U A) A) 04 UMTRI-82-45-1 INTERNATIONAL EXPERIMENT TO ESTABLISH CORRELATION AND STANDARD CALIBRATION METHODS FOR ROAD ROUGHNESS MEASUREMENTS M. SAYERS T.D. GILLESPIE C. QUEIROZ DRAFT REPORT TO WORLD BANK DECEMBER, 1982 THE UNIVERSITY OF MICHIGAN TRANSPORTATION RESEARCH INSTITUTE Technical Reprt Deceentaties Page 1. Repare me. 2. Gove me Accession Me. 3. Recipient's Ceeieg Mo. UMTRI-82-45 4. Titte wnd Suise 5. RewOt INTERNATIONAL EXPERIMENT TO ESTABLISH CORRELATION De December 1982 AND STANDARD CALIBRATION METHODS FOR ROAD 6. P0 h..ing Org..ia.io COJO ROUGHNESS MEASUREMENTS 8. Pemng Orgnsee Rert Me. 7- A"1I*0) M. Sayers, T.D. Gillespie, C. Queiroz UMTRI-82-45 9. Pee suung Organiteion Nome and Addren 10. Wwil Unit No. The University of Michigan Transportation Research Institute i. c.. GretN. 2901 Baxter Road Ltr. 5/6/82 Ann Arbor, Michigan 48109 13. T", .. Ropewo m pe,io, Cwe.d 12. Spmseeing Ageny Nom. and Addres Draft The World Bank 5/82 - 11/82 1818 H Street, N.W. Washington, D.C. 20433 14. Sonsoing ' Acy ce" -15 S.uppemwy new.s 16. Abswest An International Road Roughness Experiment (IRRE) was conducted in Brazil in May-June 1982. Vie purposes were to examine the correlations between different road roughness measurement equip- ments in use throughout the world, and to identify a standard roughness measure (an International Roughness Index) as a basis for calibrating and comparing these roughness data. The IRRE was a cooperative effort initiated by the World Bank and conducted by researchers from Brazil, England, France, and the United States, with the additional participation of equipment from Australia. The Experiment involved evaluation of the roughness on a wide range of paved and unpaved roads. The roughness was measured at a number of speeds by seven Response-Type Road Roughness Measuring Systems (RTRRMS). The longitudinal profiles were measured by a number of available methods to evaluate the profilometry techniques, and allow evaluation of various means for processing that information to analytically assign a roughness value to the surfaces. In addition, the road sites were evaluated subjectively by a rating panel for their roughness qualities. The data obtained show that correlations between any two RTRWISs are excellent when they are operated at the game test speed, and that a single equation is adequate to relate readings among devices for all types of roads. Four candidate methods for analyzing the profile measures were tested as calibration references for RTRRMSs, with various degrees of succese. The best available method proved to be a suitable calibration reference for nearly all RTRRMSs in use today, with the condition that separate calibrations be developed for certain road types. 17. Key WoMs road roughness, road profile 1 Desfshed. SteemenW measurement, ride quality, roadmeter serviceability, subjective rating, UNLIMITED profile analysis 19. Secuuity Cleesu. (of de repard 2. Seisy Clessf. (ei thia Vge) 21. . of Pe"s 22. Price NGNE NONE TABLE OF CONTENTS CHAPTER 1 INTRODUCTION. . . . . . . . . . . . . . . . . . . . . . . . . 1 Backg ound . . . . . . . . . . . . . . . . . . . . . . . . 1 Objectives of the Experiment. . . . . . . . . . . . . . . . 4 Report Organization . . . . . . . . . . . . . . . . . . . . 5 2 EXPERIMENT. . . . . . . . . . . . . . . . . . . . . . . . . . 7 Participants. . . . . . . . . . . . . . . . . . . . . . . . 7 Subjective Rating Study . . . . . . . .. . . . . . . . . . 14 Design of Experiment. . . . . . . . . . . . . . . . . . . . 16 Testing Procedure . . . . . . . . . . . . . . . . . . . . . 18 3 ANALYSIS AND FINDINGS . . . . . . . . . . . . . . . . . . . . 20 Correlation of RTRRMS Numeris. . . . . . . . . . . . . . . 23 Computation of Profile-Based Numerics . . . . ... . . . . . 31 Correlation of Profile-Based Numerics with RTRRMS Numerics . .. . ... .. . ... .. . . . . . . 35 Comparison of Profile Measurement Methods . . . . . . . . . 40 Comparison of Subjective Ratings with Roughness Measures. . . . . . . . . . . . . . . . . . . . 41 4 CONCLUSIONS AND RECOMMENDATIONS . . . . . . . . . . . . . . . 43 Correlation Between RTRRMSs in Use Today. . . . . . . . . . 43 International Roughness Index . . . . . . . . . . . . . . . 45 Recommendations . . . . . . . . . . . . . . . . . . . . . . 49 5 REFERENCES. . . . . . * * * * * . * . . . . . . . . . . . . 52 ACKNOWLEDGEMENTS The International Road Roughness Experiment (IRRE) reported here was sponsored by, a number of institutions: the Brazilian Transportation Planning Company (GEIPOT), the World Bank (IBRD), the Brazilian Road Research Institute (IPR/DNER), the French Bridge and Pavement Laboratory (LCPC), and the British Transport and Road Research Laboratory (TRRL). The Australian Road Research Board (ARRB) and the Federal University of. Rio de Janeiro (COPPE/UFRJ) provided roughness measuring equipment. The University of Michigan provided personnel and computer support through contract with the World Bank. Many individuals contributed towards the completion of the IRRE and the subsequent analyses reported here, and it would be impossible to mention here all of their names. However, the participation of the following people was invaluable for the success of this work: S.W. Abaynayaka, H. Hide, and G. Morosiuk formed the research team from TRRL; M. Boulet, A. Viano, and F. Marc formed the research team from LCPC; M.I. Izabel (GEIPOT) supervised the subjective rating study and aided in the data entry; I.L. Martins (GEIPOT), Z.M.S. Mello (IPR/DNER), and H. Orellana (GEIPOT) aided in the data entry and analysis; L.G. Campos (GEIPOT) was responsible for selection of test sites, and together with 