust and grime reduce light transmission by the front window and may cause scoring of the window or damage to the mechanical components. Some means for automatic lens washing may be an alternative to field service.

7. Ozone and acidic pollution survival: the camera housing must be impervious to the effects of corrosive atmospheric conditions that are present in some urban areas.

8. Spectral filtering: filters may assist in the elimination of certain image artifacts. A polanzing filter may reduce road glare, an IR filter may correct false imaging caused by the IR sensitivity of the camera, and an ultraviolet filter may improve contrast during overcast conditions.

9. Projectile survival: the enclosure may be required to withstand impacts from various projectiles. Outdoor CCTV caineras are often targets of vandalism.

10. Electromagnetic noise immunity: the camera must be sufficiently immune· to the effects of local sources of electromagnetic radiation, such as automotive ignition systems, highpressure vapor lamps, police radar, and mobile citizens band or cellular phone transmitters.

11. Lightning survival: the possibility exists of a direct or indirect hit by lightning. Suitable lightning protection is required to protect both the camera and other electronic devices in the signal path or connected to the same power circuit.

83

12. Power supply noise immunity: the camera and associated electronics should be tolerant of poor power quality, such as low voltage, noise, spikes, and brief interruptions.

TEST PROCEDURES AND RESULTS

Thirty-two sample video cameras were subjected to tests designed to assess their performance relative to the aforementioned requirements. The test procedures and relevance of the test results to traffic surveillance are described in the following.

Laboratory tests involved measurements of electronic parameters that underly many of the surveillance requirements. These parameters included resolution, sensitivity, noise, dynamic range, grayscale linearity, geometric linearity, flair, bloom, lag, comet tail, vertical or horizontal smear, and backfocus accuracy. In addition, the color cameras were tested for color fidelity, as indicated by color vector magnitude and phase accuracy and white balance.

The tests may be divided into two categories: static tests that involve images that contain no motion and dynamic tests that use images with moving objects or light sources.

The following static and dynamic laboratory tests were conducted. More than one parameter is measured in each test setup.

Test

Horizontal resolution Sensitivity and bloom Gray scale linearity Geometric linearity Lag, comet tail, smear Color fidelity

Static/ Dynamic

Static Static Static Static Dynamic Static

Camera Types

Monochrome/color Monochrome/color Monochrome/color Monochrome/color Monochrome/color Color

A video test bench was fabricated, upon which the camera under test is mounted and focused on a test chart or moving light source. The following video test charts manufactured by Hale Color Consultants, which conformed to EIA standards, were used:

• One RETMA resolution chart, EIA 1956, • Two 11-step grayscale reflectance charts, •One window chart, •One EIA/RETMA linearity chart, 1961, • One EIA/RETMA registration chart, and • One color calibration chart.

In addition, a "black hole" test chart for sensitivity and transient response tests was fabricated, consisting of a maximum reflectivity white chip placed in front of a 3.0-m-deep hole lined with black felt, having essentially zero reflectivity.

All tests used a set of laboratory standard Fl .4 C-mount lenses. Focal lengths were adjusted to match the various imaging chip dimensions.

The test illumination was designed to duplicate natural daylight. The NTSC illumination standard for color television is defined as CIE (Commission Internationale de l'Eclairage) Illuminant C, representative of average daylight according to available data in 1931. The definition of daylight has since been upgraded to CIE standard D65; however, Illuminant C is still the definition incorporated into the NTSC standard.

84

Therefore, Illuminant C was used as the illumination standard for these laboratory tests (2).

Illuminant C has a correlated color temperature of 6800 K. This was· achieved using 3400 K tungsten lamps with Kodak Wratten sheet filters stacked to achieve a -148 Mired shift, yielding a corrected color temperature of 6844 K. .

The use of incandescent illumination in the laboratory mtroduces a larger-than-natural IR component. The Tektronix standard light meter in this study was insensitive to IR radiation. BIA-referenced test procedures predate solid-state cameras and have not yet been updated to deal with the significant IR sensitivity. Experimentation indicated that it was not possible (or desirable) to completely remove the IR component; however, it was possible to reduce it using IR blocking filters.

Resolution

The horizontal resolution of the camera generally correlates with the amount of information present in the video signal generated by the camera. Greater resolution means that ei~her (a) for a given angular resolution requirement, a larger field of view may be imaged or (b) for a given field of view, a finer grain in the image may be discerned.

Resolution is a factor of primary importance in terms of the ability of the TOC operator to interpret the camera image on a monitor. Although the camera optics may be used to trade surveillance area for the minimum resolvable. feature size in the image, the electronic resolution of the camera is a constant, representing a product of these two factors.

Resolution as viewed by the TOC operator can also be limited by the monitor or the bandwidth of the communications path from the camera to the monitor. In view of this, it is concluded that camera resolution is important but only up to the resolution-related limits of the other components of the CCTV system.

Resolution is quantified by the number of "television lines" that can be distinguished electronically in image. This is measured as the maximum number of black and white bars of equal width that can be distinguished along the entire width (horizontal) or height (vertical) dimension of the television picture. Because the ratio of the horizontal dimension to the vertical dimension of the image is 4:3, VJ more lines are required in the horizontal compared with the vertical dimension to achieve equal vertical and horizontal resolution.

Vertical resolution is fixed by the EIAINTSC vertical line specification (525 lines interlaced). Because solid-state cameras separate line scans with separate rows of pixels, the vertical resolution is some number slightly less than 525 (depending on the number of scan lines displayed), divided by an integer (usually 1 or 2).

For solid-state cameras, horizontal resolution is fundamentally limited by the horizontal pixel density of the imaging chip. However, bandwidth limitations in the signal path may also limit horizontal resolution.

The EIA standard test chart for resolution measurement contains horizontal and vertical wedges of converging groups of lines. With the camera focused on the test chart, a single scan line is isolated using a video waveform analyzer. Increasingly narrow areas of the line wedges are scanned, and the video signal is displayed on a digital oscilloscope. The signal amplitude variation is reported (in decibels) relative to


the direct current (DC) black and white level difference. The resolution limit was defined as the line density that yields -15 dB of the DC black and white amplitude spread.

Sensitivity and Dynamic Range

Sensitivity is an indication of the ability of the camera to form an image in low-light conditions. Daytime illumination l~vels greatly exceed the lower sensitivity limits. At night, the bnghtness of vehicle headlights is much greater than the reflected light from the vehicles or roadway features. The ability to detect features in the image other than just the headlight spots depends primarily on the dynamic range of the camera and secondarily on the actual low-light limit, assuming at least some minimum level of reflected light from the features.

Most manufacturers specify sensitivity as the minimum illumination level necessary for either full or usable video. However, the definition of full or usable video is often manufacturer specific or nonrigorously defined. Measurement of sensitivity is further complicated by AGC, IR cut filters, and the spectral characteristics of the illumination itself. The video signal path gain can be increased, making a camera ~ppe~r more sensitive in terms of its output voltage versus illumination level relationship. However, the intrinsic camera noise increases proportionally.

These ambiguities were avoided by measuring camera sensitivity relative to the camera noise level, an approach that cancels the effect of any gain in the signal path that acts on both the image information and the noise. We define the lowlight sensitivity limit as the incident illumination o.n the ~lack hole chart (in lux), which yields a 0-dB RNS ratio of signal to noise for a scan line through the white chip.

The dynamic range of the camera was measured at the. s~~e time by increasing the illumination level from the sensitivity limit to the saturation limit.

The ratio of signal to noise (SIN) of a camera system is defined as the ratio between the camera peak signal output and the root-mean-square (RMS) noise output. SIN is evaluated by measuring ~he RMS noise output of the system when no light is permitted to enter the pickup device and comparing this with the rated camera output. This measurement cannot be reliably made unless the AGC and black clip circuits of the camera can be disabled, which was not possible for all cameras tested.

An attempt was also made to measure bloom during the sensitivity test. Bloom is th~ spread of the image around the original image caused by charge leakage in the pickup device. Bloom can also be observed as a result of faulty optics, usually a result of poor or nonexistent lens coatings. Although bloom can be a significant problem for tube cameras, solid-state cameras usually are unsusceptible. None of the cameras tested exhibited significant problems with bloom.

Flare is manifested as fluctuations in the black level of an image related to varying white levels. Flare is not known to be a common problem with solid-state cameras, so it was not specifically measured in these tests.

Gamma/Grayscale Linearity

Gamma is a metric of the linearity of the relationship between the incident light intensity and the signal voltage produced

MacCarley et al.

by the camera, with gamma = 1.0 corresponding to a truly linear relationship. However, a unity setting is not always desirable because the human eye, and often the monitor also, have nonlinear responses. A gamma setting of 0.45 usually produced a grayscale that appeared linear to the eye as rendered on the laboratory monitor. Gammas higher than 0.45 tended to emphasize the contrast between black and white.

Although it is important to the accurate reproduction of the image, linearity does not appear to be a factor of primary concern in traffic surveillance. From a TOC operator's point of view, the shade of gray representing a particular object in the scene is probably of little relevance (monochrome assumed). The relative intensity differences between features in the r:image convey the greatest information. Provided that the image is not overly flatt'ened out or binary from excessive contrast, deviations from perfect linearity are probably acceptable.

Linearity was tested using the Hale/EIA grayscale test chart under standard illumination. A linear response would correspond to an equal voltage difference between each of the nine gray levels on the chart. Linearity is reported as the percent average absolute difference between the signal voltage levels and truly linear increments, for a single scan line through the grayscale field.

As previously discussed, there were problems performing the grayscale linearity measurement for most of the monochrome cameras because, although the visible light reflectivity increment of each successive gray level is constant, the IR reflectivity is not. For cameras that were highly IR sensitive, the darkest gray level appeared nearly as bright as the white reference chip because of its high IR emissivity. This problem occurs even under natural daylight illumination. The only solution was to ignore the darkest gray level in the linearity measurement.

Geometric Linearity

The geometric linearity of a camera is a measure of its tendency to introduce dimensional distortion in the image. This could be an important factor in the inference of distances or shapes in a traffic scene displayed on a TOC monitor. The monitors in the TOC also introduce geometric distortion, and the human eye tends to tolerate minor distortions.

Geometric linearity was tested using a Hale/EIA geometry test chart consisting of a grid of evenly spaced dots. Ideally, this should be reproduced by the camera without any dimensional distortion. The signal from the camera viewing the test chart is mixed with a reference linear signal produced by a video signal generator. Registration of the dot pattern from the camera signal with that of the reference signal is measured on a monitor. Geometric linearity is reported as the percentage average absolute dimensional misregistration at five key positions on the test chart (center and four corners).

Tube technology (such as Vidicon) cameras are susceptible to geometric distortion because of the electron-beam scanning action that produces the video signal. This is not the case for solid-state (CCD) cameras because precise photolithography locates the imaging elements (pixels) on a wafer of silicon. Because all surveillance cameras tested (except one reference camera) were solid state, the geometric linearity of all the cameras was nearly perfect. Detected diffe,rences were prob-

85

ably more a result of optical flaws than of variations between imagers.

Color Fidelity

For color cameras, the TOC operator would expect a reasonably faithful reproduction of the colors and their relative intensities in the image. Although color fidelity is only an aesthetic issue in entertainment, it could become a critical issue in traffic surveillance. For example, a TOC operator might observe a car that appears to be a particular color on his or her monitor involved in a hit-and-run accident and then dispatch appropriate law enforcement. Poor color reproduction might cause the vehicle color to be incorrectly reported, leading to a questionable arrest by the officer.

Color fidelity is tested using a standard video test instrument called a vectorscope. Using a standard color bar chart under standard illumination, three primary colors and three color combinations are tested, each yielding a color vector displayed on a vectorscope.

Each color bar is associated with a vector with a characteristic magnitude and phase. The phase corresponds to the color hue, whereas the magnitude corresponds to the relative color intensity. A camera with perfect color reproduction would produce color vectors of the correct magnitude and phase on the vectorscope, for a line scan through the color bars. The difference between the actual vector magnitude and phase produced by the camera and the correct values is reported as magnitude and phase errors for each color vector. The absolute values of all six magnitude errors are averaged together and reported as percent average magnitude and phase errors. W~ite balance is an indication of a color camera's ability

to faithfully produce the neutral color white. True white reproduction results in a centered dot on the vectorscope. White balance is reported as the actual position of the white dot relative to the center, usually stated as a magnitude and phase deviation characteristic of a particular hue and intensity. Most of the 10 color cameras tested exhibited acceptable color fidelity. None was perfect, and two were unacceptable.

The other half of the color reproduction system is the monitor. All color monitors provide adjustments for both color hue and intensity. The monitor adjustments can be used, to some degree, to compensate for the poor color fidelity of a camera. This may be acceptable if each monitor is connected to the same camera all the time. However, in a TOC, the capability must exist for any monitor to switch to any camera. ~ny differences in color fidelity between cameras would yield distorted color reproduction on all but the original setup camera. . It is concluded that color fidelity is an issue of primary importance for color cameras. Poor color fidelity could lead to problems in traffic surveillance.

Dynamic Tests

Some metrics of camera performance are related to motion in the image. Comet tail describes a problem when a bright object moves across a dark field, leaving a decaying afterimage. Similarly, lag refers to the after-image visible when a nonsaturated (gray) object moves across a dark background.

86

These problems are not common in solid-state cameras but are sometimes observed. ·

However, vertical or horizontal smear are problems common to MOS/CCD cameras. The problem is manifested as a white vertical or horizontal bar extending from a bright-point light source in the image, across the entire image. This usually occurs only at sufficiently wide aperture settings such that the light source is saturated while the background is dark. Although not a dynamic problem, the dynamic test apparatus also facilitated the observation of vertical smear.

A camera possessing any of these problems could be seriously limited in the use of traffic surveillance because the field of view contains significant motion and numerous bright-point light sources (headlights) at night.

An apparatus was constructed on the test bench for movement of either a point light source or a gray chip across the field of view at nearly constant velocity.

Only three of the cameras tested exhibited lag or comet tail. In this study, none of these exhibited a problem significant enough to be of concern in traffic surveillance. However, all except one monochrome and one color camera exhibited problems with smear (usually vertical) at wide aperture settings.

Field Tests

Field tests were conducted at two sites. The ARDFA test track at California Polytechnic State University, a 0.55-mi straight roadway instrumented for vehicle position and velocity measurements, was used for the daytime field tests. The cameras under test were mounted on a 25-m tower at one end of the track. Position markers and sevetal typical road signs were placed along the track. Vehicles and test symbols were placed or driven along the track at various speeds and distances from the camera.

For the night field tests, a camera platform was set up on a local overpass on California Highway 101. Both approaching and departing traffic scenes were viewed. These night tests were primarily intended to evaluate low-light camera characteristics.

The field evaluations relied on human vision. Images from each camera were recorded on Super-VHS videotape. Human evaluators in the laboratory compared the recorded video images, displayed on reference monitors, with each other and

,--with synchronized and time-coded photographs taken with a 35-mm photographic earner~. Evaluators completed written questionnaires that were intended to determine both the information they could extract from the image and qualitative issues such as sharpness, clarity, and color accuracy (when applicable).

The assessment of image quality by human subjects may be considered, in one sense, the ultimate test criteria. But the limitations of the video recording process, the monitors, and the subjective nature of human reactions suggest only cautious inclusion of these results in the overall evaluation.

The ability of the human observers to identify specific features in the scene is duplicative of the more precise laboratory resolution tests. However, the relative values of color or grayscale linearity to a TOC operator are well addressed in these tests-assessments that could not be done in a laboratory. Color, to some degree, can replace resolution in aiding


a human observer in discerning features in the image. This is fortunate because color detection by an imaging chip (or chip triad) usually comes at the expense of resolution. Color information might also help to distinguish vehicles and other objects from the shadows that they cast.

TEST RESULTS

Table 1 summarizes the test results. Individual cameras are identified by desciiptor codes of the format vv:cb, where vv is the vendor code number, b is nonzero for monochrome cameras, and c is nonzero for color cameras. Complete test details, including specific camera manufacturer and model information, are available in MacCarley and Dotson (3) on public release of this document by Caltrans. The following applies to Table 1:

1. Horizontal line resolution is compared at the -15-dB point and is reported as an equivalent number of lines resolvable in the image along a single horizontal scan.

2. Low-light sensitivity is the illumination at an SIN ratio of 0 dB reported in lux.

3. Grayscale (gamma) linearity is stated as average absolute deviation from the ideal, reported in percent.

4. Geometric linearity is measured as the magnitude of the spacial misregistration over five points on the test chart. It is reported as a percentage.

5. Vertical smear (VS) and lag and comet tail (L/C) are given as simply yes or no values, indicative of whether or not these problems were observed.

6. Field test scores are reported as ratios of the total points received to the maximum number of points possible.

7. Color fidelity measurements are reported as the absolute ·phase error in degrees and magnitude error in percent over six standard color vectors.

8. Cameras are numerically rated on a scale of 1 (worst) to 3 (best) according to overall performance in the laboratory tests, field tests, and finally~ composite of all tests, indicative of the overall suitability of the camera for traffic surveillance applications. The rating system is defined as follows: 1, unacceptable performance; 2, acceptable performance; and 3, outstanding performance.


The majority of the video cameras that were evaluated would probably be suitable for traffic surveillance applications. Cameras not recommended generally were rejected for reasons of very poor resolution, image distortion, or specific operational problems. Cameras that receive high recommendations usually provided excellent resolution and adequate sensitivity and were free of any operational limitations (with the exception of vertical smear and IR sensitivity).