0. Viegas (IPR/DNER) provided day-to-day supervision and control of the IRRE; M. Paiva (GEIPOT) repaired and calibrated the GMR Profilometer, and worked together with S.H. Buller (GEIPOT) to provide technical support during the IRRE. Aid in the planning of the IRRE was provided by an expert working group that included W.R. Hudson, R. Haas, V. Anderson, R.S. Millard, and W. Phang. A number of institutions contributed with comments, suggestions, and criticisms during the planning stage of the experiment, including the NITRR (South Africa), ARRB (Australia), the CRR (Belgium), the National Swedish Road and Traffic Research Institute, and the Institute for Transport Economics (Norway). CHAPTER 1 INTRODUCTION Background The roughness history of road surfaces has long been recognized as an important measure of road performance. As used in this report, the word "roughness" means the variations in surface elevation along a road that cause vibrations in traversing vehicles. By causing vehicle vibrations, roughness has a direct influence on ride comfort, safety, and vehicle wear [1]. In turn, the dynamic wheel loads pro- duced are implicated as causative factors in roadway deterioration. As a consequence, the characterization and measurement of road roughness is a major concern of highway engineering worldwide. As the highway networks in the United-States and other developed countries near completion, the maintenance of sophisticated quality at minimum cost gains priority. In sophisticated management systems, roughness measurements are an important factor in makiag decisions toward spend- ing limited budgets for maintenance and improvements. In the United States, ride comfort has been emphasized because it is the manifesta- tion of roughness most evident to the public. This philosophy has resulted in the concept of Present Serviceability Rating [2] used broadly throughout the United States to judge road roughness quality. In less developed countries, the same concerns face administrators from the very beginning; faced with limited resources, they must choose between quantity and quality in the development of public road systems. Optimizing road transport efficiency involves tradeoffs between the high initial costs of smooth roads and subsequent high maintenance and user operating costs of poor roads. Hence, studies of the road-user cost relationship to roughness are underway in India [3], Brazil [4,5], Kenya [6], and other locations. User costs are generally quantified 1 in terms of fuel, oil, tires, maintenance parts, maintenance labor, and vehicle depreciation. Other costs--often excluded from these analyses--that are also a consequence of roughness involve speed limitations, accidents, and cargo damage. A persistent problem in these studies is characterizing the rough- ness of a road in a universal, consistent, and relevant manner. The popular methods now in use .I;e based on either profilometry or measure- ment of vehicle response to roughness. These latter measurement methods make use of an instrumented vehicle that produces a numeric that is a measure of the vehicle response to road roughness as the vehicle traverses the road at a constant test speed. The systems have acquired the name Response-Type Road Roughness Measuring Systems (RTRRMSs) and represent development from a practical approach to the problem without a thorough technical understanding. As a result, the relationship between different measurement methods is uncertain, as is also the relevancy to ride com- fort or road-user costs. Nonetheless, most of currently popular instru- mentation systems share a commonality in configuration and operation, and are in such widespread usage that drastic changes in measurement methodology are not imminent. The users of RTRRMSs recognize that the roughness numeric ob- tained from one of these systems is the result of many factors, two of which are road roughness and test"speed. Other factors, that affect the responsiveness of the vehicle to road excitation at its traveling speed, can be difficult to control. While great effort is spent limit- ing the variability of these other factors, there is growing agreement that some variation can still persist between RTRRMSs and that even the most carefully maintained systems should be independently calibrated occasionally. Recent research on the variability of RTRRMSs, funded by the National Cooperative Highway Research Program (NCHRP) has indicated that the only calibration approach that will be valid for any roughness level or surface type is an empirical correlation of the RTRRMS with a reference measurement [7]. The calibration is performed by running the RTRRMS over -a number of "control" road sections that have been con- currently measured by the reference method. Validity of the calibration is guaranteed by selecting control sections that cover the roughness range of interest for each surface type, and performing separate 2 calibrations for each combination of surface type and measurement speed. The key to this approach is the ability to assign reference roughness levels to the control sections, This requires the ability to accurately transduce the longitudinal profiles of the control sections, in the wheel tracks traversed by the RTRRMS. It also 'requires a method for processing the profile data to yield a single roughness measure for the correlation. Although there is general agreement worldwide that RTRRMSs.pro- vide useful and meaningful data, and that they must be calibrated to a reference, there is no consensus as to how they should be operated, and what calibration reference should be used. In response to this need, the World Bank proposed that roughness measurement devices representative of those in use be assembled at a common site for an International Road Roughness Experiment (IRRE) to determine correlations among the instru- ments and encourage the development and adoption of a single Rough- ness Index (RI) to facilitate the exchange of roughness-related information. The IRRE was held in Brasilia, Brazil during May and June of 1982. Research teams participated from the Brazilian Transportation Planning Company (GEIPOT), the Brazilian Road Research Institute (IPR/DNER), the British Transport and Road Research Laboratory (TRRL), the French Bridge and Pavement Laboratory (LCPC), and The University of Michigan Transpor- tation Research Institute (UMTRI-formerly the Highway Safety Research Institute (HSRI)). Each of these agencies has been responsible for obtaining and analyzing road roughness data in different parts of the world, using different methods and analyses [4,5,6,7,8,9]. The IRRE included the participation of a variety of equipment: seven RTRRMSs, two methods for directly measuring profile, and two vehicle-based profilometers. Four road surface types were included: asphaltic concrete, surface treatment, gravel, and earth. At the finish of the experiment, all of the sections were evaluated by a panel of raters. 3 Objectives of the Experiment The project had three immediate objectives toward the ultimate goal of defining and validating a RI for all international use: Objective 1: To establish the correlation between different RTRRMSs. Different RTRRMS measures can be made somewhat "equivalent" through calibration. The IRRE should help determine the degree of equivalency between the measurements obtained by different systems, and the ranges of roughness, surface type, and operating speeds over which these equivalences are valid. Objective 2: To evaluate profile-based roughness measures for the calibration of RTRRMSs. Although there is agreement that profile measurement is needed to determine a RI for calibrating RTRRMSs, a number of analysis methods have been proposed to define RI. While they are generally unique, one aspect they all have in common is that they emphasize the roughness in only a por- tion of the wave number (wave number = l/wavelength) range of the profile. Depending on the form of the analysis, the waveband selected may simulate the response of a vehicle (Quarter-Car Simulation, as developed for the NCHRP [7]); it may correspond to a waveband related to pavement features ("waveband analysis," as employed by LCPC in the APL 72 analysis [10]); or it may produce a statistic that has been empirically correlated with other measures (RMSVA [11,12]- used in Brazil and Texas-and QI [4,5--the standard rough- ness scale in Brazil). Prior to the IRRE, none of the potential RI analyses had been demonstrated on unpaved roads. Objective 3: To evaluate candidate profile measurement methods. One of the problems in transferring methods worldwide is that certain equipment may be feasible in one country but 4 not another, for technical, political, or economic reasons. For example, the rod and level survey method is a labor- intensive method that is well suited to countries with low labor costs, whereas certain vehicle-based profilo- meters may require the technical support that is only.to be found in the more developed countries. In the past, specific analysis methods have been associated with parti- cular measurement methods. There is a need to determine whether different measurement methods can provide acceptable accuracy for calibrating RTRRMSs to an international roughness standard. A major requirement of any "profilometer" system is that it is able to be calibrated independently, with- out resorting to correlations with another instrument that measures road roughness. Four profile measurement methods were employed in the IRRE: th - rod and level survey method, the APL Trailer from LCPC, an experimental Beam device from TRRL, and a vehicle-mounted GMR Profilometer. At this time, the measurements have only been analyzed for the first three of these methods. Report,Organization The main purpose of this report is to document the experiment and data. This.involves describing the participating equipment, pre- senting the summary roughness numerics obtained from the RTRRMS, describing the subjective rating procedure, and presenting those candi- date RI statistics that have been calculated from the profiles. Many of the descriptions are technical and detailed, and most of the data, needed for further analyf3es, will not be of interest to the average reader. Therefore, this main report is limited to an overview of the IRRE and the results of the first analyses. The bulk of the technical information is sorted and presented in the attached Appendices A-G. The next chapter describes the experiment and the participating equipment. .Chapter 3 discusses the analyses performed to date, and presents the important findings from the data obtained in the IRRE. 5 Chapter 4 ties these findings into the immediate needs of the world highway community by discussing the relevancy of the different find- ings to the objectives of the project. When possible, specific recommendations are made concerning future use of RTRRMSs. Recommenda- tions are also made regarding further analysis of the IRRE data with the objective of the development of a universal RI, and validation of future roughness measuring systems. 