Operational problems of critical concern are those related to the basic usefulness of the camera in its intended application. Synchronization problems, serious image distortion, extreme grayscale nonlinearity, very poor color trueness (phase error), chronic backfocus problems, excessive dead pixels, unusually poor resolution, or unusually low saturation limits are considered causes for a recommendation against a camera.

TABLE 1 Summary of Camera Test Results

I lorizontal Low Light Gamma Color Fidelity D)11amic Field Field Rating

Camera Perf. Cost Resolution Scnsiti\"ily Linearity Geometric (color cameras only) Saturated Unsal. Test 1 Test2 by Class Conunents

Codc Class Class"' (Lines at -15 dB (Lux at 0 dB (scaled Linearity (mean abs. error) lloriz. Scan Vertical Scan (comp. (comp. Lab Field Over-

of d~ 11:1111ic range) sfn ratio) a\"g. dc\".) l\lag.(%) I hase(Deg VS UC vs UC L/C score) score) Tests Tests all

IOI l\lcd. Low 280 0.004 1.70% <0.5% yes no yes no no 2li32 12122 l 2 l Low res., back focus problem

102 l\lcd. l\!od. 500 (-6.78 dl3) 0.004 1.80% <0.75% yes no yes no no 28132 10\22 3 2 2

103 i\lcd. l\!od. 400 (-10.35 dl3) 0.003 1.80% <0.75%1 yes no yes no no 27132 9122 3 I 2

210 Color lligh 290 0.81 0.80% <0.5% 28.39 7.33 yes no no no 110 I 5135 15125 2 2 2

202 lligh Jligh 600 (-6.16 dl3) 0.004 3.00'>.;, <0.5% yes no yes no 110 27132 18122 3 3 3 Highest rated monochrome camera

201 !\led. l\lod. 362 0.006 3.30% <0.5% yes no yes 110 110 19\32 17122 2 2 2

310 Color lligh 278 5.13 1.20% <0.5% 45.9·1 4.17 110 no 110 no 110 16\35 14125 2 1 2 Immune to \"ertical smear

'.\01 lligh lligh 203 0.00·1 2.20% <0.5~·o yes 110 yes 110 110 19132 13\22 l 2 I Poor res., \"cry high IR sensiti\"ity

302 Low l\lod. 250 (-10.95 dB) 0.856 1.60% <0.5% 110 110 110 110 110 8\32 13122 2 1 2 Inunune to \"ertical smear

1401 !\led. l\lod. 400 0.003/0.284 1.90%'2.10% <0.5% yes 110 yes 110 110 25132 15\22 2 3 2

l·I02 lligh lligh 376 0.004.10.155 l.90'%/1.00% <0,75% yes yes yes yes 110 21'32 9122 2 2 2 Lag and comet tail

1410 Color Jligh 450 (-4.27 dl3) 0.585 1.20% <0.5% 14.85 5.5 yes no 110 110 no 19\35 18\25 3 2 3 Excellent res. for color camera

901 lligh l\lod. 485 0.002i0.00 I 1.40% <0.5% yes yes yes yes 110 27\32 16122 2 3 2 Lag and comet tail

1101 l\lcd. Low 450 (-8.08 dl3) 0.003 3.30% <0.5% yes no yes no 110 23\32 J41.22 3 2 2

1102 lligh l\lod. 489 0.003/0.074 4.10% <0.5% yes no yes no no 25\32 14\22 3 2 2 Very nonlinear grayscale

1110 Color l\lod. 396 1.976 3.80% <0.5% 54. l 10.83 yes no yes 110 110 18\35 11\25 2 2 2 High res. but poor color fidelity

1001 l\lcd. l\lod. 467 0.005 1.40% <0.75% yes no yes no 110 22132 15\22 2 2 2

1002 lligh Jligh 512 0.005 1.80'% <0.5% yes no yes no no 21\32 11122 3 1 2 Dead pixels

610 Color Jligh 283 0.49 l.60<?o <0.5% 21.6 11.83 yes no yes 110 110 21\35 14\25 2 2 2 Good scnsiti,·ity for color, low res.

601 !\led. l\lod. 400 O.Q05 1.40% <0.5% yes no yes nq 110 24\32 11122 2 2 2

1301 l\kd. Low 450 (-11.90 dB) 0.01 3.40% <0.75% yes no yes no no 28\32 13\22 2 3 2

401 lligh l\lod. 470 0.346 2.10% <0.5% yes no yes no no 23\32 20122 2 3 2 IR-inunune, noise problem

410 Color l\lod. 450 (-8.09 dB) 1.113 0.90%1 <1.0% 35.37 9 yes no yes no no 21\35 22\25 2 3 2

501 lligh Low 550 (-12.74 dB) 0.009/0.014 1.60% <0.5% yes no yes no 110 21\32 15\22 2 3 2

502 !\led. ~lod. 450 (-10.42 dB) 0.004/0.004 2.20~·i> <0.75% yes 110 yes 110 110 21\32 14\22 2 2 2

510 Color High 294 1.722/1.722 1.80% <l.0% 18.38 13.17 yes no yes no 110 19\35 18\25 2 2 2 Poor color "trueness" (phase error)

810 Color Iligh 263 0.319/0.314 1.10% <0.5% 28.98 10 yes no yes no· 110 16\35 19\25 l 2 1 unstable color balance

801 l\led. l\lod. 400 (-11.83 dB) 0.002/0.002 3.20% <0.75% yes no yes 110 no 20\32 15\22 3 2 2 Electronic auto-iris

1201 l\led. Mod. 459 0.005 1.60% <0.5% yes yes yes yes yes 24\32 13\22 2 2 2 Lag and comet tail

1210 Color Iligh 328 1.58 1.20% <0.75% 29.96 10 yes no no no 110 21\35 21\25 2 3 2

701 f\lcd. Low 254 0.004 l.30% <0.5% yes no yes 110 no 23\32 11\22 1 2 1 Low resolution, image distortion

710 Color Low 260 2.47 1.60% <0.5% n/a sec test yes no 110 no no 13\35 10\25 1 1 1 Poor resolution, sensitivity, S)nch

•Price Class: <$500 Low, $500<cost<$1000 Moderate, >$1000 11;&11

88

Operational issues of less concern, although still important, include moderate grayscale nonlinearity, a few dead pixels, minor image artifacts (such as dot-grid pattern noise), color intensity fidelity (magnitude error), poor sensitivity, marginal resolution (at least 250 lines), and lag/comet tail problems, if not excessive.

Vertical smear potentially could be a serious impediment to nighttime traffic surveillance. Vertical smear problems prevent the use of wide apertures at night. A wide aperture is necessary if the TOC operator needs to see complete vehicles rather than just headlight pairs. Indeed, the excellent lowlight sensitivity of most cameras is of no value if a bright headlight spot in the image causes vertical smear. With as many as 100 cars in the field of view, 200 bright vertical smear lines render the image useless. This is especially true if the camera is to be used as the input to a traffic video image processing system.

All monochrome cameras tested that were not equipped with IR block filters were sensitive to IR radiation, usually in the 1- to 3-µm near IR range. This radiation is invisible to the human eye and generally correlates with the heat emanating from or reflected by a surface in the scene. IR sensitivity causes false intensity levels in the image: black tires and hot asphalt surfaces appear white. A red car appears whiter than a green car of equal visible reflectivity. It is difficult to say summarily whether this is a real problem in traffic surveillance because enough other visual queues exist in the image to correctly identify surfaces regardless of temperature. Color cameras, by virtue of their color distinguishing mechanism, are insensitive to IR radiation, at least relative to the monochrome cameras tested.

Human subjects in the field tests seemed to accept color information in exchange for decreased resolution. Although color information will never substitute for the resolution required to read a sign or identify a particular vehicle model, it could aid considerably in identifying particular vehicles or distinguishing a vehicle from its own shadow.

Quoted camera costs generally correlated well with performance as measured by the tests in this study, although a few significant exceptions were encountered. However, high cost is often associated with special features, such as a ruggedized


housing or accessible controls, which were not primary evaluation factors in our study. Within the context of the overall system, the cost of the camera is probably a minor issue. Consider that the cost of the environmental enclosure and the remotely controlled pan-tilt-zoom mount for a camera usually exceeds the cost of the camera itself. In view of the installation and maintenance expense, as well as the projected service lifetime, it is recommended that the camera purchase price be considered only as a secondary issue.

Overall, it is concluded that most of the higher-performance cameras surveyed would be adequate for roadway surveillance applications, although significant deficiencies were noted in several cases. The ideal video camera for roadway surveillance would probably be a solid-state color camera with at least a horizontal resolution of 450 lines; a sensitivity of 0.5 lux; and complete immunity to bloom, lag, comet tail, and especially vertical smear. At the time of this evaluation, such a camera was not commercially available. The vertical smear problem is the most noteworthy deficiency, and further development is suggested to eliminate this problem. The rapid pace of video technology may be expected to bring significant improvements within the next few years.

ACKNOWLEDGMENTS

This work was supported by the California Department of Transportation and FHW A.

REFERENCES

1. EIA Standard RS-170, Rev. TR-135. Electronic Industries Association, Washington, D.C., 1957.

2. Benson, K. B., ed. Television Engineering Handbook, Section 2.3. CIE System, McGraw-Hill, New York, 1986.

3. MacCarley, C. A., and S. Dotson. Evaluation of Closed-Circuit Television Technology for Application in Highway Operations. Final Project Report, Caltrans Contract 51J932. California Poly-technic State University, San Luis Obispo, 1992. ·

Publication of this paper sponsored by Committee on Vehicle Counting, Classification, and Weigh-in-Motion Systems.


Traffic Sensing System for Houston High-Occupancy Vehicle Lanes

CLYDE E. LEE AND LIREN HUANG

The typical high-occupancy vehicle (HOV) facility in Houston, Texas, is a single, 22-ft-wide, reverse-flow lane situated in a freeway median and separated from the adjacent freeway main lanes on each side by a concrete median barrier. Arrays of inductanceloop vehicle detectors in the pavement, along with remotely controlled television cameras on high poles, are used routinely for surveillance and traffic monitoring activities. As part of a research study designed to identify and evaluate traffic sensors that feasibly can be used in lieu of the loop detectors, especially on bridges, the Center for Transportation Research, the University of Texas at Austin, designed, installed, and evaluated a traffic data acquisition (TDA) system that features a pair of infrared light beam sensors and a microprocessor. Evaluation of the system showed errorless detection of the direction of travel-critical information for managing a reverse-flow HOV lane-and perfect counting of vehicles, even during a period of heavy rainfall. The TDA system also produces speed, headway, and vehicle-length data. Digital data to and from the system can be transmitted over conventional communication links. The sensors have been operational for the past 13 months without adjustment or maintenance. The TDA system potentially has traffic-monitoring applications other than on the Houston HOV lanes.

A planned 154-km (96-mi) system of high-occupancy vehicle (HOV) lanes is being implemented in Houston, Texas, jointly by the Texas Department of Transportation and the Metropolitan Transit Authority of Harris County (METRO) (1). More than half the system is now operational. Typically, the single reverse-flow HOV lane 6.7 m (22 ft) wide is situated in a freeway median and is separated from the adjacent freeway main lanes on each side by a New Jersey-shape concrete median barrier 0.8 m (32 in.) high. Access to the lane is provided either by grade-separated interchanges with ramps connecting to surface streets or terminal facilities or by slip ramps through openings in the concrete median barrier. Manually operated gates across the openings are used along with official traffic control signs and signals to direct traffic in the proper direction of flow. Buses, vans, and other vehicles with two or more (three or more at certain sites and times) persons are allowed to use the HOV lanes during specific periods, but trucks, vehicles towing trailers, and motorcycles are prohibited at all times.

METRO has responsibility for enforcement and operation of Houston's HOV lane system. Routine surveillance and traffic sensing activities include the use of remotely controlled television cameras on high poles overlooking selected sections of the lanes along with arrays of inductance loop vehicle detectors embedded in the pavement and connected via hard-

Center for Transportation Research, University of Texas at Austin, Tex. 78712-1076.

wire links to a communication center. Work is currently under way to expand and improve these activities with color television equipment and fiber-optics communication links.

Over the years, the inductance loop detectors have provided the primary sensor elements for traffic sensing and control, but an inherent characteristic of this type of sensor, which requires sawing grooves in the roadway structure to install the wire loop, has limited its applicability, particularly on bridges. The reliability and durability of the loops and the communication system along with the time and expense of installing and maintaining the loops under traffic also have been matters of concern.

As part of a research study conducted by the Texas Transportation Institute (TTI), Texas A&M University, to identify and evaluate feasible alternative traffic sensing systems for the Houston HOV lanes, the Center for Transportation Research (CTR), the University of Texas at Austin, designed, installed, and evaluated a traffic sensing system that· uses a pair of infrared light beam sensors and a microprocessor. This traffic data acquisition system is described.

TRAFFIC SENSING REQUIREMENTS

The principal reason for continuous sensing of vehicular traffic in the HOV lane is to detect wrong-way movements so that warning procedures, which might prevent accidents and minimize disruption to normal traffic operations, can be implemented quickly. Additionally, it is desirable to collect statistical data about the number of vehicles of various types that use the lane, speed, and time headway between successive vehicles. These data are important for planning new facilities or modifying the existing ones, developing various operating strategies, monitoring bus activities, evaluating HOV lane efficiency and safety, possil;>ly detecting traffic incidents, and providing warnings to speeding vehicles at critical locations. All these functions require a sensing system and an associated information-processing system that is easy to install, reliable, durable, inconspicuous, inexpensive, and capable of producing the desired types of data.

TRAFFIC DATA ACQUISITION SYSTEM

The traffic-sensing system designed for application on the Houston HOV lanes consists of three functional components: (a) sensors, (b) a signal processor, and (c) a communication interface. It satisfies the traffic sensing requirements stated earlier. For convenience, the system will be referred to as the

90

SENSOR SOURCES ~ ~ 2 ft

SENSOR DETECTORS

+ HOV LANE .. (REVERSE FLOW)

CONTROL UNITS

COMMUNICATION CABLE

LAPTOP MICROCOMPUTER

FIGURE 1 Configuration of TDA system.

traffic data acquisition (TDA) system. The overall configu- . ration of the system is shown in Figure 1, and each component is described briefly.

Sensors

Sensors chosen for this application are through beam, modulated infrared (wavelength, 880 nm) photoelectric devices. The source unit emits a modulated (switched off and on) light beam through a glass lens in a 2-degree, full-conical radiation pattern, and the detector unit has a 2-degree, full-conical field of view. Depending on the type of control unit selected for use with the sensors, the rated maximum effective distance from source to detector is either 21 or 113 m (70 or 370 ft). In the Houston HOV installation, the actual distance between the source (mounted atop the concrete median barrier on one side of the lane) and the detector (mounted atop the concrete median barrier on the other side of the lane) is only 7 m (23 ft); therefore, ample reserve sensitivity is available even though the beam is partially obstructed by dust, grime, fog, smoke, or rain. The source and detector units are connected individually to a control unit via a two-wire, 22-gauge, shielded cable.

The control unit, supplied by 115 V of alternating current, provides modulated power to the light-emitting diode source and conditions the signal from the detector to produce a virtually instantaneous (about 1-msec response time) on-to-off indication when the light beam is blocked by an opaque object. Sensitivity to the proportion of blockage required for switching is adjustable. The output device of the control unit is an optically isolated transistor switch that is connected to the solid-state logic circuits of the signal processor.

Two infrared light beam source-detector pairs are used in the TDA system. The source units are_ contained in an aluminum box 0.1 m2 (4 in. 2) x 0.7 m (28 in.) long cemented to the top of the concrete median barrier on one side of the HOV lane, and the detector units are mounted similarly on the opposite side. Thus, the two sensing light beams are 0.86 m (34 in.) above the road surface and 0.6 m (2 ft) apart in the direction of traffic movement. Anytime that an opaque object blocks a light beam, the transistor switch in the control unit provides a change-of-state (on to off) message to the signal processor.


Signal Processor

The TDA system signal processor unit is based on a Motorola MC68HC11E9 8-bit, single-chip microcontroller that is supplemented with other instrumentation circuits. This unit, which uses a general-purpose 12-V direct current power supply and hard-wire connections to the sensor controllers, receives the on and off signals from the infrared light beam sensor controllers along with signals from an auxiliary signal-sampling generator. These signals are converted to digital form, and programmed computations are performed to produce the desired traffic data. These data include the following:

• Sequential number of vehicle, • Direction of travel, • Speed [mph (km/h)], •Length of vehicle [ft (m)], •Headway [sec (time behind previous vehicle)], and •Time of passage [hh,mm,ss].

Ideally, the signal from a vehicle-presence sensor will consist of a single pulse corresponding to the length of the vehicle. The duration of the pulse multiplied by the speed of the vehicle gives the calculated length of the vehicle. This requires that some part of the vehicle be present in the sensing zone continuously as the vehicle passes. For the TDA system, the approximately conical effective sensing zone of the infrared light beam that is aimed perpendicularly across the HOV lane 0.86 m (34 in.) above the road surface is about 0.1 m (5 in.) in diameter at the center of the lane, and proportionally smaller as the blocking object is located closer to either lane edge. All buses and vans and most passenger cars break the beam continuously as they pass, but some low-profile cars do not. As these low vehicles pass, various objects such as roof posts, seats, and people interrupt the beam to produce a series of short-duration pulses. To make a logical approximation of the length of such vehicles, software in the TDA system signal processor applies a digital filter to the signals from the downstream sensor to group sequential pulses of 250 msec or shorter duration into a single pulse with effective duration from the beginning of the first pulse (begins after a 250-msec period with no pulse, i.e., the minimum gap between successive vehicles) to the end __ of the final short-duration pulse. Thus, the calculated vehicle· length is the composite length of objects and the intervening spac~s when there is less than a 250-msec time gap between successive objects. Vehicle-length data from the TDA system, will make low-profile vehicles appear shorter than their actual overall length, as only the portion of the vehicle higher than 0.86 m (34 in.) is sensed. This can be used in data analysis, however, to distinguish such vehicles from other types.