6 CHAPTER 2 EXPERIMENT This chapter describes the physical aspects of the International Road Roughness Experiment (IRRE). It summarizes the methods used to acquire roughness data, the ranges of road and operating conditions covered in the IRRE, and the testing procedure. Participants The experiment included the participation of 11 pieces of equip- ment, which are separated into three categories in this report: Response-Type Road Roughness Measurement Systems (RTRRMSs), direct profile measurement, and indirect profile measurement. Appendix A pro- vides a technical discussion for each piece of equipment and offers much greater detail than the following overview. RTRRMSs -- All of the RTRRMSs that participated in the IRRE consist of a vehicle equipped with special instrumentation. Although different designs are employed, all of the instruments are theoretically measuring the same type of vehicle response: an accumulation of the relative movement of the suspension between axle and body. The measure- ments obtained with these instruments are in the form of discrete counts, where one count corresponds to a certain amount of cumulative deflection of the vehicle suspension. When the host vehicle is a passenger car, the instrument is mounted on the body, directly above the center of the rear axle. Alternatively, some are mounted on the frame of a single-wheeled trailer to one side of the wheel, directly above the axle. Seven RTRRMSs participated in the IRRE: 1. Mays Meter Systems. Three RTRRMSs were provided and operated by the Brazilian Transportation and Planning 7 Company (GEIPOT). These consisted of Chevrolet Opala passenger cars equipped with Mays Meters, manufactured by the Rainhart Company of Austin, Texas (13], as modified by the researchers of the international project, "Research on the Interrelationships Between Costs of Highway Construction, Maintenance, and Utilization" (ICR). The modifications were made to eliminate the strip-chart recorder normally used to read roughness measurements, replacing it with an electronic counter with a digital display (4]. The modified meters produce a display for every 80 meters of road travel, which is shown until the next 80 m is reached. The meter can also be adjusted to display every 320 m. 2. A Bump Integrator (BI) unit. This instrument was pro- duced and operated by the Transport and Road Research laboratory (TRRL) from the United Kingdom [9]. It was installed in a Chevrolet Caravan, which is a station wagon from the same automotive family as the Opala used for the Mays Meter systems. 3. A NAASRA Roughness Meter. The instrument was provided by the Australian Road Research Board (ARRB) [14], and operated by the research team from TRRL. It was installed in the same Caravan station wagon as the BI unit, and all measures made with the NAASRA and BI units were made simultaneously. 4. Bump Integrator Trailer. The BI Trailer, produced and operated by TRRL, is a single-wheeled trailer equipped with a BI unit (see Figure la) [91. It is.based on the old BPR Roughometer design [15], but has undergone a great deal of development by TRRL. 8 a. Bump Integrator Trailer - -r~ ik b. BPR Roughameter made by Soiltest, Inc. Figure 1. Two RTRRMSs Based on the BPR Roughometer Design. 9 5. BPR Roughometer. A Road Roughness Indicator, made by Soiltest, Inc., of Evanston, Illinois [16] is owned by the Federal University of Rio de Janeiro (COPPE/UFRJ) and was operated by personnel from the Brazilian Road Research Institute (IPR/DNER). The trailer is built to the specifications of the BPR Roughometer (see Figure lb) [15]. Normal measurement speed for the two trailers is 32 km/h (20 mph). A standard speed does not exist for the Mays Meter systems, although 80 km/h (50 mph) is the speed recommended by the manufacturer and widely used in the United States. Standard speeds in the ICR project were 80, 50, and 20 km/h, although in actuality little data were collected at 20. Standard test speeds-for the NAASRA Meter are 50 and 80 km/h. Direct Profile Measurement -- Two methods were used to obtain the elevations of the longitudinal profile of each wheel track over a test section. Each method uses a fixed horizontal reference as a datum line. Measures are then made of the distance between this datum and the ground at specific locations that are at fixed intervals apart. .One method is the traditional rod and level survey, shown in Figure 2. A surveyor's level provides the datum, while datum-to-ground measures are made with a marked rod. The level has a range of about 100 m. When it is moved to a new location (station), the change in elevation is established so that measures made from different.stations are equivalent. Using a measurement interval of 50 cm, a trained crew of three can survey both wheel tracks of two 320 m test sections in an eight-hour working day. The second method used in the experiment is based on an experi- mental instrument being developed by TRRL-the "TRRL Beam"--that is shown in Figure 3. The horizontal datum is provided by an aluminum beam three meters in length. The ground-to-datum measures are made with an instrumented assembly that contacts the Beam on precision rollers. To operate the device, the Beam is leveled by an adjustment at one end, and the sliding assembly is moved from one end of the Beam to the other. 10 TN Figure 2. Rod and Level Survey of Loneitudinal Profile. 11 皺 ，柚 Figtlre 3. Meas嘴elㅍent of Longi仁udinal Profile wi仁h TRRL Beam. l2 The moving assembly contains a microcomputer that digitizes the measures at pre-set intervals of 10 cm and prints them on paper tape. A trained crew of two or more can survey two wheel tracks of a 320 m test section in one day. Indirect Profile Measurement -- The two vehicle-based systems that participated are designed to measure longitudinal profile over a selected wave number range (wave number = l/wavelength) that is of interest. In both cases, an inertial datum is used that is not fixed, but provides a reference valid only for frequencies above a certain limit. The General Motors Research (GMR) Profilometer (also called a Surface Dynamics Profilometer), manufactured by K.J. Law, Inc., of Farmington, Michigan, uses an accelerometer to provide the reference datum [17]. The datum-to-ground measure is made by a follower wheel instrumented with a potentiometer. The signal from the accelerometer signal has no meaningful information at very low frequencies that cor- respond to long wavelengths, therefore the "profile" signal is inten- tionally filtered to remove the very low frequency content. The result is intended to be a signal with the same wave number amplitude content as the true profile for wave numbers corresponding to frequencies above the filter frequency at the test speed. The G'MR Profilometer had not been in use for several years before the IRRE and as a result, considerable effort was spent preparing it for the IRRE. The effort was justified by the need to obtain profile measures on the smoother sections with smaller measurement intervals and better resolution than was possible with the rod and level method. Another reason was that the IRRE offered a chance for side-by-side comparison between the profilometer and the rod and level method, to establish a roughness range over which the profilometer results are valid. Due to an almost endless series of problems-mostly related to the vehicle portion of the profilometer-it was able to obtain data on little more than half of the sections, although the roughness range covered was probably already beyond its capabilities. Measurements made with the profilometer were made much later than most of the others. The combina- tion of limited time and limited computer facilities made the complete 13 processing of all profilometer measurements impossible. Early analyses of the TRRL Beam data were promising, so the decision was made to stop all work involving the profilometer to concentrate on other tasks. No further mention of the GMR Profilometer is made in this report. The second profilometer, made by the French Bridge and Pavement Laboratory (LCPC), is called the Longitudinal Profile Analyzer (APL) trailer and shown in Figure 4. The trailer has a design that isolates its response solely to profile inputs. Movements of the towing vehicle, applied at the towing hitch-point, do not elicit any measurement. The datum consists of a horizontal pendulum that has an inertial mass, a spring, and a magnetic damper. The response of the pendulum is designed to provide a correct datum for frequencies above 0.5 Hz. The trailer wheel also acts as a follower wheel, and has been designed to allow measurement with fidelity for frequencies up to 20 Hz [18]. The wave number band measured by the APL trailer is determined by its measurement speed, as its true response is always over the frequency range of 0.5-20 Hz. The APL trailer is nearly always used in conjunction with one of two standard analyses, called the CAPL 25 and the APL 72 [10,18]. These analyses require that the trailer be towed at specific speeds (21.6 km/h for the APL 25 and 72 km/h for the APL 72), and that the test sections be of certain length (integer multiples of 25 r for the APL 25, and multiples of 200 m for the APL 72). Subjective Rating Study After the completion of the experiment (for the RTRRMSs), all test sections were evaluated by a panel rating process, documented in Appendix D. In this study, a panel of 18 persons was driven over the sections and asked to provide a rating ranging from 0 to 5. All panel members were driven in Chevrolet Opalas at 80 km/h over the paved sections, and 50 km/h over the unpaved sections. 14 4 a. APL Trailer L variable 50 cm environ b. Inertial Reference of APL Trailer. Figure 4. The APL Profilometer. 15 Design of Experiment Forty-nine (49) test sites were selected in the area around Brasilia. Thirteen of these were asphaltic concrete sections; 12 were sections with surface treatment; 12 were gravel roads; and the remaining 12 were earth roads. All of the candidate sections were rated with a Mays Meter-based RTRRMS, to ensure that the selected sections demon- strated a uniformly spread range of roughness. Generally, six levels of roughness were sought for each surface type, with two sections having each level of roughness as measured by the Mays Meter RTRRMS. All sections were fairly homogeneous over their lengths and were on tangent roads. Each section was 320 meters long. This length was selected based on the following considerations: 1. RTRRMSs are limited in precision, resulting in random error if the sections are too short. Standard test lengths in use throughout the world range from 0.16 km to over 3 km. A length of one mile (1.6 km) is common in the United States. 2. The Mays Meters used in Brazil can only be used on sections with lengths that are integer multiples of 80 m. 3. The process of measuring profile by the rod and level method is slow and tedious. Given the number of sections, the available time, and the available manpower for the survey crews, sections much longer than 320 m were not possible if all sections were to be profiled. 4. Some of the necessary combinations of roughness/surface type/homogeneity/geometry/traffic density/location were difficult to find. The difficulty was increased with test length. 