Direction of travel is determined by the order in which the two light beams are broken. Speed is calculated from the time required for the front of a vehicle to travel the 0.6-m (2-ft) distance between sensors. Headway is the interval between the arrival of the fronts of successive vehicles at the downstream sensor. The number of vehicles is counted as each vehicle leaves the sensing zone of the downstream sensor. The time of passage is determined from the internal clock in the microcontroller.

Lee and Huang

Communication Interface

In the TDA system, the communication interface between the signal processor software and external devices is implemented in software and passed through the RS232C communications port of the Motorola MC68HC11E9 microcontroller. Both signal-processing and communication software are retained in the electrically erasable programmable readonly memory (EEPROM) of the microcontroller. An IBM or compatible microcomputer has been programmed to communicate with the TDA system signal processor unit through its RS232C communications port for gathering and storing traffic data. The data are displayed in real time on the screen of the microcomputer. Control signals can also be sent via the RS232C communications port and modems to traffic warning devices on site or at remote locations. The programs stored on the microcontroller EEPROM are changeable from a remote location through the communication link.

91

speeds were matched to the headway condition, that is, the following vehicle was traveling at a speed lower than the vehicle ahead of it. Vehicle 38 appears to have been a low, small car that was traveling rather fast and at a relatively long distance behind the car ahead.

EVALUATION

The TDA system sensor hardware was installed and adjusted on July 18, 1991, in the HOV lane on I-10 near the Post Oak access gates (west of the interchange with I-610). Personnel worked in the elevated lane for about 1 hr during the earlyafternoon period when the lane was cleared to reverse the direction of traffic flow. Cable installation beneath the surface to connect the sensors to the control units and signal processor in a cabinet at ground level required another 2 hr. The sensors have needed no further adjustment or maintenance; they have

· now been in service for 13 months.

Sample of '{raffic Data

Table 1 presents a sample printout of traffic data collected on a laptop microcomputer. The TDA system was operating on the I-10 HOV lane west of downtown Houston near the interchange with I-610. Traffic was eastbound in the morning hours. Judging from the length and speed of vehicles, Vehicle 30 was a single-unit bus, and Vehicle 33 was an articulated bus-trailer vehicle. Actual measurements of buses in a nearby bus terminal have confirmed these length values within 0.3 m (1 ft). The two passenger cars (Vehicles 34 and 35) following the articulated bus were operating at short headways, but their

On April 16, 1992, a signal processor unit was connected to the sensor controllers in the cabinet beneath the HOV lane. The RS232C port on the unit was connected via a communication cable to a similar port on an IBM-compatible microcomputer located in the METRO surveillance center building about 100 ft away. With assistance from TTI and METRO personnel in the center, a television surveillance camera, connected to a monitor and to a video recorder, was aimed at the HOV lane where the TDA system was installed. During the periods given in Table 2, traffic data from the TDA system and video images of the passing vehicles were recorded simultaneously. The periods encompassed morning, afternoon, and early-evening hours as well as a time when directional

TABLE 1 Sample of Traffic Data from TDA System for HOV Lane on 1-10 at 1-610 in Houston

Begin Date: 04-17-1992 Begin Time: 07:48:05

Vehicle No. Direction Speed (km/h) Length (m) Headway(s) Time

30 31 32 33 34 35 36 37 38 39

WE 53 7.01 86.939 08:09:46 WE 74 5.18 4.730 08:09:51 WE 66 3.35 11.689 08:10:03 WE 50 17.07 11.102 08:10:14 WE 48 3.66 1.369 08:10:15 WE 42 5.18 1.490 08:10:17 WE 84 3.35 47.779 08:11 :05 WE 66 3.66 3.961 08:11 :09 WE 79 2.74 30.150 08:11 :39 WE 60 4.27 7.410 08:11 :46

TABLE 2 Comparison of Vehicles Counted by TDA System and by Observers Viewing Recorded Video Images

Date

4/16/1992 4/17/1992 4/17/1992 4/17/1992

Time

16:56:00-18:54:00 07:48:05-09:49:18 12:30:00-14:30:00 14:30:00-15:48:00

TOTAL

Number of Vehicles

TOA System Video

552 192 225 ill

1224

552 192 225 ill

1224

92

flow was changed from west to east to east to west, and ,a period (2:30 to 3:48 p.m.) with heavy rainfall.

Human observers subsequently viewed the recorded video images and counted all vehicles in the HOV lane,. by direction of travel, during exactly the same periods when counts from the TDA system were recorded. It was possible to check the synchronization of the two counts by using easily recognized vehicles such as buses. '

There was perfect agreement between the vehicle counts for all periods, as indicated by the data in Table 2. Likewise, there was no error in the direction of travel indicated by the TDA system.· Although there were no quantitative· measures against which to compare the other TDA system data, speed, vehicle. length, and time headway values all appeared to be reasonable and consistent with the corresponding relative values that could be judged from the recorded video images. There was no degradation of performance during heavy rain.

It is appropriate to point out that this application of the TDA system was for a single traffic lane 6.7 m (22 ft) wide. The system also has potential use at other one-lane sites, such as freeway entrance or exit ramp terminals (e.g., to detect wrong-way movements, to count and classify vehicles by length, and to measure speed), interchange ramps (e.g., to activate speed-,warning devices at curve·s), and automated toll gates through which vehicles move at a constant speed. When the infrared light beam is directed across two or more traffic lanes, the detrimental effects of simultaneous arrival of vehicles must be evaluated. In some instances, these effects may be considered to be insignificant. The system can also be used to detect over-height vehicles and to trigger warning devices.

The TDA system described in this paper was a first-generation development that has not yet been produced in quantity; therefore, costs that are comparable with those of other traffic sensing systems do not exist. It should be feasible to manufacture and install the system for less than about $3,000/site, excluding communication linkages. Installation can be made at many sites without blocking the traffic lane. At other sites, a shallow saw cut in the pavement might be necess.ary to accommodate a small cable that ·connects the source and receiver units of th~ sensors. The TDA system on the Houston HOV lane has opeiated for more than a year without maintenance or service.

SUMMARY

The traffic sensing system that has been developed especially for use on Houston's HOV lane network is easy to install, reliable, durable, inconspicuous, relatively inexpensive, and capable of producing certain essential data that are needed for managing and operating the extensive mileage ·of specialpurpose lanes that are being implemented. A pair of infrared light beam sensors is teamed ·with a programmable micro-


controller and associated communication links to comprise the basic TDA system. The sensors, mounted in small protective metal boxes atop the concrete median barriers on each side of the HOV lane, have been operational for more than a year without adjustment or maintenance. Installation time for personnel in the elevated traffic lane was about 1 hr, and no modification to the roadway structure or to the barriers was involved. Evaluation of the TDA system, by comparison with recorded video images of vehicles traveling in the HOV lane, indicated error-free performance in detecting the direction of travel for each vehicle and in counting vehicles in several time periods during 2 days. Heavy rainfall experienced in one period had no adverse effect on performance. TDA system data for speed, vehicle length, and time headway were all reasonable and consistent with a small sample of measurements made on test vehicles driven through the system and with qualitative judgments based on observing the video images. Data are produced in digital format and can be handled over long distances via conventional communication links. The TDA system should be considered as a feasible alternative to the part of the existing sensing system that depends on inductance loop detector vehicle sensors. Production costs for the system have not yet been determined, but installed cost, excluding communication links, is estimated at about $3,000/site. Although the TDA system was designed for the Houston HOV lane, other applications certainly are possible.

ACKNOWLEDGMENTS

The research reported herein was conducted under an interagency agreement between TTI, Texas A&M University, and CTR, University of Texas at Austin. The TTI study was sponsored bythe Texas Department of Transportation and METRO, Harris County. Motorola, Inc., contributed hardware com-

. ponents, reference materials, and advice through its Univer. sity Support Program.

REFERENCES

1. Christiansen, D. L. Status and Effectiveness of the Houston HighOccupancy-Vehicle Lane System, 1988. In Transportation Research Record 1280, TRB, National Research Council, Washington, D.C., 1990, pp. 119-130.

2. Juds, S. M. Photoelectric Sensors and Controls: Selection and Application. Marcel Dekker, Inc., New York, 1988.

3. Lipovski, G. J. Single- and Multiple-Chip Microcomputer Interfacing.Prentice-Hall, Inc., Englewood Cliffs, N.J., 1988.

4. M68HC11 Reference Manual. Motorola, Inc., 1989.



Cluster.Analysis of Arizona Automatic Traffic Recorder Data

JOE FLAHERTY

Monthly factor data were used as input data for cluster analysis of 28 permanent traffic volume counters installed .in Arizona. Monthly factors are the ratio of monthly average daily traffic to annual average daily traffic (AADT). Cluster analysis is a statistical procedur·e that reveals natural groupings in data. There are two types of clustering methods: hierarchical and nonhierarchical. Hierarchical methods use a successive series of either mergers or division. Nonhierarchical methods group objects into a collection of clusters "K." Monthly factor data .for each location collected over 5 years were used in the cluster analysis. The group mean monthly faetors of the groups that were determined and the monthly factors of each location were applied to the appropriate randomly selected daily traffic count. These counts were proxy variables for short-term 24-hr counts. Statistical analysis was used to determine the "best" method for deriving monthly factors and also provide~ the best estimates of.AADT. From the results of this analysis, it was determined. that the two primary groups derived from using four clusters were the best and the most stable of all the variations used in the analysis. The statistical analysis revealed that the results obtained from using the grouped mean monthly factors of this variation were marginally better than those from the other variations. The two distinct groups that were determined. as a result of the analysis are. quite stable with respect to time and provide an estimated level of precision that was greater than acceptable. ·

The purpose of this study is to assess in as objective a manner as possible the feasibility of implementing the procedures recommended in FHWA's Traffic Monitoring Guide (1) (TMG) for expanding. short-period traffic counts to estimates of annual average daily traffic (AADT).

The TMG suggests that the best approach to use is one that omits as much subjectivity as possible and is based on sound statistical procedures. It recommends that for the purpose of developing estimates of AADT, automatic traffic recorders (ATRs) with similar patterns of monthly variation be grouped together and the means of the monthly factors of the groups be used to expand short counts to estimates of AADT.

The grouping procedure that is recommended in the TMG is a computerized statistical technique called cluster analysis. Cluster analysis is used to discern the groups. Short-term traffic counts can be simulated from daily ATR data and factored volumes using the current method can be compared with factored volumes derived from these simulated counts adjusted by the factors of the appropriate groups as determined by the cluster analysis.

The monthly factors for each of 5 years for the 28 A TRs that were installed at various locations throughout Arizona

Transportation Planning Division, Arizona Department of Transportation, Phoenix, Ariz. 85007.

were used to conduct the cluster analysis. The monthly factors ~re the ratio of monthly average daily traffic to AADT.

CLUSTER ANALYSIS

Cluster analysis is.a multivariate procedure for detecting natural groupings in data. In one respect, it is similar to discriminant analysis in which the researcher seeks to classify a set of objects into groups. The difference is, however, that unlike discriminant analysis neither the identity nor the number of groups in the data set is known. Stated another way, discriminant analysis is: .a classification m:ethod ·that pertains to a known number of groups. The operational objective of any classification method is to assign·an observation to one of'the ·groups. 'Cluster analysis differs from discriminant analysis in that it is a more mdimentary technique.

In cluster analysis, no prior· assumptions are made concerning the number of groups or the. group structure. The grouping is accomplished on the basis of similarities or distances. The necessary input is data from which similarities can be computed. In the context of this project, itis the 12 monthly factors for each A TR.

There are two basic types of clustering methods: hierarchical and nonhierarchical. Hierarchical methods use either a series of successive mergers or· successive divisions. Agglomerative hierarchical methods. begin with ·individual objects. Initially, there are as many clusters as there are objects. First, the most similar objects are grouped. These groups are then merged according to their similarities. This process continues until ultimately the similarity decreases and the groups are fused into one cluster.

Divisive hierarchical methods work, as the name suggests, in just the opposite manner. An initial group of objects is divided into two subgroups so that the objects in one group are most distant from . the objects in the other. These subgroups are then divided in the same· manner. This process is continued until ultimately there are as many subgroups as there are objects.

The results of both of these methods are often displayed in a dendrogram. A dendrogram is a two-dimensional diagram that depicts the mergers or divisions that have been made at sequential levels. ·. Nonhierarchical clusteriiig methods group objects into a

collection of clusters, K. K, the number of clusters, may be prespecified or determined by the clustering algorithm. These methods ordinarily begin with either an initial set of seed points that form the nuclei of the clusters or an initial partition of objects into groups. The beginning configuration should

94

be relatively free of overt bias. This can be assured by randomly selecting seed points or by randomly partitioning the objects into initial groups.

The Systat software package for microcomputers (2) was used to conduct the cluster analysis of the A TR monthly factor data. The cluster module of this software package employs both hierarchical and nonhierarchical algorithms.

Hierarchical clustering was the first method applied to these data. If the input data are a rectangular matrix, a distance matrix is computed as a first step. If they are a symmetrical matrix, the input data will be used directly for computing distances. The output is a dendrogram. A dendrogram is analogous to a tree diagram. It displays the linkage of each object or group of objects as a joining of branches in a tree. The base of the tree is the linkage of all the clusters into one cluster, and the ends of the branches point to each object.

The dendrogram is displayed or printed so that the most similar objects are closest to each other in the branch ordering. Additionally, the cluster diameters (joining distances) are printed on the extreme right of the dendrogram. Thus the analyst can see the clusters that are being joined and the distances at which the joining occurs. If the A TR station numbers are input as character variables, they will appear on the extreme left of the dendrogram.

The nonhierarchical method available in Systat (2) is the K-means method. This is an iterative procedure that assigns objects to nonoverlapping clusters. The number of clusters can be prespecified. The number of prespecified clusters can be as large as the number of cases. The default number is 2. The number of iterations can also be prespecified; the default is 50.

The K-means algorithm produces the selected number of clusters by maximizing between, relative to within-cluster variation. It is analogous to a one-way analysis of variance with the number of groups unknown, and the largest F-value is sought by reassigning objects to each group. The output is tabular with summary statistics for the number of clusters. Additionally, the members of each cluster are identified and the statistics for the variables that are being clustered are included. Note that all data outputs referred to in this paper, are available from the author.

These statistics are, in the aggregate, the sum of squares between clusters and the degrees of freedom, the sum of squares within clusters and the degrees of freedom, and an F-ratio that describes the between-cluster variability relative to the within-cluster variability. The statistics for each cluster contain the minimum, maximum, and mean values of the monthly factors. Also included is the standard deviation of the monthly factors and the joining distance of each of the cluster members.

The data for the first year were first analyzed with the joining method. The dendrogram was useful for depicting which ATR stations group together and where they group. It is difficult and. cumbersome to use the dendrogram for any fruitful analysis.

The K-means method was applied to the same data. This method was the best of the two because the output was in a format that was more fruitful for determining the results of the analysis. The cluster members were clearly identified. The mean, minimum, and maximum, and the standard deviation of the group members were included in t~e output tables.


The distances were displayed next to each A TR station number. It was also helpful that this method allowed the number of clusters to be prespecified and thus varied. Varying the number of clusters and accepting the "best" results is one way to circumvent the overt bias alluded to above.

The K-means method was applied to this data set with the number of groups varied from two to nine. The fact that the number of groups could be preselected was useful for two reasons. First, it allowed the analyst to determine how the ATRs were related to each other. Second, it allowed the analyst to see how strongly they were related to each other as the number of groups was increased.

As a result of following the above procedure, a few things became apparent. The first of these was two A TR stations that apparently were not related to each other or to any other ATR station in the other groups. Second, for the stations within the groups, it appeared that the similarity of the pattern of the monthly factors was more a function of geography and topography than functional classification of the highway on which the ATR station was situated. Additionally, the population of the surrounding area did not appear to provide much of an explanation as to why the ATRs grouped as they did.

As the number of groups was increased, beginning with six groups, the two largest groups, based on the number of members, remained relatively constant, but members of the smaller groups were being transferred to new.groups. The implication of this is that less and less information was contained in these groups, and they were more a function of white noise.

The same approach to cluster analysis was applied to the other four years of A TR monthly factor data. The number of groups that were prespecified was again varied from two to nine. The joining method was used to cross check the results of the K-means method. The results of these analyses were by and large consistent with the results obtained from the analysis of the first-year ATR monthly factor data. The two A TR stations that were outliers in the first year's analysis were also outliers in the other years. Again, there were two primary groups in which the same A TR stations consistently grouped with each other. Also, as with the first-year data, increasing the number of groups led to the "unstable" stations forming new groups.

It was then decided that the two ATR stations that were consistent outliers be excluded from the cluster analysis because they obviously had monthly factors that were vastly different from the other ATR stations and from each other.

These two ATR stations, one near the primary entrance to Grand Canyon National Park, and the other on the primary route to Puerto Penasco (Rocky Point), Mexico, have tremendous variation in their monthly factors. Station 17, near the Grand Canyon, has monthly factors of an average of 1.66 in July and an average of 0.45 in December. Station 26 on the route to Rocky Point is somewhat the reverse with average monthly factors of 0.70 in August and an average of 1.32 in December.