16 The major disadvantage of the 320 m test length was its incom- patibility with the APL 72 requirement of a multiple of 200 m length. This incompatibility was not known by the Brazilian team at the time of site selection, and could not be -corrected for the equipment. The APL 72 measurements were obtained for the 200 m section completely contained within the 320 m test section, whereas the APL 25 output was the mean of the 12 APL values obtained over the test section. Measurements were made with the RTRRMSs at four speeds when possible: 20, 32, 50, and 80 km/h. The 32 km/h speed is standard for the BPR Roughometer and the Bump Integrator from TRRL. The 80 km/h speed (50 mph) is the most common measurement speed for RTRRMSs in the United States, and is recommended by several manufacturers of RTRRMS instruments. The other speeds of 20 and 50 were used as standard speeds in the ICR project. The APL trailer was operated at its standard speeds of 21.6 and 72 km/h. The levels of roughness went to sufficiently high levels that high- speed measurements were not expected to be within the allowable range for any of the equipment on the roughest unpaved sections. The operators of the instruments were given the option of declining to make any measurements that they felt would either be invalid or damaging to the equipment. Because of the relatively short section lengths, several measure- ments were made with the RTRERSs to improve precision and demonstrate repeatability. The RTRRMSs that were based on passenger cars made five measurements at each speed when possible, while the trailer-based systems made three runs in each wheel track. The sequence of tests was scheduled with several goals in mind. From a statistical point of view, it is helpful to randomize the sequence of each variable (roughness, surface type, speed, instrument). At the same time, any measurements that risk damage to the instruments should be scheduled last when all of the low-risk measurements have been completed. Transit time to and from the sections is minimized by scheduling all measures in one day for sections that are near each other. The sequence of testing is included as Appendix H. All of the 17 paved sections were tested before the unpaved sections, in an order dictated according to geographical convenience. The paved sections were not measured in any particular order in terms of their roughness. The smooth and moderate unpaved sections were measured according to geographical convenience, while the very roughest were measured last. Because of the logistics involved when a number of RTRRMSs are making measures on the same section, all repeats were made at one test speed before continuing to the next speed. The sequence of test speeds was randomized for each section when possible. However, some of the test sites were adjacent sections of road which were both tested in one pass of the RTRRMS; the same speed sequence was necessarily used for these tests. Testing Procedure The experiment took place over a period of one month, beginning on May 24 and ending on June 18, 1982. All of the vehicles under- went a speed calibration on the first day, based on a precision transducer on the APL trailer, which was in turn checked by stopwatch. During the following month, about 1-1/2 weeks were unscheduled, allow- ing make-up runs for the equipment that had experienced problems. The research teams from GEIPOT, TRRL, and LCPC operated their equipment, while the vehicles were driven by employees of GEIPOT. The tests were performed in caravan fashion, with all of the measures being made by the RTRRMSs at one speed before beginning the next speed. The testing was supervised by two test site controllers who kept track of the progress of each system. Occasional spot checks were made of the test speed with stopwatches to confirm that the test speeds were being maintained by the drivers. The APL trailer, which operated at different speeds, did not follow the caravan, but made its measurements as needed on the same sites as the others. The test sites were all located within a 50 km radius of the garage at GEIPOT used for storage and repair of equipment. The drive from the garage to the test sites served as a warm-up, to allow the 18 shock absorber and tire temperatures to stabilize. The test sites on unpaved roads required that the last 10 minutes of driving be over unpaved roads, so that the RTRRMSs were never operated "cold" on any surface type. An exception to this was the BPR Roughometer, which was towed only on the actual test.sites to minimize the damage that seemed to occur on a daily basis. The direct measures of profile were much slower than those of the RTRRMSs, and were made on different days. Measurements with the rod and level were made on all of the paved sections before the experi- ment, and repeated for many of the sections during the experiment. When testing proceeded to the unpaved sections, the rod and level measures were made immediately (two days or less) before the RTRRMS tests. The TRRL Beam did not arrive until the end of the experiment. Measures made with the Beam were made after the RTRRMS testing, on sites selected by the TRRL team to cover a range of surface types and .roughness conditions. Ten sites were completely profiled by the Beam. An additional eight wheel tracks were profiled on sections that dis- played nearly identical roughness levels on the right and left wheel tracks (as measured by the BI trailer). Repeat runs with the BI trailer on the sections that were profiled were used to confirm that the roads had not changed between the RTRRMS measures and the Beam measures. (The IRRE took place during the dry season, and as usual, there was no rain during the months of June, July, and August. The unpaved roads used for test sites normally saw little traffic. Marks were made to define the test wheel tracks with paint on the paved roads, lime on the earth roads, and with colored ribbon nailed to the surface of the gravel roads. Even at the end of July, the markers were still intact.) 