B~c;~µse of their location and the vast swings in their monthly factors, it was obvious that they each had unique patterns of variation primarily influenced by recreational activity. These stations could be considered as one group each.

These stations and their monthly factors were deleted from each year's ATR monthly factor data set. The data sets were

Flaherty

reanalyzed in the same manner as described earlier. The results of running this analysis again were consistent with those of the prior analysis except that there were no consistent outliers.

The stations in the two primary groups were consistent with the prior analysis. Again, the functional classification of the highway on which the ATRs were located was not a determinant of how the stations clustered. It again appeared that the underlying reasons for the groupings were geography and topography.

One important consideration and objective of deriving ATR groupings has to be the stability of the groupings over time. To check for the stability of the groups, the data were summarized in a table to determine which stations were grouping together in each year and over the 5-year period.

The information in the table was then transferred to maps for ease in determining the geographic and topographic distributions for each ·year and for the 5-year period. The map for the entire 5-year period confirmed the basic stability of the two primary groupings over the 5-year period. There were, however, some aberrations with respect to these groupings. It appeared that A TR Station 28 did not have the same pattern of stability over the 5-year period that could be expected with respect to its location. It is located in Tucson, which is an urbanized area with a relatively low elevation. On further investigation, it became apparent that the reason it did not conform to expectations with respect to consistent group

\ membership was that in 1 month the traffic at the location was abnormally low. This abnormally low volume was attributable to traffic restrictions that were imposed because of construction.

If anything, these findings point out the need for cogent analysis of the results of the cluster analysis output. Cluster analysis is a powerful analytical tool for discerning "natural" groupings of data. However, the analyst must have an understanding of the data that are being analyzed. In this instance it is imperative that the analyst be, or have available as a resource, someone who is knowledgeable about traffic conditions in proximity to the A TR locations and the state as a whole.

The crucial point to be made in this context is that no analysis, in a strict sense, is completely objective. The conclusions reached as a result of the analysis, must be reasonable and justifiable. Thus it is reasonable to expect that ATR Station 28 would, over the 5-year period, be in Group 1. If it is not, then why is it not? From the discussion above, it is clear that in 1 month of a year the traffic volumes were so divergent from the norm that it led to an unexpected result.

In addition to the individual plots of monthly volumes relative to annual volumes for each A TR in each of the 5 years another graphic tool of analysis was employed in this study. The data set contained the 12 monthly factors for each of the 5 years for each Of the ATRs. The data were smoothed by a distance-weighted least-squares algorithm. As the name implies, this algorithm fits a line through a set of points by leastsquares regression. Every point on the smoothed line is a function of a weighted quadratic multiple regression on all the points. This procedure produces a true locally weighted curve through the points. This algorithm permits the surface to flex locally to better fit the data.

95

This plotting procedure is ordinarily used to determine the shape of the function needed to regress one variable on another when the analyst is uncertain of the functional form. They are used here as a post hoc indicator of functional form to clarify the results obtained from cluster analysis. Examples of these plots are shown in Figures 1 through 4. They are representative of Groups 1 and 2, inconsistent and recreational, respectively. The figures show patterns that on a stationby-station basis are consistent with the results of the cluster analysis.

STATISTICAL VALIDATION

The determination of the groups is not the culmination of the analysis. In some respects it is only the beginning. The use of group monthly factors to adjust short period counts must be validated. This validation should be based on a comparison of the present method to the alternative method under development. This comparison can be made by synthesizing short counts and applying both the mean monthly factors derived from the cluster analysis and the monthly factors of the individual A TRs. The "known" AADTs from the ATRs serve as a benchmark for the validation of the factoring approaches.

Randomly selected Monday, Tuesday, Wednesday, and Thursday volumes from the ATRs were used as proxy variables for short-term traffic counts. Data bases were created for each of the 5 years. These simulated short counts can then be adjusted by the monthly factors developed from the cluster analyses and compared with simulated daily volumes adjusted by the monthly factors of each ATR and the unadjusted simulated short period counts.

The daily volumes for each of the A TR stations were adjusted by monthly factors developed in four different ways for comparison purposes. The four monthly factors and the way they were developed are the monthly factors of each A TR and the group monthly factors for three groups, four groups, and five groups, as determined from the respective cluster analysis.

The hard-copy output of the statistical analysis contained the number of cases, the minimum value, the maximum value, the range, the mean value, the standard deviation, the standard error, and the coefficient of variation for each type of the simulated volumes in the data files. They represent, respectively, the AADT, the unadjusted daily volume, the estimated AADT adjusted by the ATR station's own monthl~ factors, and the estimated AADT using the appropriate group monthly factors derived from the cluster analysis for three, four, and five clusters.

The monthly factors that were used to calculate the· simulated estimated volumes were analyzed in a manner similar to the analysis described immediately above. Data files containing the monthly factors of each A TR and the appropriate group monthly factors from the cluster analyses were used to obtain the same statistics as with the simulated volumes. It was of course unnecessary to use a random sample because this comparative analysis was based solely on the monthly factors. In addition to comparing the summary statistics for each A TR, statistical comparisons were made on each group as determined by the cluster analyses.

Cl)

a: f,P5 0 .... 0 "(

&a.. f,PP

0

. FIGURE 1 ATR Station 14 in Phoenix.

Cl)

a: f,.I 0 ....

- (;,)

."( f,P ll..

•)..

..... _::C .... p,8 ~

0 :5

FIGURE 2 A TR Station 6 near Show Low.

>.... :c .... 2! 0 :e

f,4

FIGURE 3 ATR Station 25 in Yuma.

FIGURE 4 . ATR Station 17 near Valle.

o e~

98

These summary statistics provide an objective basis on which the efficacy of factoring and a comparison of the various methods of factoring can be made. The different factored simulated short-term counts can be compared on the basis of standard deviations, coefficients of variation, and so on.

-These comparisons were made on a station-by-station basis and on a group basis. In virtually every run for all the stations for each year there was an improvement, on the basis of the summary statistics, using simulated factored short counts to estimate AADT rather than unadjusted simulated short counts. The standard deviations were lower and the coefficients of variation were lower.

Although the goal of this research is to develop as objective a method as possible of estimating AADT from short-term counts, it must be remembered that no method is completely objective. The results obtained from computerized statistical analysis cannot be just blindly accepted. They must be interpreted and they must make sense. They are an aid to, not a substitute for, informed decision making. In the context of this analysis, the various statistical analyses had to be consistent and make sense both on an annual basis and over the 5-year period.

From the cluster analysis the procedure using four prespecified groups was the most consistent, particularly with respect to Groups 1 and 2. There was a significant decrease in the joining distance in going to four prespecified groups from three prespecified groups.· There was little or no change in the joining distances when going to five prespecified groups from four.

The summary statistics for the monthly factors for each A TR and for Groups 3 through 5 when sorted and analyzed by the various groups also showed that, particularly with respect to Groups 1 and 2, the four-cluster variation was best. The standard deviations, standard errors, and coefficients of variation were consistently the lowest for this variation compared with the other variations. By and large the same results were obtained when the data were analyzed by the ATR stations within the various groups. These results were con-sistent over the 5-year period. ·

In all the variations of the monthly factor data that were analyzed, the A TR stations in Groups 3 through 5 were inconsistent-especially with respect to the number of stations in each group and membership of the group. These various group member stations did not group together consistently and bounced from group to group from year to year.

The A TR stations that consistently were in Groups 1 and 2 with the four groupings as determined from the cluster analysis show a consistent pattern of variation in each year and over the 5-year period. The inconsistent A TR stations when graphed over the 5-year period show .the pattern of variation characteristics of Group 1 in some years and Group 2 in others. The ATR stations at Valley and Why have characteristics that are different from any other A TR stations and

. each other. · When an analysis of variance was conducted on the random

samples of short counts factored by the four different methods, the null hypotheses that the means of each were the same was easily rejected because the samples were different once the short counts were factored. The mean-square error provided the most useful information, particularly for assessing the efficiency of the estimators. The mean-square error for


the within group was the lowest for the estimates obtained by using the group monthly factors derived from the four clusters. Virtually every measure led to the conclusion that the groupings determined by the K-means = 4 method provided the best groupings both on an annual basis and over the 5 years.

As mentioned earlier, the stations that were inconsistent with regard to group membership have similarities in their patterns of monthly factors with the stations in either Group 1 or 2, but they vary from year to year. This observation, however, raises the question, What about those ATRs that are inconsistent with respect to how they group? They do not consistently fall in Group 1 or 2. They do not consistently group with each other. When some of them do group together, there are not enough of them to provide a statistically valid sample.

Looking at the A TR data on an annual basis was not very fruitful. The inconsistencies in the monthly factor data for these A TRs that were the cause of the instability of their grouping over time could not be circumvented by this backdoor approach.

Because of the inconsistency and instability of these A TRs, which is largely caused by the apparent erraticism of the variation in monthly traffic volumes, it is better to exclude these A TRs from the groupings than to attempt to force them to fit into one of the two groups or into one of their own.

One way to perhaps circumvent this problem would be to conduct short-term counts at those times of the year when seasonal adjustments are not necessary. If the ratio of monthly ADT to AADT is approximately 1 then it would not be necessary to adjust a short-term count conducted in that time period for seasonal variation. A data base that contained the monthly factors for those A TR stations that were inconsistent over the 5 years was created.

The individual monthly factors for each of these A TRs were averaged over the 5-year period. The results of this procedure were not encouraging to say the least. None of these ATR monthly factors was on average approximately 1 for any month.

On the basis of the analyses just described, it appears clear that in Arizona there are two distinct, clearly defined, consistent groups whose mean monthly factors can be applied to short-term (24-hr) counts conducted in their respective domains to arrive at a reliable estimate of AADT. The results of the cluster analysis were not quite in conformance with the results expected. The expected results were that almost every ATR station would fall into a clearly defined group. This type

·of result clearly was not the case. in this study. ·Two possible explanations for this difference from expected

results may be (a) the number of years of data that were analyzed in this study and (b) Arizona's skewed population distribution and topographical divergence.

The two groups have one group in which the A TRs are situated at relatively low elevations and in or near the Phoenix and Tucson metropolitan areas. The second group consists of A TRs situated at relatively high elevations with relatively high volumes in the summer and relatively low volumes in the winter. They are Groups 1 and 2, respectively. Most of the A TR stations that were inconsistent have the characteristics of the two aforementioned groups but they vary from year to year as to which group they resemble. These ATRs are mostly in the western half of the state and along I-40 from Flagstaff

Flaherty

and west. Two ATRs are clearly recreational: ATR 17 near the Grand Canyon and A TR 26 on the primary route to Puerto Penasco, Mexico.

The two distinct groups that were determined as a result of the analysis are quite stable with respect to time and provide a greater-than-acceptable estimated level of precision. It is anticipated that the cluster analysis will be conducted on an annual basis and that the results will be incorporated into the traffic counting program.

Using the group monthly factors will facilitate the assignment of short-term count segments of the state highway system because approximately 75 percent of them are in the domain of the A TRs in Group 1 or 2. The remaining 25 percent will have to be assigned to specific ATRs for adjustment purposes. Seasonal counts will be needed to make the assignments of these count sections to specific A TRs and to delimit the domains of Groups 1 and 2.

99

ACKNOWLEDGMENTS

This study was supported by FHW A and Arizona Department of Transportation (ADOT). The author acknowledges the helpful comments offered by Ed Green and Dale Buskirk of ADOT.

REFERENCES

1. Traffic Monitoring Guide, FHWA, U.S. Department of Transportation, 1985.

2. Sys tat Software Package for Microcomputers. Sys tat, Inc., Evanston, Ill., 1990.



Vehicle Configuration Influences on Weigh-in-Motion Response

P. E. VAN NIEKERK AND A. T. VISSER

Statistics on axle and vehicle masses traditionally have been collected by stopping and weighing vehicles on axle or whole-vehicle weighers. This technique is still applicable but only on roads carrying low truck volumes or in instances in which only a small sample of the population is required. On heavily trafficked roads this technique is extremely hazardous. Because of the inherent limitations of static weighing, considerable developments have taken place in the last decade on systems that permit the collection of data on axle and vehicle masses. These data are recorded while the vehicles travel at normal highway speeds without interference to the traffic stream. These systems are commonly known as weigh-in-motion (WIM) systems. The dynamic pavement loading, by simulation, of various vehicle configurations on a range of pavement roughnesses where WIM installations could be made are reported. In the evaluations the mass variations are attributed to the dynamic influences, although the precision of the measuring system is not taken into account. The simulation procedure is presented and its use is justified by comparing simulated results with measured axle masses. Variation in all the axle loads of various vehicles along sites that typically could be appropriate for WIM, ideally having a pavement serviceability index greater than 3.0, are presented. The influence of the systems flush with the surface and placed on the surface are considered. The effectiveness of WIM to measure vehicle masses is then presented. Finally the implications of these results of WIM accuracy, calibration, and weighing are discussed.

Statistics on axle and vehicle masses traditionally have been collected by stopping and weighing vehicles on axle or wholevehicle weighers. This technique is still applicable but only on roads carrying low truck volumes or in instances in which only a small sample of the population is required. On heavily trafficked roads this technique is extremely hazardous. Because of the inherent limitations of static weighing, considerable developments have taken place in the last decade on systems that permit the collection of data on axle and vehicle masses. These data are recorded while the vehicles travel at normal highway speeds without interference to the traffic stream. These systems are commonly known as weigh-inmotion (WIM) systems.

A number of WIM systems are available on the market. For example, eight different systems are under evaluation on 1-95 at the Pennsylvania and Delaware border (J). These systems are either built into the pavement flush with the surface or fastened onto the surfacing. A range of technologies is employed: for example, strain measurements, capacitive sensors, and piezoelectric systems.

P.E. van Niekerk, VIAED, P.O. Box 35256, Menlopark, 0102, South Africa. A.T. Visser, Department of Civil Engineering, University of Pretoria, Pretoria, 0002, South Africa.

Although attempts are made to calibrate the systems before installation, calibration to assess the local site conditions is essential (2). Previous experimental work (3) also showed that calibration with one vehicle type does not necessarily apply to other vehicle configurations. More recently it was reported ( 4) that the front axle of a five-axle rig is consistently underreported. A need therefore exists for a fuller understanding of dynamic axle loading effects to be able to compensate for the inconsistent results that are found in practice.

Recently a research project on dynamic heavy vehicle influences was completed (5), and this provides a basis for addressing the issues mentioned above. The aim of this paper is to report on the dynamic pavement loading, by simulation, of various vehicle configurations on a range of pavement roughnesses where WIM installations may be made. In the evaluations the mass variations are attributed to the dynamic influences although the precision of the measuring system is not considered. ·

The paper briefly presents the simulation procedure and justifies its use by comparing simulated results with measured axle masses. Variation in all the axle loads of various vehicles along sites that typically would be appropriate for WIM measurements [pavement serviceability index (PSI) > 3.0] are presented. The influence of systems flush with the surface and placed on the surface also is considered. The effectiveness of WIM to measure vehicle masses is then presented. Finally the implications of these results on WIM accuracy, calibration, and weighing are discussed.

METHOD OF ANALYSIS

A wide range of factors influence the magnitude of the dynamic loading effects and their spatial distribution (5). However, only the following factors are relevant for normal highway vehicles traveling over WIM sites and were considered in this study:

•Vehicle configuration; •Vehicle speed; • Road roughness; and • WIM equipment arrangement.

A vehicle simulation program, tire force prediction program (TFP) (6), was used to determine the relative influence of each of these factors. The program's suitability was validated by comparing the program output with measurements obtained during field tests.

van Niekerk and Visser

Because this comparison was found to be acceptable, the simulation program was used to determine the influence of the aforementioned factors. The vehicle configurations included

• A two-axle rigid truck: Truck Type 1; •A three-axle rigid truck: Truck Type 2; •A two-axle tractor with a single-axle trailer: Truck

Type 3; · •A two-axle tractor with a tandem trailer: Truck Type4; •A two-axle tractor with a tridem trailer: Truck Type 5;

and • A three-axle tractor with a single-axle trailer: Truck

Type 6.

To permit the ready comparison of results for each vehicle, the ,responses for each of the vehicle configurations at each site were determined and normalized by dividing the dynamic load by the static axle load.

Vehicle speed is the primary determinant of the dynamic loading and two vehicle speeds, namely 5.5 and 44 km/hr were used in evaluating vehicle effects, but 80 km/hr was used to evaluate speed effects. To assess pavement conditions, three different road profiles with a riding quality of 3.8, 3.3, and 3. 0 PSI were used.

The normalized responses were determined at regular '150-mm intervals along the road profiles, and these responses were then used to show the variation of the dynamic loading along a road. The influence of different WIM installations was handled similarly, except that the dynamic load on the WIM was compared.

COMPUTER SIMULATION .TECHNIQUES

TFP Program

The TFP program is designed to enable the highway engineer to predict tire forces that occur as a vehicle travels along a road in a straight line. The road profile is defined by points at 150-mm intervals representing the elevation of the road for the two wheel-paths.

The TFP program can simulate four basic vehicle configurations:

•A rigid truck; •A tractor-semitrailer combination; • A tractor-semitrailer combination with one full trailer;

and •A tractor-semitrailer combination with two full trailers.

The tire forces predicted by the TFP program are for a road section traversed by a vehicle at constant velocity. These tire forces are applied at the tire-pavement interface .. The tire characteristics and inflation pressures are quantified through the tire spring rate. No turning or braking maneuvers or roll effects can be simulated. This program is useful for the formulation of policy governing highways that normally involve vehicles moving at constant speed along a straight section of roadway.

101

The model used consists of two planar rigid-body inertial masses representing the tractor sprung mass and the semitrailer sprung mass. Each vehicle component's mass is constrained to move vertically (heave) and rotate (pitch) in the direction of travel. The vehicle suspension is represented by a parallel combination of springs, a viscous damper, and a coulomb damper at each axle. The viscous dampers represent shock absorbers and in practice appear only on the front suspensions. The coulomb dampers exert a constant force against the direction of relative motion between the tire mass and the vehicle mass.

The model of the tractor-semitrailer-full trailer used in developing the program is shown in Figure 1. Various profiles that previously had been surveyed were selected and analyzed to determine which would be used for the simulations. From the rod and level data for the sections the root-mean-square vertical acceleration (RMSVA) was determined (7). From a correlation with the RMSV A a PSI was calculated for each section. Finally three sections were used in the simulations. The three profiles included a rough profile (PSI = 3.0), a medium profile (PSI = 3.3), and a smooth profile (PSI = 3.8).

The need for assessing the pavement profile is highlighted by previous findings (8) that the dynamic effects of the pavement profile for approximately 150 m in advance and 30 m after the WIM site could affect the response at that point. The 30-m length after the WIM is to ensure.that the influence of dynamics from the leading axles on the trailing axles is considered.

Validation of Simulation Program

The TFP computer simulation program was verified by using more general simulation programs and comparing these results with those from field tests. One particular field comparison incorporated the use of piezoelectric film strips to correlate simulated results ( 6). From these tests it was concluded that the TFP predictions bound the range of field data.

Simulation Process

The simulations undertaken for the current study involved the determination of the force profiles for each of the vehicle types at a specific speed and over a specific road profile. The dynamic forces were determined at 150-mm intervals and then integrated to determine the gross vehicle mass (GVM) on a 600-mm section because this is the approximate width of WIM platforms. These GVM values were determined for consec-

I

i 0 II 0 I

iiii ii FIGURE 1 TFP tractor-semitrailer-full trailer model.

102

utive 600-mm sections by moving the 600-mm section forward in 150-mm intervals.

. Once this was completed for all the vehicle types, each GVM profile was normalized by dividing the dynamic load by the static mass of the individual vehicles. The various normalized distributions were then plotted to determine whether any similarity existed for the various vehicles. Variation of the relative static mass of a particular vehicle from that of a rigid calibration vehicle (two-axle rigid truck) was also determined. This profile could be used to assess the error resulting from the use of a specific vehicle to calibrate WIM instruments at a particular site.

A further analysis of the dynamic response of the front and leading rear axle of all the vehicles was done at two different speeds and over three road profiles. The response of corresponding axles in various vehicles as well as the response of various axles of specific vehicles was compared.

SIMULATION RESULTS

Vehicle Configuration

Table 1 presents the simulation of the gross vehicle mass for the various vehicle configurations traveling over a medium profile at a speed of 44 km/hr. Furthermore, as a basis for

, comparison, the deviation from that of a two-axle vehicle was determined by subtracting the deviation from the static of the two-axle vehicle from that of the other vehicle configurations at each simulation point along the profile. This gives an indication of the error that would be made if only a two-axle vehicle were used to calibrate the WIM equipment.

From the GVM simulation analysis results the following were found:

•The dynamic component of the GVM is not constant for all types of vehicles.

•No particular vehicle can be taken to represent the worst or best dynamic influence throughout the road section analyzed. The same vehicle does not necessarily retain the same relative ranking for all the profiles considered.

• The variation of the relative response of a particular vehicle from that of the standard vehicle (Truck Type 1) is not constant throughout a particular profile.

• There appears to be no similarity in the responses of the various tractor-trailer combinations when compared with both the standard vehicle and other trailer combinations.


•At 5.5 km/hr all the vehicles' response profiles lie within a very narrow range of + 101 to -99 percent of the static load.

When considering the relative dynamic influence of various axles of a vehicle the following were observed:

•The two-axle rigid vehicle's front and rear axle follow a similar pattern, with the rear axle experiencing the higher relative loads. This effect is expected because the rear-axle response at a specific point along the profile is influenced by the response of the preceding axle to that same point, whereas the reverse is not true. This tendency can be seen in Figure 2, which shows the dynamic responses of the individual axles of a two-axle rigid truck traveling at a speed of 44 km/hr over a profile of PSI = 3.8.

• Relative responses for tractor-trailer combinations also show similar patterns for the front and rear axles, but these patterns are not similar to those of the standard vehicle. Peak relative responses for the various vehicle configurations occur at different points along the profile.

•The tractor-trailer combination's responses do not show constantly higher or lower responses for an axle at a particular point. At some points the front axle may have a higher relative response than the rear axle; at other points the contrary may be true. Figure 3 shows the dynamic responses of the individual axles of a two-axle tractor with a single-axle trailer.

Vehicle Speed and Road Roughness

Influence on GVM

The simulation results obtained for a two-axle rigid vehicle traveling over the three profiles considered and at speeds of 5.5, 44, and 80 km/hr are summarized in Figure 4. The following was observed:

• The speed of the vehicle greatly influences the dynamic response. The response of Vehicle 6 traveling at 5.5 km/hr on a rough profile (PSI = 3.0) deviates from the static GVM within a range of + 3 percent. Traveling at 44 km/hr the same vehicle's response deviates from the static GVM within a range of + 15 percent to -15 percent of the static GVM.

•No particular vehicle can be taken to represent the worst or best relative dynamic influence throughout the section analyzed. The same vehicle does not necessarily retain the same

TABLE 1 Dynamic GVM Response of Various Vehicles Traveling at 44 km/hr over a Profile of PSI of 3.0

% Deviation from static % Deviation from standard Vehicle configuration Maximum Minimum Average Maximum Minimum Average

Truck type 1 15,0 0,0 8,8 0,0 0,0 0,0 Truck type 2 4,6 -5,1 -1,9 13,2 5,1 8,5 Truck type 3 8,6 -8,2 0,1 19,7 -0, 1 8,6 Truck type 4 4,6 -6,l -1,7 11,5 4,9 9,0 Truck type 5 5,0 -5,9 -0,2 14,0 3,6 9,3 Truck type 6 8,2 -1,4 3,5 11,0 0,2 6,0

van Niekerk and Visser

relative ranking for all the profiles considered or throughout a particular profile.

• The speed of the vehicle greatly influences the relative dynamic response. The relative response of a two-axle tractor with a single-axle trailer traveling at 5.5 km/hr on a smooth profile (PSI = 3.8) lies in a range between 98 and 102 percent of the static GYM. Traveling at 44 km/hr the same vehicle's response lies within a range between 91 and 109 percent of the static GYM.

Influence on Individual Axles

The dynamic component is influenced by the profile of the road surface. A specific change in the profile does not, however, influence all the vehicle types in the same manner. Fig-

(/) (/) co E 1.20 u

·.::; ~ (/)

B 1.10 Q)

> -~ Qi ..... 1.00 Cl c :c co 0

-; 0.90 .E co c > 0 0.80 -+------,-----r----.-------,-----i

0 5 10 15 20

Chainage (m)

- Rear Axle * Front Axle

FIGURE 2 Dynamic responses of individual axles of two-axle truck traveling at 44 km/hr over profile of PSI of 3.8.

25

103

ure 5 shows the variation in the dynamic response of the rear axle of a two-axle vehicle for various road profiles at a PSI of 3.0, 3.3, and 3.8, respectively.

WIM Equipment Arrangement

Two types of WIM systems were investigated. One was a plate installed flush with the road surface, and the other was a capacitive mat placed on the road surface. Simulations were done on the three previously described profiles, and the axle loads were determined on a 600-mm planar section at a particular distance along the road section. The simulations were then repeated after the 600-mm planar section had been raised by 6 mm to represent a WIM mat placed on the road surface. These simulations were undertaken at a vehicle speed of 44

(/) (/)

E 1.20-+-----------------------1 ()

·.::; co u; B 1.10-1-~~--.$~-".lt-----+--+----++---~~~--i Q)

> -~ Qi ..... 1.00-+----+-+---~--!'--\-___ .._.,l'r>-t--++--+-;>t-T---;~ ......... ---; Cl c :c co 0

-; 0.90 -+----------71::~-----T-r-------; .E co c > 0 0.80 -+------r-----.---------,------,--~

0 4 8

Chainage (m)

12

*Tractor front axle - Tractor rear axle

16

FIGURE 3 Dynamic responses of individual axles of two-axle tractor with single-axle trailer traveling at 44 km/hr over profile of PSI of 3.8.

3o~-5-.-5-k-m-/h-~--44-km-/h-~-~8~0-k_m_/_h~

() ·;:;

20 .................... .

~ 10 ................ .. (/)

E _g c

0 ~ -0 '-'-' ·;:; Cl)

·;;: Q)

0

-10 .. .................

-20 3.0 3.3 3.8 3.0

• > 1,.

I

3.3 3.8 3.0 3.3

Road Profile Roughness (PSI)

-

3.8

•Average

~Maximum

DMinimum

FIGURE 4 Dynamic response (GVM) statistics for two-axle truck traveling over range of profiles at three different speeds.

104

km/hr and from the previous results it is evident that the dynamic response found would increase as the vehicle speed increased.

The relative responses of the individual axles of a range of vehicle types, traveling over a 6-mm-thick WIM installation is shown in Figure 6. These were determined for a traveling speed of 44 km/hr and a road roughness PSI of 3.3.

For all the profiles considered and for all the vehicles simulated there was an increase in the dynamic response of all axles when the mat was placed on the road surface. This increase of between 6 and 13 percent was evident for all axles and on all profiles. This increase is however compensated by the calibration of the equipment. After correction this would translate to a maximum 7 percent deviation from static.

Cl) Cl)

ro E 1.20 u ·~ Cl)

B 1.10 Q)

> ·;; ro.

"ID ... 1.00 Cl c: :0 ro 0

-;:; 0.90 .E ro c: > 0 0.80 -+------..----~-----,-----..------i

0 5 10 15 20 25

Chainage (m)

-o- Smooth Road *Medium Road * Rough Road

FIGURE 5 Dynamic response of rear axle on two-axle vehicle (44 km/hr).

Cl) Cl)

ro E .g 1.20 ~ Cl)

B Q)

> ·;; ro ~ gi 1.10 :0 ro .2 u .E ro c: > 0 1.00-+------.------.------.------..----~

AXLE 1 AXLE 2 AXLE3 AXLE 4 AXLE 5 AXLE 6

- TRUCK 1 + TRUCK2 "*- TRUCK3 -a- TRUCK4 * TRUCK5 + TRUCK6

FIGURE 6 Relative response of individual axles of various vehicles traveling over 6-mm-thick WIM installation .(road PSI = 3.3) at 44 km/hr.


The WIM results were determined over a 50-m road section. For the raised WIM the capacitive mat was moved forward in 600-mm increments, and the responses were determined for each position. The simulation results showing the deviation from the static load for a two-axle. rigid truck traveling at 44 km/hr are summarized in Figure 7.

The following were found:

• For the WIM mat placed on the surface there was no generalized tendency for the dynamic load to move closer to the static axle load as the pavement roughness improved. For the mat placed flush with the surface the simulation yielded results that did not necessarily show the same tendencies as were shown for the WIM mat simulations.

•For both the raised and flush arrangement the front axle's relative response did not necessarily produce the smallest spread for the range of vehicles considered. No one specific axle was found to produce the best case throughout. The finding that the lowest variance between static and dynamic weight occurs in the second or third axle of a loaded five-axle semitrailer seems to hold (4). The axle with the lowest variance is, however, not always the same.

•There was also no relationship between the front axle's responses and those of the other axles. The front axle's responses could thus not be used to determine or predict the accuracy of any other axles. Any calibration of the WIM equipment would thus have to incorporate both the vehicle configuration and the axle number.

•For the smooth profile and for the case in which the WIM apparatus is placed flush with the surface, the front axle's responses were nearest to the responses of the static axle load.

CONCLUSIONS

Vehicle Configuration

The simulations undertaken again show that the use of any one specific vehicle to calibrate the WIM equipment can lead to serious errors. The suitability of a specific vehicle for use as a calibration standard is influenced by the road profile, the WIM arrangement, and the calibrating and operating speed.

The general approach of using a multiaxle vehicle for calibration is confirmed. The calibration would have to allow for both the vehicle configuration and the axle number when determining the correction to be used to determine the static load.

Vehicle Speed

The speed of the vehicle being weighed influences the variation of the response along a road section. As the speed increases the variation in response also increases. The specific increase varies from axle to axle and from vehicle configuration to configuration. The multiaxle vehicles generally have one axle that is least influenced by the speed. However, which specific axle is applicable is not always consistent and varies with the speeds considered.

van Niekerk and Visser 105

3o WIM 6 mm above road WIM flush with road Front axle Rear axle Front axle Rear axle

20 ........... .

cfi. u -~

~ Vl

E ._g c 0

10

0

:~ -10 > Q)

0

-20

•Average

~Maximum

0Minimum

3.0 3.3 3.8 3.0 3.3 3.8 3.0 3.3 3.8 3.0 3.3 3.8

Road Profile Roughness (PSI)

FIGURE 7 Dynamic response of two-axle vehicle for various WIM arrangements.

Road Roughness

The simulations have shown that roughness variation, even within good payments (PSI > 3.0), have a marked influence on the response. The results for the rough, medium, and smooth profile (PSI = 3.0, 3.4, and 3.8, respectively) show a distinct reduction in the response variation.

The general recommendation that WIM installations be used on pavements with PSis greater than 3.0 can thus still lead to meaningful variations in dynamic responses. Once again the specific response will differ for each vehicle configuration. Some vehicles will react best on one pavement but will not be suitable on another.

Weigh-in-Motion Equipment Arrangement

The study has shown that the lead in profile and placement of the WIM equipment has a great influence on the response of a vehicle. This influence is however compensated by the calibration of the equipment.

The response of a vehicle at a specific point is also influenced by the roughness of the lead-in and exit road section. The placement of any system would thus require consideration of the general site condition at a potential installation point. In this regard the study by Cunagin et al. (8) suggests that a 150-m lead-in and a 30-m exit section need to be considered.

The results show that the vehicle dynamics, and thus the WIM results, could be different between two profiles with identical PSis. This difference is because vehicle response is a combination of profile frequency and amplitude, as well as vehicle configuration, mass distribution, and suspension characteristics. The combinations are therefore limitless. The results also emphasize that even with good calibration, there will be a significant fluctuation in the errors obtained with WIM. Users therefore should be made aware of the potential accuracy of WIM measurements.

RECOMMENDATIONS

Similar to findings reported by Purdhoe ( 4) it is recommended that a range of vehicles be used to verify the weighing system. Furthermore, calibration factors should be determined for various vehicle configurations so as to limit the variation between the static and WIM results.

The use of a flush WIM arrangement is recommended to . supply road loads. As suggested previously (2), the WIM

systems currently available are not suitable for law enforcement purposes but rather should be used for vehicle screening. It has also been stated (9) that the WIM technology can provide reliable trend data to indicate where and when enforcement efforts may be needed.

The purpose of calibration has not yet been refined, and more work is needed to ensure reliable and consistent results at different sites and for various vehicles. The calibration curves must be set up to allow for all the various vehicle configurations that will be operating at the particular installation site.

As shown with the comparison of field (3) and simulation results it is suggested that vehicle simulation techniques be used to identify suitable WIM sites. The further calibration and installation should then be done for the WIM equipment at these identified sites.

REFERENCES

1. Petropola, T., G. Wass, R. Marshall, T. Neukam, D. Bochenek, D. Taylor, and E. Biggs. Joint Pennsylvania/Maryland Weigh-inMotion Demonstration Project. Report Interstate 83. PennDOT/ MD SHA/FHW A. Aug. 1991.

2. Slavik, M. M., and A. T. Visser. Development of a Weigh-inMotion Network in South Africa. In Transportation Research Record 1272, TRB, National Research Council, Washington, D.C., 1990.

106

3. Viljoen, A. W., and F. C. Rust. Interim Results of the Dynamic Axle Load Study Conducted at Boekenhoutkloof. Technical Note TP/30/87. National Institute for Transport and Road Research, CSIR, Pretoria, South Africa, April 1987.

4. Purdhoe, J. Slow Speed "Dynamic" Axle Weighers: Effects of Surface Irregularities. Research Report 134. U.K. Transport and Road Research Laboratory, Crowthome, Berkshire, England, 1988.

5. Van Niekerk, P. E., and A. T. Visser. Dynamic Effects of Heavy Vehicles on Road Pavements. RDAC PR91/280. South African Roads Board, Pretoria, March 1992.

6. Fernando, E.G., R. L. Lytton, W. L. McFarland, J. L. Memmott, F. Helin, and A. N. Jamy. The Florida Comprehensive Pavement Analysis System. Florida DOT State Project 99000-172. Texas Transportation Institute, Texas A&M University, College Station, April 1991.


7. Visser, A. T. A Correlation Study of Roughness Measurements with an Index Obtained from a Road Profile Measured with a Rod and Level. Technical Report No RC/2/82. National Institute for Transport and Road Research, CSIR, Pretoria, South Africa, March 1982.

8. Cunagin, W. D., S. 0. Majdi, and H. Y. Yeom. Intelligent Weighin-Motion Systems. In Transportation Research Record 1311, TRB, National Research Council, Washington, D.C., 1991, pp. 88-91.