19 CHAPTER 3 ANALYSIS AND FINDINGS The data obtained from the IRRE are possibly the most compre- hensive ever obtained in the field of road roughness measurement. Each RTRRMS produced five or six repeat roughness measurements for each of the 49 test sections for each of the three or four measurement speeds. Every section was profiled by the rod and level survey method at least once, yielding 1,282 elevation measurements for every one of the 70 section profiles obtained. LCPC provided profiles as measured with the APL trailer. In the APL 25 configuration, 97 of the 98 wheel tracks were profiled, yielding 1,281 numbers per wheel track. In the APL 72 configuration, each wheel track was described by a series of 6,401 digitized values. The experimental Beam from TRRL was used on 28 wheel tracks, providing 3,201 measures for each. In addition, all 49 sec- tions were rated subjectively by 18 panel members. All of the data had to be converted to a format suitable for computer analysis and checked for errors. The achievement of this task is one of the major accomplishments of the project. Three computer systems were employed in parallel to prepare the data for analysis. The rod and level survey measures were copied by typists into the IBM 370 computer system at GEIPOT. The RTRRMS data, the subjective ratings, and the elevation readings from the TRRL Beam were all typed into an Apple II microcomputer, using special entry and checking programs written for the project. The electronic analog signals produced by the APL trailer were digitized for plotting with a system based on a European ITT microcomputer that is compatible with the Apple II made in the United States. Programs were prepared to store the APL data on the floppy diskettes used by the Apple. 20 The analysis of all this data included five distinct tasks: 1. Correlation of reponse-type road roughness measurement systems (RTRRMSs) with each other. This study is straightforward: the RTRRMS data cover seven instru- ments, four surface types, up to four measurement speeds, and 12 or 13 test sites in each category. 2. Computation of profile-based roughness numerics. Profile signals that are entered on a digital computer consist of a large number of individual sampled elevation values that must be processed to yield a roughness numeric. Several processing methods are in use and have been pro- posed as candidates for defining an international standard roughness scale. Two of these that are widcAy known were applied to all of the profile signals obtained by the different methods. The first is the QI analysis, used in Brazil, that was developed as part of the ICR project and is based on two RMSVA statistics calculated for different baselengths. The second is the RARV developed for the NCHRP in the United States that is based on a Quarter-Car Simulation (QCS). APL roughness numerics were provided by LCPC based on the measures made by the APL trailer. 3. Correlation of profile-based numerics with RTRRMSs. The calibration of a RTRRMS is accomplished by regressing measures obtained on control road sections with correspond- ing reference measures, obtained by profiling the control sections and processing the profiles according to a standard method. In devising a good calibration method, the influences of variables that are known to affect both measures must be considered. 4. Comparison of profile measurement methods. No profile measurement method is without error, but some methods are superior to others when a certain analysis procedure is required. Five independent profile measurements were made 21 in the experiment; at the time this Draft Report was prepared, measures from four of the methods had been processed (the GMR Profilometer was excluded). Although direct comparisons between profiles (point for point, PSDs, etc.) are desirable, the time constraints for the Report made it necessary to use RTRRMs numerics obtained with each "profile" to indicate the reproducibility of the measures and the limitations of the different profiling methods for RTRRMs calibration. 5. Comparison of subjective ratings with other measures. In some applications, the relationship between rough- ness measurements and opinion of the roughness by the using public is -important. The panel ratings obtained for the test sections can shed light on these relation- ships. Prior to analyzing the RTRRMS data, the "raw" measures recorded in the field were averaged and rescaled to the same engineering units to facilitate direct comparisons between the different systems. The actual measure is inevitably a number of counts produced in a test. These counts were rescaled by the amount of suspension deflection in either direction needed to produce one count, and divided by the length of the test section to produce a measure of vehicle response that has the units: meters of suspension deflection accumulated in one traveled kilometer (m/km = slope x 1000). Users of this type of measure often refer to it by the _nits used, such as "counts/km," "inches/mile," or "mm/km." A more technical name is "Average Rectified Slope" (ARS) [19], which is used in this report, while recognizing that the variable whose slope is being rectified and averaged is difficult to visualize. The two RTRRMSs that are based on the BPR Roughometer design (see Figure 1) obtained measures of the left- and right-hand wheel tracks separately. The two ARS values were averaged for each section for com- parison with the single measures obtained from the RTRRMSs based on passenger