9. Basson, J. E. B., A. T. Visser, and C. R. Freeme. In-Motion Weighing of Vehicles on Heavily Trafficked Roads. In Transportation Research Record 1200, TRB, National Research Council, Washington, D.C., 1988.



Influence of Vehicle Speed on Dynamic Loads and Pavement Response

PETER E. 5EBAALY AND NADER TABATABAEE

Weigh-in-motion systems have been used extensively to measure dynamic loads imparted by traffic vehicles. One of the major uses of these load data is to evaluate the equivalent single-axle loads (ESALs) generated by each load level. The cumulative ESALs are then used in the design or rehabilitation procedures, or both, fot the existing road. In situ pavement response parameters, such as the strains at the bottom of the asphalt concrete layer, can also be used to evaluate ESALs. The findings of a research program aimed at evaluating the effect of vehicle speed on the measured dynamic loads and pavement response are documented. The data were measured through a full-scale field experiment. The analyses of the data indicated that vehicle speed has a significant effect on both the measured dynamic loads and the actual response of the pavement system. However, the effects of vehicle speed on dynamic loads and pavement response are not identical. For example, higher vehicle speed generates higher dynamic loads, whereas the strains at the bottom of the asphalt concrete layer are significantly reduced as the speed increases. This discrepancy has been shown to ha've a great impact on the final design of the pavement system.

The dynamic interaction that occurs between the loading vehicle and pavement system plays an important role in the development and progression of pavement damage. Both truck dynamics and payement response characteristics must be considered when determining the extent of the pavement damage and, more importantly, identifying the means by which the pavement damage can be reduced.

Weigh-in-motion (WIM) technology has experienced considerable progress in the past 10 years. Several new WIM systems have been developed with various levels of cost and expected reliability. The main objective of these WIM systems has been to measure the dynamic loads imparted by the vehicle traveling at highway speed. In the process of measuring the dynamic loads, several factors were found to significantly influence the WIM data: degree of road roughness, vehicle speed, load level, and the WIM calibration process. Several studies currently are under way to improve the quality of the WIM data and establish uniform calibration techniques. The study reported in this paper represents an effort to investigate the reliability and repeatability of the WIM data and its correlation to pavement response.

In the case of pavement response characteristics, there are two basic approaches to handling this problem: theoretical modeling and in situ instrumentation. Several computer models are available for computing stresses, strains, and displacements in layered systems. Theoretical pavement responses can

P. E. Sebaaly, Department of Civil Engineering/258, University of Nevada, Reno, Nev. 89557. N. Tabatabaee, Department of Civil Engineering, Sherif University, Tehran, Iran.

be computed if the materials are characterized properly and the loading conditions are well defined. In most cases, material properties are very difficult to define and loading conditions are assumed static for the purpose of simplifying the analysis. In situ instrumentation of pavement systems offers an alternative approach by which the actual pavement responses are measured without making any simplifying assumptions. In this research the strains at the bottom of the asphalt concrete (AC) layer were measur~d by strain gauges installed in the wheel track of the test section.

OBJECTIVE

The objectives of the res_earch presented in this paper can be summarized as follows:

• To study the effect of vehicle speed on the variability and magnitude of the WIM data,

• To investigate the effect of vehicle speed on the variability and magnitude of tensile strains at the bottom of the AC layer, and

• To investigate any correlations between the effect of vehicle speed on the WIM data and on in situ pavement strain.

PAVEMENT SECTIONS

A flexible pavement section was constructed at the Pennsylvania State University test track. The following table gives the properties of the test section as evaluated from falling weight deflectometer (FWD) testing:

Layer Description

AC surface Crushed aggregate base Sub grade

Thickness (cm)

15 20

381

Moduli (MPa)

2,550 207 152

The variations in axle load, axle configuration, and vehicle speed implemented in this program yielded a wide range of pavement responses and dynamic load magnitudes that provided for an extensive evaluation program.

TESTING PLAN

The objective of this study was to evaluate the variations in the dynamic loads and the strain response at the bottom of the AC layer as a function of speed, axle load level, and axle configuration. Earlier studies have shown that the effect of

108

tire pressure on the strains at the bottom of medium to thick AC layers, such as the one evaluated here, is insignificant (1). Therefore, only the following test conditions were varied:

• Load level: empty, intermediate, and fully loaded; •Axle configuration: single-drive axle, front tandem, and

back tandem; and •Testing speed: 32, 56, and 80 km/hr.

Four replicate measurements were collected for each combination of test variables. One full measurement consists of measuring the dynamic loads of the individual axles with one WIM system and the strain response at the bottom of the AC layer with two strain gauges. The two strain gauges were installed at different locations along the longitudinal direction of the test section. The tire pressure remained constant at 861 kPa.

LOAD MEASUREMENT

The WIM system of the Pennsylvania Department of Transportation (PennDOT) was used to evaluate the dynamic load variations at the test sections. This is a Golden River portable system (2). The t~st truck (tractor-trailer combination) was loaded at the three levels-empty, intermediate, and fulland the individual axles were weighed statically and as they ran over the WIM pads. Four replicate measurements were taken for each combination of vehicle speed, axle load, and axle configuration. The WIM system was placed after the test sections to avoid dynamic excitation of the test truck as it approache.s the test section. Table 1 gives a summary of the


data for the empty load level with the single- and tandemaxle configurations: that, at the speed of 80 km/hr, the dynamic axle load levels deviate the most from the static load levels for all of the axles (i.e., single-axle and front and back tandem axles). Tables 2 and 3 present the data under the intermediate and fully loaded levels, respectively; they show a constant trend where the load on the back tandem axle decreases as the speed increases. Figures 1 through 3 show the relationship between the coefficient of variation and speed for the single-drive and tandem axles, respectively. The data in these figures indicate that the variability of the dynamic loads at the empty load level is highly dependent on the speed. At the intermediate and full load levels, the effect of speed is insignificant except in the back tandem axle case (Figure 3).

The difference between the static and dynamic loads is another important factor when considering pavement loading. The majority of the pavement design and analysis procedures consider static loads. Therefore, the differences between static and dynamic loads would indicate how conservative or nonconservative these assumptions are. In this analysis, the difference between static and dynamic loads is calculated as the mean of dynamic loads at a given speed minus the static load. Figures 4 through 6 show the difference between static and dynamic load data as a function of vehicle speed for all three axles. The differences are expressed in arithmetic values instead of absolute values to differentiate among the cases in which the dynamic loads are higher or lower than the static load.

The data in Figures 4 through 6 indicate that, in the majority of cases (17 out of 27 combinations), the dynamic loads are higher than the static loads (i.e., a positive difference). It is

TABLE 1 Dynamic Loads from Single-Drive and Tandem Axles for Empty Load Level

Speed Single Front Back (km/h) Axle Drive Tandem Tandem

0 40 14 15

32 44 25 23 32 42 25 23 32 44 25 23 32 38 26 22

Mean 42 25 23 STD 2.6 0.5 0.5 CV (%) 6 2 2

56 40 22 22 56 49 26 22 56 49 30 24 56 43 23 21

Mean 45 25 22 STD 4 3 1 CV (%) 9 12 5

80 46 34 30 80. 48 43 31 80 46 41 25 80 66 48 30

Mean 52 42 29 STD 10 12 3 CV (%) 19 29 10

TABLE 2 Dynamic Loads from Single-Drive and Tandem Axles for Intermediate Load Level


0 59 61 45

32 53 43 57 32 52 44 52 32 53 45 57 32 53 43 55

Mean 53 44 55 STD 0.5 1 2

. CV (%) 1 2 4

56 55 50 47 56 63 59 43 56 59 55 45 56 61 56 37

Mean 59 55 43 STD 3 4 4 CV (%) 5 7 9

80 69 71 46 80 64 64 34 80 67 69 33 80 68 68 33

Mean 67 68 36 STD 2 3 6 CV ( % ) 3 4 17

TABLE 3 Dynamic Loads from Single-Drive and Tandem Axles for Fully Loaded Level


·o 81 103 78

32 71 81 76 32 75 89 78 32 70 83 77 32 75 87 77

Mean 73 85 77 STD 2 3 1 CV (%) 3 4 1

56 80 93 88 56 80 92 90 56 84 91 91 56 80 90 88

Mean 81 91 89 STD 2 2 1 CV (%) 2 2 1

80 87 82 79 80 88 83 78 80 92 89 86 80 86 84 82

Mean 88 85 81 STD 2 3 4 CV (%) 2 4 5

110

also seen that, at the empty -load level, the dynamic loads are always higher than the static values. In the intermediate and fully loaded cases, the differences are evenly distributed between negatives and positives. This distribution indicates that, at the empty load level, the test truck is experiencing a higher dynamic effect than at the other two load levels.

STRAIN MEASUREMENT

This testing program was conducted on an instrumented flexible pavement test section with the properties shown earlier. The instrumentation of the test section consisted of strain gauges at the bottom of the AC layer in the longitudinal direction. The gauges were located in the outer wheel track at various stations throughout the sections. The strain gauges

35

30

~ c: 25 0

~ .iij 20 > 0 c 15 CD ·o :e 10 CD 0 (.)

5

~---: ~

0 0 20 40 60 80 100

Vehicle Speed (km/h)

1-- Empty --+-- Intermediate - Full

FIGURE 1 Effect of vehicle speed on variability of dynamic loads, single-drive axle.

35

30

~ c: 25 0

~ "iij 20 > 0 c 15 CD ·u :e 10

CD 0 (.)

5 _,.

0 0 20 40 60 80 100


1-- Empty --+-- Intermediate -. Full

FIGURE 2 Effect of vehicle speed on variabHity of dynamic loads, front tandem axle.


were of the H-gauge type, which are installed during the construction of the test section (Figure 7).

The data analyzed in this paper were collected from strain gauges located at Stations 10 and 29. The station number indicates the distance in feet from the beginning of the section. Table 4 gives the strain measurements under the single-drive axle and 861 kPa of tire pressure. In the case of a tandem axle, unlike the WIM data, only the strain values under the back tandem axle were extracted from the actual measurement. The data indicated that the maximum tensile strain at the bottom of the AC layer always occurred under the back tandem axle (3). Table 5 summarizes the strain data under the back tandem axle.

It is well known that the strains at the bottom of the AC layer are highly sensitive to the pavement temperature at the time of testing. In this study, the pavement temperature was

35

30

~ c: 25 0

~ .iij 20 > 0 c 15 CD-

"(3

:e 10 CD

~ 0 ()

5

0 0 20 40 60 80 100


1-- Empty --+-- Intermediate -. Full

FIGURE 3 Effect of vehicle speed on variability of dynamic loads, back tandem axle.

25

z 20 ~ "O

15 ra .3 10 0 5 ~

Ci5 0 "O

-5 ra 0

....I 0 -10 ·e

-15 ra c: >. 0 -20

-25 0 20 40 60 80


1-- Empty --+-- Intermediate - Full

FIGURE 4 Effect of vehicle speed on difference between dynamic and static loads, single-drive axle.

100

Sebaaly and Tabatabaee

25

z 20

~ "'O

15 ro .9 10 (.)

5 1a ti5 0 "'O

-5 ro 0

...J (.) -10 .E

-15 ro c: >-0 -20

-25 0 20 40 60 80

Vehicle Speed (km/h}

1--- Empty -+- Intermediate - Full

FIGURES Effect of vehicle speed on difference between dynamic and static loads, front tandem axle.

25

z 20

~ 15 "'O ro .9 10 (.)

5 ~ ti5 0

I

"'O -5 ro

.9 (.) -10 .E

-15 ro c: >-0 -20

-25 0 20 40 60 80

Vehicle Speed (km/h}

1--- Empty -+- Intermediate - Full

FIGURE 6 Effect of vehicle speed on difference between dynamic and static loads, back tandem axle.

100

100

measured throughout the various layers. However, for the data analysis presented in this paper, the temperature effect is insignificant because the various replicate measurements at all three speeds were collected within 15 min. Average pavement temperature is not expected to vary significantly within 15 min; therefore, no temperature adjustment was necessary. The effect of the transverse location of the truck relative to the strain gauges was handled by using an ultrasonic distancemeasuring device and accepting only the replicates that were within 2.5 cm of each other.

The analysis of the strain data follows the same procedure as that for the WIM data. Because the two strain gauges are located at different stations, the effect of vehicle speed on the strains at various locations can also be compared. The data in Tables 4 and 5 indicate that the effect of speed on the variability of strain is insignificant. However, the truck speed has a tremendous effect on the magnitude of the measured .

end anchors

~

,J i

strain gauge embeded in

epoxy-fiberglass

10mm

102mm

FIGURE 7 Typical H-type strain gauge.

111

15mm

strain. Figure 8 shows the relationship between the strains at Station 10 and the vehicle speed for all load levels. By varying the speed from 32 to 80 km/hr, the strains are reduced by 50 percent in almost all cases. Theoretically, the effect of the viscoelasticity of the AC layer can be the reason for this large reduction in the strains. Because of the viscoelastic nature of the AC material, the material will show stiffer behavior under shorter loading times. The shorter loading times occur at higher speeds, which explains the observed large reductions in the strains under higher speeds.

Another way to assess the effect of vehicle dynamics on pavement loading is to compare the measured strains at the two locations as a function of speed for all load levels. For this analysis, the nonuniformity of the pavement material between the two stations must be taken into account. For this purpose, FWD testing was conducted at both stations, and the layer moduli were backcalculated. This analysis indicates that the difference between the material properties at the two stations is very small and would have an insignificant effect on the strains. Figure 9 shows the percent difference in strains as a function of vehicle speeds for the intermediate and full load levels. The data from the empty load level were inconsistent. The data in Figure 9 indicate that the strains at various locations in the section are highly dependent on the speed. However, the major significance of the speed occurs between 32 and 56 km/hr. These data clearly indicate the significance of truck dynamics on the response of the pavement.

EFFECT OF VEHICLE SPEED ON PAVEMENT LOADS AND PAVEMENT RESPONSE

This analysis will investigate the possibility of relating the effect of speed on both the loads and the strains. Because they represent the pavement response, the strains should be directly influenced by the magnitude of the applied load. However, strains in the AC layer are highly dependent on the properties of the material. Therefore, the thought relationship may not be easily found, especially because AC material is viscoelastic in nature.

The load and strain (Station 10) data under the single-drive axle are analyzed in this section. The objective here is to prove whether there is any correlation between the effect of speed on load and strain. Load levels were measured at 0, 32, 56, and 80 km/hr, whereas strains were measured only

TABLE 4 Longitudinal Strains at Bottom of AC Layer with Single-Drive Axle and 125-psi Tire Pressure

LOAD LEVEL

Speed Empty Intermediate Full

Sta 10 Sta 29 Sta 10 Sta 29 Sta 10 Sta 29

32 74 115 148 159 284 354 32 83 118 153 174 319 368 32 81 130 158 174 357 392 32 84 118 140 159 315 373

Mean 81 120 150 166 319 372 STD 4 6 7 7 26 13 CV ( % ) 5 5 4 4 8 4

56 73 73 102 137 218 283 56 80 78 91 130 218 278 56 78 83 98 130 220 280 56 77 83 99 132. 220 282

Mean 77 80 98 / 132 219 280 STD 3 4 4 3 1 1 CV (%) 4 5 41 2 1 1

80 58 69 60 81 127 166 80 60 69 61 83 136 169 80. 56 66 60 81 136 179 80 53 64 60 88 145 179

Mean 57 67 61 83 136 173 STD 3 2 1 3 6 6 CV (%) 4 3 1 4 5 3

TABLE 5 Longitudinal Strains at Bottom of AC Layer with Tandem Axle and 125-psi Tire Pressure

LOAD LEVEL

Speed Empty Intermediate Full

Sta 10 Sta 29 Sta 10 Sta 29 Sta 10 Sta 29

32 34 56 133 142 292 338 32 43 56 133 135 283 311 32 39 64 132 154 272 336 32 42 59 127 140 280 299

Mean 40 59 131 143 282 321 STD 4 3 2 7 8 17 CV (%) 10 5 2 5 3 5

56 17 27 98 118 186 239 56 18 25 94 118 188 249 56 18 27 91 122 179 234 56 17 25 94 118 185 238

Mean 18 26 94 119 185 240 STD 1 1 3 2 4 5 CV (%) 4 5 3 2 2 2

80 22 17 66 88 80 122 80 24 20 66 91 85 125 80 20 20 74 91 84 130 80 21 17 73 93 98 135

Mean 22 18 70 91 87 128 STD 2 1 4 2 8 5 CV (%) 7 7 6 2 9 4

Sebaaly and Tabatabaee

under speeds of 32, 56, and 80 km/hr. Therefore, the 32-km/hr speed is used as the base level for both load and strain data. At each speed level, the percentage of difference is evaluated for both the load and the strain. The percentage of difference is defined as the difference between the measurement (i.e., load or strain) at any speed minus the measurement at 32 km/ hr divided by the measurement at 32 km/hr multiplied by 100.

Figure 10 represents the effect of speed on the measured load and strain. The figure shows that the speed has a significant effect on both the load and the strain. However, the two effects are reversed. In the case of load, the higher the speed, the larger the measured load; whereas, in the case of strain, the higher the speed, the lower the measured strain. This indicates that, even though a vehicle speed increase, vehicle dynamics, and, therefore, higher dynamic loads are generated, the viscoelastic behavior of AC materials greatly outweighs this effect.

0 500

0 ! 400 (.) <: (5'

300 E 0 :i= 0

200 CD Q)

£

~ ca 100 c: ·~

Ci5 0 -----0 20 40 60 80 100 Vehicle Speed (km/h)

- Single, Empty -1- Single, Int. - Single, Full

-a- Tandem, Empt - Tandem, Int. __..._ Tandem, Full

FIGURE 8 Effect of vehicle speed on tensile strain at bottom of AC layer.

50

l 40

(/) c: ·~ 30 Ci5 .£ Q)

20 0 c: l!? Q) :i:: i5 10

0 0 20 40 60 80 100


- Single, Int -1- Single, F ---+-- Tandem, I -a- Tandem,

FIGURE 9 Effect of vehicle speed on strains at various locations within test sections.

113

EFFECT OF FACTORS ON PAVEMENT DESIGN

The objective of this analysis is to evaluate the effect of the various factors on pavement design and analysis. First, it i's necessary to define the uses of the WIM load data and the strain data in the pavement design process. The WIM load data are widely used to predict the 80-kN equivalent singleaxle loads (ESALs) generated by the passage of the weighted axle. These ESALs are generated on the basis of the AASHTO load equivalency factors (LEF) (4). The strain data can also be used to predict ESALs that are based on a mechanistic fatigue criterion, such as the one recommended by Finn et al. (5).

N1 ~ 15.947 - 3.291 log(E) - 0.854 log ( 1~3) (!)

where:

N1 = number of ESALs needed to cause fatigue failure, e = tensile strain at bottom of AC layer (microstrain),

and E = modulus of AC layer.

Therefore, by using the WIM load data and the strain data, two types of LEFs, one using the AASHTO approach and one using the fatigue criterion, can be evaluated. The fully loaded case of the single-drive axle had a mean value of 81 kN at 56 km/hr (Table 3), which is very close to 80 kN. Therefore, the fully loaded case of the single-drive axle at 56 km/hr is considered the standard axle load. To obtain AASHTO LEFs on the basis of the WIM data, the structural number (SN) and the ~erminal serviceability of the test section are needed. The SN is evaluated as follows: ·

(2)

where D; is thickness of layer i and a; is layer coefficient for layer i. In this case, the SN is calculated to b~ 4.0 and the

100

"2 80

~ 60 0 40

"tJ 111 ~ 0 20 d ~~ Q) 0 0 .:::::::: c:: l!! -20 Q) = 0 -40 "E Q) -60 e Q) '

-80 Cl..

-100 40 50 60 70 80 90 100


- Strain, Empty Loa -+- Strain, Int. Load - Strain, Full Load

-s- Load, Empty Load ~ Load, Int. Load .......- Load, Full Load

FIGURE 10 Comparison between effect of vehicle speed on dynamic loads and strains.

114

terminal serviceability index (PSI) is assumed to be 3.0. Using the SN of 4.0 and terminal PSI of 3.0, AASHTO LEFs for the single-drive axle were obtained:

Speed Load Level

(km/hr) Empty Intermediate Full

32 0.11 0.254 0.74 56 0.14 0.39 1.00 80 0.23 0.58 1.34

The LEFs based on the fatigue failure criterion were evaluated as follows:

• Assume that the intermediate load test at 56 km/hr is the standard load.

•Use the strain values from Table 4 and the AC moduli given earlier in the fatigue failure criterion to evaluate the N1 for each combination of speed and load level.

• Finally, evaluate the LEF as the ratio of the N1 at any combination of speed and load level to the N1 value at the 35-km/hr speed and intermediate load level.

The following table summarizes the values of the LEFs for the single-drive axle on the basis of the fatigue failure criterion:

Speed Load Level

(km/hr) Empty Intermediate Full

32 0.04 0.29 3.44 56 . 0.03 O.o? 1.00 80 0.01 0.02 0.21

A comparison of the data in the preceding tables indicates that there are significant differences between the two types of LEFs. The WIM data indicate that the LEFs increase as the speed increases, whereas the strain data indicate the opposite. As a result of this contradiction, major differences can be expected in designing pavement structures on the ·basis of the AASHTO design guide or the mechanistic approach.


In this paper, the effect of vehicle speed on both the measured load and pavement strain response has been analyzed. The WIM technology was used to measure the dynamic loads imparted by the moving vehicle. The pavement strain response under dynamic loads was measured by strain gauges embedded into the AC layer. On the basis of the analysis of the data, the following conclusions can be drawn.

• The WIM data indicate that in the majority of the cases the dynamic loads are higher than the static loads. It was also shown that at the empty load level, the truck would experience a higher dynamic effect than at the intermediate and full load levels.

• The variability of the dynamic loads at the empty load level (i.e., coefficient of variation of four replicates) is highly dependent on the speed. At the intermediate and full load levels, the WIM measurements were more repeatable at various speeds.

• The strain data indicate that the speed has a significant effect on the strain response of flexible pavement. By increasing the speed from 32 to 56 km/hr, the tensile strains at the bottom of the AC layer are reduced by 50 percent.


• The strain data indicate that the strain response at various locations of the road is highly dependent on the vehicle speed. However, the major significance of the speed occurs between 32 and 56 km/hr. Both the WIM and strain data indicated that the effect of vehicle dynamic are more significant at the empty load level.

• The LEFs analysis revealed very interesting facts about the discrepancies between using the WIM data and the use of the mechanistic approach. This contradiction between the two approaches has a great impact on the current practice in pavement design and analysis. Currently the majority of highway agencies follows the AASI:ITO design guide for the design of new pavement, whereas overlays are being designed by either an empirical approach or a mechanistic analysis. The data given in the preceding tables indicate that by using the WIM data and the AASHTO LEFs, a very conservative estimate o~ ESAL is obtained compared with the fatigue failure criterion approach. For mechanistic overlay analyses, there is an even more serious problem. The majority of the mechanistic overlay design procedures currently used by highway agencies are based on theoretical analysis by which the strains are evaluated. The computed strains are then used in a fatigue failure criterion to predict the pavement life. Finally, the predicted pavement life in terms of the number of ESALs is compared with the expected ESALs obtained from WIM data or other traffic analyses. This analysis process contains two contradictory approaches that shoulc;l not be combined because of their inconsistencies.

•Finally, on the basis.of the analysis presented in this paper, it is evident that more rational pavement analysis models should be investigated. The ideal pavement analysis model should consider the dynamic nature of traffic loads and the viscoelastic properties of the AC material. In addition, the current practice of using .the WIM .data with both new design and overlay design procedures must be seriously investigated.

ACKNOWLEDGMENTS

The authors thank FHW A for the financial support for the project and the Pennsylvania Department of Transportation for providing the WIM system.

REFERENCES

1. Sebaaly, P. E., and N. Tabatabaee. Effect of Tires and Pressures on Pavement Performance. Final report submitted to Goodyear Tire and Rubber Company. Report PTI9014. Pennsylvania Transportation Institute, University Park, Oct. 1989.

2. Sebaaly, P. E., W. Cunigan, and T. Chizewick. Truck Weight Data Processing, Storage and Reporting. FHWA-PA-89-040:89-11. Pennsylvania Department of Transportation, Harrisburg, July 1990.

3. Sebaaly, P. E., N. Tabatabaee, B. T. Kulakowski, and T. Sullivan. Instrumentation for· Flexible Pavements-Field Performance of Selected Sensors. Final Report, Vols. 1 and 2. FHWA-RD-91-094. FHWA, U.S. Department of Transportation, Oct. 1991.

4 .. Guide for Design of Pavement Structures. AASHTO, Washington, D.C., 1986.

5. Finn, F., C. L. Saraf, R. Kulkarni, K. Nair, W. Smith, and A. Abdullah. NCHRP Report 291: Development of Pavement Structural Subsystems. TRB, National Research Council, Washington, D.C., 1987.



Results of Weigh-in-Motion Project in France: 1989-1992

B. JACOB, c. MAEDER, L. A. GEORGE, AND M. GAILLAC

A national research and development project began in France in early 1989 on weigh-in-motion (WIM) techniques and devices, under the leadership of the Laboratoire Central des Ponts et Chaussees. The objectives, organization, and main results of this project are presented; a new generation of WIM station recently developed by Electronique Contr6le Mesures within this project is described in detail; the application of WIM by a major French motorway company (Cofiroute) is explained; and some new concepts for the acceptance of WIM stations are discussed.

A weigh-in-motion (WIM) project was prepared during 1988 in response to expressed national needs. The Bridge, Road, and Transport Administration represented by its technical institutions [such as the Public Works Research Laboratories-Laboratoire Central des Pon ts et Chaussees (LCPC) and its regional offices, LRPCs-and Service d'Etudes Technique des Routes et Autoroutes (SETRA)], the major motorway companies (including Cofiroute), and French manufacturers [including Electronique Controle Mesures (ECM)] had acquired so.me experience since the late 1970s, in vehicle WIM. These research works were developed for various purposes, such as pavement and bridge design and maintenance and statistical knowledge of traffic loads or enforcement.

After more than 14 years of development of WIM techniques using piezoelectric ceramic cables in France (1-3), it became useful to collect and compare the experiences of the various participants and to more fully consider the growing needs of the customers. It was also necessary to design and implement a new generation of WIM stations with the latest improvements in electronic computer hardware and software, and also signal processing of piezoelectric sensors. Therefore, it was intended to make the most of the experience acquired, to rationalize and coordinate research and development resources, and above all to bring all the participants together around common objectives by taking advantage of their complementary natures.

The project responds to three main demands of Public Works and Transport:

• Pavement design and maintenance on highway and motorway networks, revi~ion of the pavement design code, knowledge of the aggressiveness of heavy traffic, and the provision of some tools to assist repair and maintenance policies;

B. Jacob, Laboratoire Central des Ponts et Chaussees, 58 boulevard Lefebvre, 775015 Paris, France. C. Maeder, Electronique Controle Mesures, 4 rue du Bois-Chene-le-Loup, Pare d'Activites de Brabois, 54500 Vandoeuvre-les-Nancy, France; ECM, Inc., 15635 Vision Drive, Plugerville, Tex. 78660-3203. L.A. George and M. Gaillac, Cofiroute, 42 avenue Raymond Poincare, 75116 Paris, France.

•Bridge design and maintenance, including revision of existing bridge loading codes and the preparation of the new Eurocodes, the checking of some particular or exceptional projects, and fatigue studies for steel or composite bridges under traffic loads; and

• The survey and control of loads on roads to enforce the law, and the gathering of statistical data for economic and safety purposes.

ORGANIZATION OF WIM PROJECT

The WIM project was conducted by the project manager of LCPC and the project assistant manager of LRPC in Trappes. The following organizations participated:

•Public works laboratories and technical centers (the Ministry of Transportation);

•Motorway companies: Cofiroute, Autoroutes du Sud de la France, and Union des Societes d'Autoroute a Pesage (USAP); and

•Manufacturers and private companies: ECM, Drouard (a company that installs sensors in the roads), and Alea tel.

The project was divided into five working groups, whose objectives follow:

•Working Group 1. Development of a new Hestia WIM station.

•Working Group 2. Research on piezoelectric sensors, resin, and installation techniques, signal processing, and calibration.

•Working Group 3. Multiple-sensor WIM system. •Working Group 4. New types of WIM sensors and optic

fibers. •Working Group 5. Customer requirements and project

evaluation.

The total budget is roughly 12 million FF ($1.00 U.S. 5.88 FF, 1993) in_cluding more than 70 percent in personnel costs. Sixty percent of the budget is covered by the Ministry of Transport through the LCPC; the rest is funded directly by nongovernmental participants. Altogether, about 40 engineers and technicians are working part time on the project. The planned project lifetime is 4 to 5 years, and the project was completed in 1993.

RESULTS

The first results were presented in 1991 (4). The following are the main results obtained after 4 years:

116

Working Group 1

In the planning, design, and development of the new Hestia WIM station by ECM, the new station was built according to specifications prepared by the French administration in 1988. It satisfies the demands of the customers and takes advantage of the most recent developments in electronic computer hardware and software and experience in piezoelectric sensors and signal processing. Some tests have been made since 1990 with calibrated trucks and real traffic vehicles to finalize the Hestia station. Now more than 25 of these stations are in operation around the world. Additional technical details are given later.

Working Group 2

Significant improvements of the WIM technique by piezoelectric sensors were obtained up to now. A simple new algorithm was achieved for signal integration to improve accuracy and avoid some double- or triple-axle misdetection. An automatic self-calibration procedure was introduced by the Centre d'Etudes Technique de l'Equipement (CETE) de l'Est, using some characteristics of common heavy trucks, and was implemented in Hestia.

To optimize response quality and the durability of sensors in the pavement, studies and comparisons of various resins were made for mounting the sensor in the road. These studies led to sensor installation and calibration guidelines specific to the pavement properties (5). Quality-control tests of the sensors before and after specific to installation are also being developed.

Working Group 3

To reduce the variance of the measured axle loads and close the gap between the dynamic and static loads that are caused by pavement roughness and vehicle vibrations, a multiplesensor WIM system was devised, using more than two sensors per lane. An advanced signal-processing technique based on a weighted linear regression and a learning set of vehicles was developed to combine the individual records of one axle on each sensor and to properly estimate the static load; this procedure was tested by simulation ( 6) and is now being tested on real roads.

Working Group 4

A new type of sensor with optic fibers may offer at a low cost a powerful WIM system and provide more information than existing sensors. A feasibility study was conducted with onroad testing and real vehicles, which showed the ability of such a system to measure axle loads with very good sensitivity. A cooperative effort with Alcatel-Cables is underway to develop a prototype of a new optical WIM system.

Working Group 5

The requirements of all WIM customers were collected, especially on data transmission and remote control of WIM


stations. New devices were then adapted to make them compatible with existing data transmission systems.

Motorway companies have played a large part in the development and testing of the Hestia station. Recommendations for WIM location and suitable pavement characteristics for measurements with an acceptable or high standard of quality have been published (7).

A data base of traffic load data collected on all types of roads was created at LCPC in 1990. More than 150 traffic records over continuous periods between 1 and 8 weeks, recorded at more than 50 WIM locations since 1982, are already available. They contain altogether data on more than 2 million trucks.

SECOND GENERATION OF WIM SYSTEM: HESTIA

Layout of Sensors on Pavement

One induction loop and two LCPC-patented piezoelectric sensors (Type E) are mounted per traffic lane. They are laid out as shown in Figure 1.

General Presentation of Hestia Station

The basic design of the station was determined by three criteria, which since 1988 appear to be essential:

1. Use of one intelligent detector (DU) per traffic lane to collect the data relative to each vehicle;

2. Use of a central unit (UC) to manage all the intelligent detectors and process their data to provide information according to customer requirements (the UC also communicates with the outside); and

3. Use of a standard European format.

SENSORS ,,,.'' -----....,,

I I

I - - - - ~·@!!I _ - - - - ... O_ --1

I I I l.OOm I i- ----,

j l.OOm I -----1

•L _. - ll - - _1!.. - - - L - -s __ IN_SPECllON - - - - - - - - - - - - HOLE

- - CABINET

FIGURE 1 Installation of WIM station.

Jacob et al.

The structure of the Hestia station is shown in Figure 2. The Hestia DU detectors (2 and 4) receive the information from the piezoelectric sensors of Lanes A and B and from the loop· detectors connected to the loops. In addition, each detector receives information termed anticoincidence from its right adjacent lane. Each detector is made of two cards: a four-layer digital card and a two-layer analog card. For each vehicle passing in Lane A, the Hestia DU detector provides all the information given in Table 1.

The system works by sampling during the induction loop detector switching time. The speed is derived from time Tl

/ /

FIGURE 2 Design of Hestia.

TABLE 1 Output of Station

Measuring elements Unit

date d-m-y

hour h-mn-s

choice of vehicle sensor 110

number of lanes 0 to7

validation character decimal

category 0099

speed (km/h) mph

inter-vehicle time ms ors

time spent on loop ms

nuinber of axles decimal

s 20

total weight 100 x lb

0.1 T

weight of each axle 100 x lb

0.1 T

inter-axle distance. ft: 100

117

and the distance between the sensors. The distances between axles are calculated from the times T2, T3 . . . , and from the vehicle speed. The vehicle category is deduced from distances and weights. The RS232 serial link, the connector of which is located on the front panel, allows communication with the detector. The DU-UC dialog is carried out on the bus by break management with a system to avoid any data interference and to allow a very high throughput. Simulation trials carried out show that four vehicles, each having five axles, passing simultaneously on four different lanes, are handled by the system without any perceptible delay.

WIM Detector WIM : A VC Central Unit

0:0 x

0:0 x

X:O 0

x : x 0

x : x 0

X:X 0

X:X 0

x : x 0

x : x 0

X:X 0

X:O 0

X:O 0

X:X 0

118

The UC interacts with a maximum of eight lane detectors, with the external memories (711 to 7/8) made up of one to eight 1-megabyte (M) cards, backed up by lithium cells, with the Alarm card, and with the outside via the RS232 (or using 300- to 9600-baud modems). Only the UC has a battery-backed real-time clock. Each DU detector and the UC are driven by a 16-bit CMOS 80C186 microprocessor. This unit provides a visual display of the traffic in real time and statistics in four user formats, some of them compatible with Lotus 1-2-3. It also records, if needed, a file with the results vehicle by vehicle and lane by lane.

The power supply is provided by direct current (DC)/DC convertors (15) with electrical voltage decoupling and integrated smoothing from an 85-Ah load battery. The battery charger can be supplied by 220 or 115 V (50 or 60 Hz) mains or by an 880- x 445- x 36-mm solar panel.

The Hestia station is available in two versions: a doublecased fixed system for a maximum of eight traffic lanes and 8 M of memory, and a portable system for a maximum of two lanes and 5 M of memory. The station can be integrated into a traffic management network.

Signal Processing for Axle Load Calculation

Signal processing is carried out independently on each of the two piezoelectric sensors. It is therefore possible to compare the results obtained from each of the two sensors to determine whether the speed was correctly measured.

In addition, the use of two sensors improves the accuracy, either by averaging the two measurements or by selecting the more accurate sensor. Axle loads are computed with the help of the surface pulse produced during the passage of an axle.

Automatic Self-Calibration Algorithm

The original automatic self-calibration algorithm mentioned earlier provides at regular intervals the value that must be allocated to the unit square to calculate the integral of the signal produced by each axle. This value determines the axle loads. The areas used for the load calculation are first corrected for the vehicle speed.

The automatic self-calibration algorithm originates from a statistical study of vehicle parameters. In this study, it was pointed out that the first axles of some vehicles (termed characteristic) whose gross weight is above 30 T have a very low dispersion of the load around the mean of 6.1 T (on French national roads); in addition, the gross weights of these vehicles are centered on 40 T (other values may be determined for U.S. traffic). Then a weighted moving average of these parameters is continuously computed and fitted on the given target values to make the self calibration.

Station Initialization

Detector parameterizing is carried out from the RS232 serial link with the help of a PC/AT-compatible microcomputer, either from command words or from the user-definable ECoM software. The parameters to be defined are the distances be-


tween sensors, the choice of calibration method (testing vehicle or self calibration with its constants and target values), the sensor(s) used for weighing, and some additional choices, such as the loop size to measure the vehicle length.

EVALUATION OF TRUCK AGGRESSIVENESS ON MOTORWAY NETWORK

Presentation of Cofiroute and Utility of Weighing

Since 1970 Cofiroute, a private toll highway company, has been building and managing a network that spreads toward the western and central parts of France (Figure 3). In 1992 its network consisted of 732 km, making it the fourth longest in France. It is made up of four highways: AlO from Paris to Poitiers, All from Paris to Le Mans and from Angers to Nantes, A81 from Le Mans to Vitre, and A 71 from Orleans to Bourges.

The evaluation of pavement structure conditions calls for a precise knowledge of the heavy traffic that has used it from the day it was put into service. For a toll highway network, it is relatively easy to know the volume of heavy traffic with the counting systems used for controlling toll receipts. However, this information does not usually determine whether the vehicle is loaded. Thus, the weight supported by the pavement remains unknown. Moreover, toll rates include in the same category different types of vehicles, the aggressiveness of which

. can be very different. Many campaigns using WIM systems have been completed

on the French national roads network. Nevertheless, nothing revealed the possible error in using the results of these campaigns to evaluate heavy traffic on the highways that were

FIGURE 3 French and Cofiroute's motorway network.

• Hestia axle 2

15% ~Saft axle 2

1.0% ~ Hestia axle 3

5%

~ 0% ctl .....

:;, (.> -5% (.> <(

-10%

-15%

-20%

2 3 4 5 6 7 8 9 10 11 12 13 14 15

Run number

FIGURE 7 Double-axle weights on three-axle truck.

Last, axle spacing obtained by Hestia is excellent because it is often within a range of 5 ot 7 cm (2 or 3 in.) above or below statical value. The maximum recorded value is 9 cm (4 in.).

Additional Statistical Results Obtained with Hestia

Other experiments were performed on French highways in 1992. They gave additional results about accuracy on traffic flow and speed. A comparison with videotape recording of 3,000 vehicles showed a difference of less than 0.4 percent on traffic flow and less than 0.3 percent with a 22-category classification. A sample of 150 vehi~les whose speed was previously measured by a standard radar gave 95 percent of speeds with an accuracy of 2 percent. No deviation was more than 3 percent, and accuracy was constant from 60 to 170 km/hr (37 to 110 mph).

Last, an experiment was made in July 1992 with three Hestia under the following conditions: (a) two sensors per station, (b) 30 m (98 ft) between each station, and (c) indepen.dent self-calibration for each station. The first results emphasize that multisensor weighing appears to improve accuracy and lessen deviation. To set an example, with only one sensor, maximum deviation was 13 percent for an axle and 8 percent for gross weight. With three pairs of sensors both maximum deviation and gross weight were reduced to 4 percent.

CONCLUSIONS

In addition to fulfilling the presented objectives and achieving the announced and expected devices, the goals of this project are to make WIM tools and their applications better known and to -provide national technical and financial support to research and development actions (R&D). Consequently, customers are much better informed about new developments; they are associated with the R&D actions and support

them. Their needs and demands are more seriously and more quickly considered during development. Manufacturers and suppliers also get substantial technical assistance from laboratories and the administration, which puts the road and highway networks and some technicians at their disposal for testing sensors and 'stations. Recent measurements obtained on a motorway network with a Hestia device have given satisfying results with respect to French recommendations for the acceptance of WIM stations. Similar results were obtained with a SAFT 2000. Therefore, France now has at its disposal two types of operational WIM station. By the end of 1992, Cofiroute, a private toll motorway company, had started the equipment of its network with two Hestia stations.

REFERENCES

1. Besnard, S., and M. Siffert. Les Cables Piezo-electriques: Une Innovation pour I' Analyse de Trafic (in French). Travaux, N'.572, Dec. 1982, pp. 84-88.

2. Jacob, B., and M. Siffert. A High Performance WIM System by Piezo-Electric Cables and Its Applications. Proc. 1st International Conference on Heavy Vehicle Weights and Dimensions. Kelowna, B.C., Canada, June 1986.

3. Siffert, M., and B. Lescure. Evaluation of Heavy Vehicle Traffic and Its Applications to Pavement Structural Design. Proc. 6th International Conference on Structural Design of. Asphalt Pavements, Vol. 1. University of Michigan, Ann Arbor, July 1987.

4. Jacob, B., et al. Compte-Rendus de la Journee Nationale de Presentation du Projet Pesage en Marche (in French). LCPC, Paris, France, Sept. 1991.

5. Projet Pesage en Marche: Recommandations pour le Chaix et la Pose des Capteurs de Pesage (in French). Rapport du SG 2. Paris, France, 1990 (revised 1992).

6. Eymard, R., F. Guerrier, and B. Jacob. Measurement of Axle Weights by Multiple Sensor Weighting. Proc., CERRA-ICASP'6, Mexico, June 1991.

7. Projet Pesage en Marche: Recommaridations pour le Chaix des Sites de Pesage et la Reception des Stations (in French). Rapport du SG 5. Paris, France, June.1992.

Jacob et al.

cisely, it needed to know which results could be obtained with Cofiroute's own pavement and traffic.

Here accuracy consists of a set of criteria that includes bias, average, standard deviation, and statistical range of measurements. The axis of ordinates in the following figures gives (in percentage points) for each test run the following quantity:

DW- SW SW

where DW is dynamic weight and SW is statical weight. On one of the company's best sites in terms of geometric

and pavement characteristics, two pairs of sensors were placed 100 m (330 ft) from each other. In this manner complementary measurements could be performed as well with a SAFT 2000, another French device. The experiment lasted 2 days, and the two stations worked alternately on each site. Five test vehicles were used: three two-axle trucks, one three-axle truck, and one four-axle articulated truck. For reference purposes, statical weights are given in Table 4.

Statical axle spacing values, where d;i is the distance between axle i and axle j, are shown in the following table:

Truck Number

1 2 3 4 5

Results

d12 [m(ft)]

3.20 00.50) 2.74 (9) 4.45 (14.6) 4.00 (13.1) 3.50 (11.5)

d23 [m(ft)]

1.35 (4.4) 7.50 (24.6)

d34 [m(ft)]

1.35 (4.4)

The Hestia station was connected the night before the beginning of the experiment to self-calibrate. First, there were a few test runs to calibrate each station. The target value for self-calibration of Hestia was increased by 5.2 percent after these runs. No further calibration was made after the experiment had actually started. Final results discussed here come from the average of the two sensors for Hestia and from only one sensor for SAFT 2000. The first remark concerns the effect of Hestia's self-calibration algorithm on results (Figures 4 and 5). Accuracy was clearly getting better in the afternoon

TABLE 4 Statical Weights

truck n° axle 1 axle 2 axle 3 axle 4 gross weight

1 1.25 2.70 3.95

[2700] [5900] [8700]

2 2.75 3.05 5.85

[6000] [6700] rt2900]

3 5.35 12.60 17.90

[11800] [27800] f39400]

4 5.50 9.65 8.55 23.85

[12100] [21300] [189001 f52600]

5 5.90 12.80 10.50 8.25 37.45

[13000] [282001 f23100] f18200] f82500l

Note: Weights are given in the thousands of kilograms (pounds in.brackets).

121

of Day 1 (Runs 7 and up). Besides, in a general way test runs of Day 2 give better results, particularly with the articulated truck, whose pattern is considered characteristic.

On Day 2, dynamic weights on single axles ranged from - 6 to 16 percent with Hestia and from - 15 to 11 percent with SAFT 2000 for the first axle of the three-axle truck; results were similar for the articulated truck (Figure 6).

An analysis of the results given for each elementary axle composing a double axle shows that they are within the range of Category B, except for two values, with the three-axle truck (Figure 7). The two stations give results in Category C with the articulated truck.

Two things are worth noting: first, accuracy is almost always greater on the last axle, particularly for the articulated truck; second, better accuracy is obtained if a multiple-axle weight is globally considered.

10%

53 II ~ 0% -··r-tl+l+l+m'·- -+-+-l ~ -5%

13 -10% (.) <( -15%

-20%

-25% 2 3 4 5 6 7 8 9 10 11 12 13

Run number

FIGURE 4 Effect of Hestia's self-calibration on accuracy, gross weight (Day 1), three-axle truck.

5%

~ 0%

m -5% :; 8 -10%

<( -15%

-20% 2 3 4 5 6 7 8 9 10 11 12 13

Run number

FIGURE 5 Effect of Hestia's self-calibration on accuracy, gross weight (Day 1), articulated truck.

15%

10%

~ 5%

m lo..

0% ::::i (.) (.) <( -5% I

!ill Saft 2000 axle 1 I -10%.

-15%

~ Hestia axle 2

0 Saft 2000 axle 2

2 3 4 5 6 7 8 9 10 11 12 13 14

Run number

FIGURE 6 Dynamic single-axle weights for articulated truck.

TABLE 2 Choice of Operational Site

Quality of site

Good Fair

Rutting (max. depth in mm) <5 < 10

Rigid and semi- Average deflection (in hundredths of < 10 <20

mm)

rigid pavement Difference between axis and edge +/- 3 +/- 10

(in hundredths of mm)

Flexible Average deflection (in hundredths of mm) <30 <60

pavement Difference between axis and edge +/- 5 +/- 10

(in hundredths of mm)

Evenness Short waves

(in APL 72 marks) Long waves

Acceptance of WIM Stations

Because the fundamental element most responsible for the accuracy of the results is the sensor-pavement pair, it is necessary to ensure that the sensors are well placed, both mechanically and electrically. In a general way, the authors refer to the recommendations for the placement of piezoelectric sensors, a guide published by the LCPC (5).

The calibration method followed is different for each category of station. For example, Category D involves checking vehicle flow, speed, and classification using a sample of 20 to 50 vehicles.

Other categories require a test method with one or many types of vehicles, for example:

• Category C: One truck with at least one multiple axle [a single axle of about 13 T (28 ,600 lb) and a double axle between 19 and 21 T (42,000 and 46,000 lb)];

•Category B: One truck with two or three axles, and one articulated truck with four or five axles [a single axle between 8 and 13 T (17 ,600 and 28,600 lb) and a double axle between 13 and 21 T (28,600 and 46,000 lb)].

>7 >6

>8 >7

First, each test vehicle is statically weighed: each axle must be weighed independently, then each axle group, and finally the whole truck. For multiple axles, all axles are simultaneously weighed; the operator must ensure that the sum of all axle weights is equal to the gross weight (within the accuracy of the weigh bridge used).

Second, each vehicle is run at a constant speed [between 40 and 90 km/hr (25 and 55 mph)] until 10 significant values are obtained. Tolerance is one value out of range for 10 significant measurements in Category C and for 20 in Category B.

Levels of tolerance concern the extreme values obtained. For weights, the reference is statical weight. These levels are summarized for each data item in Table 3.

Concrete Example: Results in Mer (July 1992)

Purpose and Conditions of Experiment

Before Cofiroute invests in a couple of WIM stations, the company wanted to test and evaluate the new Hestia according to the methodology described in this paper. More pre-

TABLE 3 Levels of Tolerance in Statical Weight for Categories B through D

For every significant run

data Cat.B Cat.C Cat.D

traffic flow per pattern +/- 3% +/- 5% +/- 10%

traffic flow +/- 1% +/- 3% +/- 5%

soeed +/-4% +/-6% >6%

axle-spacing +/- 10 cm +/- 30 cm +/- 50 cm

[0.3 ft] [l ft] f 1.6 ft]

single axle weight +/- 10% +/- 15% > 15 %

multiple axle weight +/- 10% +/- 15% > 15 %

elementary weight of each +/- 15% +/- 20% >20%

axle composing a multiple

axle

gross weight +/- 10% +/- 15% +/- 30%

Jacob et al.

being managed. Some considerations could even challenge such a hypothesis. For example, the proportion of long-distance traffic on highways will certainly involve a large number of trucks from countries where the legal axle load is lower than that in France (130 kN). Because of this, it was decided in the beginning of the 1980s to determine as exactly as possible and to follow up over a period of time the aggressiveness of heavy traffic. Only specific instrumentation makes it possible.

Equipment of Coflroute's Network

Using data from toll statistics, the network managed by Cofiroute was first divided into sections of homogeneous truck traffic. Therefore it was possible to localize a limited number of traffic sensors on each of these sections.

Because the cost of continuous measurements on each section were too high in 1981, it was necessary to determine a reliable sampling method for measurement campaigns. The analysis of toll statistics showed the optimum duration of onsite real traffic measurement campaigns and the best time of year. Systematic studies comparing toll statistics with the results of these campaigns are being completed. These studies will determine the values of aggressiveness coefficients applicable to each class of vehicles in the toll tariffing system, thus making it possible to know the number of equivalent standard axles independently of the toll statistics on each highway section of homogeneous traffic.

WIM Station Evolution

SAFT 16 (Station d' Analyse Fine du Trafic) was developed 12 years ago for research needs, but it was decided that it would be used for operational measurements on this network. Since then, other equipment has been developed in France, such as the SATL in 1983 [Station d' Analyse du Trafic Lourd (Heavy Traffic Analysis Station)], which can collect and store accumulated data for each type of axle load (single, tandem, and triple axles). Recently a second generation of stations has appeared, one that uses new electronics and computing devices, such as Hestia and SAFT 2000.

In 1992, a couple of new stations were used; they were more efficient than the old ones in the following ways:

•Greater accuracy in the whole range of axle load (from 10 to 150 kN);

•Bigger data storage capacity on the site; • Less electrical consumption; and •Greater ability to record traffic data (up to eight lanes).

These new stations are permanently installed on a highway section and linked through a communication network to the local maintenance center. These first devices will be quickly followed by others, as the highway network is equipped with a traffic management system. Most of these traffic stations will be able to distinguish only cars and trucks; calculate traffic flow, speed; occupancy rate, and so on; and send the aforementioned data to the maintenance center every 6 min. About 10 of them will be able to collect detailed axle loads and spacings, vehicle by vehicle (using piezoelectric sensors).

119

Evaluation of Needs and Acceptance of WIM Stations

Let us call a WIM station the piezoelectric sensors and the electronic system that analyzes the signal. Depending on the nature of the needs, the authors have adopted the methodology concerning the choice of a site and the acceptance of such a station, described previously (6). This methodology was defined in cooperation with the LCPC and the other WIM project members.

Choice of Operational Site

The needs can be summarized in the three points that follow, in order of increasing accuracy. Each corresponds to a category of WIM station:

•Category D: To have a basic idea of weights to classify vehicles using their load,

• Category C: To build histograms of loads, and • Category B: To determine as exactly as possible each axle

load or multiple-axle load and the gross weight.

(Category A would concern future stations with a higher level of accuracy.)

The choice of a site is closely bound to the purpose of the measurement (7). In view of these needs, the authors determine three classes of sites (fair, good, excellent) satisfying some geometric criteria and pavement characteristics (rutting, average deflection, and evenness). This evaluation is made from 200 m before the placing of sensors to 50 m behind.

Geometric criteria are the same for every class of site:

• Longitudinal gradient less than 2 percent, with no break in the slope,

•Cross-fall less than 3 percent, and •Radius of curvature greater than 1000 m (3,300 ft), but

a straight line is preferred.

Pavement characteristics such as evenness and cracks may affect the accuracy of the results. In addition, deflection and rutting have an impact on both the reliability and the life of the sensors.

Concerning deflection, the level of tolerance depends on the type of pavement, as given in Table 2 (10 mm = 0.39 in.). The following are noteworthy:

• The profilometer APL 72 gives a mark every 200 m ( 660 ft), ranging from 1 to 10 (10 means excellent); it is then necessary to determine more precisely the area in which the sensors should be placed to avoid a particular point (8).

• Deflection is measured every 20 m by measurements in the axis and the edge of the pavement; only the greater of the two values is kept, and then the average is calculated every 200 m.

A good balance between site and station is important too: even with a station in Category B, accuracy cannot be guaranteed on a fair site. On the other hand, it is useless to require a good site for a station in Category D.

transportation research recordonlinepubs.trb.org/onlinepubs/trr/1993/1410/1410.pdf ·...

Documents