cars. Appendix B contains all of the data from the RTRRSs, and also presents the summary results obtained by averaging repeat runs. When comparing measures from RTRRMSs over more than a single test speed, a 22 statistic called Average Rectified Velocity (ARV) has been proposed as a more useful roughness measure that is based on the raw reading of a RTRRMS (7,20]. Therefore, ARV values of all the RTRRMSs are also pre- sented in Appendix B. Finally, Appendix B contains the QI and RARV values obtained by processing profiles (as described later in this chapter). Correlation of RTRRMS Numerics A calibration procedure that can be used with different RTRRMSs can have only limited effectiveness if the different RTRRMSs are pro- ducing raw measures that are largely unrelated. No transformation will make the measures compatible if different systems rank the same set of roads in dissimilar order by roughness. Since the equivalence between measures based on separate RTRRMSs obtained with an independent cali- bration is always "second best" to a direct side-by-side comparison of the RTRRMSs, the correlation study was performed to determine the highest levels of agreement that are possible. This provides a standard that can be applied later for evaluating different candidate calibra- tion methods. In this correlation exercise, reported in Appendix C, the measures of each RTRRMS were regressed against those of every other for each of the 40 possible combinations of speed and surface type that exist when both instruments are operated at all four of the test speeds. A number of effects and interactions were examined. 1. Comparison of Measures Made at Different Speeds -- The most important finding of this study is that the best correlations between two RTRRMSs are obtained when the instruments are operated at the same test speed, even when the test speed is not "standard" for one of the instruments. For example, the BI trailer is normally operated at 32 km/h, while a Mays Meter system is typically operated at higher speeds. Figure 5 shows that the measures obtained from the BI trailer at 32 km/h are correlated with those of the Mays Meter at 50 and 80 km/h, and that 23 V) TE LLI 5 10 15 2 25 MAYS METER #2 - M/KM (.. = 50> BUF¢FCERP: Cih Tg- äF TE BF-EEDS:EO ' IGI Ii L 2 4 6 10 12 14 MAYS METER #2 - M/KM (V = SIE, F"EED : SO B2 Figure 5. Comparison of RTRRMS Measures Taken at Different Speeds. 24 different regressions are required for different surface types. In contrast, Figure 6 shows a single relationship between the two instru- ments when the measures are made at the same speed of either 32 or 50 km/h. Besides showing a more simple relationship, the amount of scatter is greatly reduced. 2. Effect of Including Different Surface Types -- Separate regression equations are usually not needed when the measures of two RTRRMSs are made at the same speed, while different regression equations are required if the speeds differ. 3. Effect of Including Different Speeds -- Even when comparing measurements made by different RTRRMSs at the same speed, several "standard" speeds could be required, depending on the intended uses of the measurements. A separate regression equation is usuall, needed for each speed when the measures are expressed in the form of an ARS (m/km, mm/km, in/mi, etc.). That is, the equation used to "convert" BI readings taken at 50 km/h to Mays Meter readings at 50 km/h is not valid for converting measures made by the BI at 32 km/h to Mays Meter measures made at 32 km/h. The errors are largest on smoother sections. 4. Effect of Converting Measures to ARV -- When the counts obtained from the roadmeter instruments are converted to "accumulated deflection per unit time," the result is the Average Rectified Velocity (ARV) of the suspension. Given the configuration of most roadmeters now in use, ARV is most easily calculated by multiplying the ARS numeric by the test speed, with an optional units conversion (see Appendices C and F for details). When data are taken at just one speed, the choice between ARS or ARV as a roughness measure is irrelevant: the two statistics differ only by a constant scale factor which is eventually eliminated through calibration to a reference. But when data taken at different speeds are compared, the two statistics have different inter- pretations. ARV is a direct measure of vehicle response: a higher ARV value always indicates more vehicle vibration, regardless of the 25 ,ler 5 l1 15 2@ 25 3Ø MAYS METER #2 - M/ýKM !URF~ACE : FCA TD R: TEF -J , rn 9-4 -74 .I I I-.*I 5 1 15 23 25 3 MAYS METER #2 - M/KM BUF¯3CS CA ~!E~ - 3R TE CSPFý'ENED a. S C Figure 6. Comparison of RTRRMS Measures Taken at Identical Speeds. 26 circumstances causing the excitation. On the other hand, ARS measures taken at different speeds are not comparable because the variable whose slope is being measured has a definition that depends on speed. This speed effect confounds with nonlinearities in the vehicle and roadmeter, requiring the separate calibration curves for each speed noted above (or alternatively, the resignation to accept larger amounts of scatter and less accuracy in the calibration). Figure 7 shows that when all of the measures are expressed as ARV, a single relation can be used for all speed/surface type combinations. 5. Limitations of Different RTRRMSs -- None of the speed/ surface type conditions seemed uniformly good or bad when comparisons were made between RTRRMS measures made at the same speed. Most of the instruments were capable of testing almost the full roughness range available. Still, the individual RTRRMSs did show different limitations. Correlations involving the Soiltest BPR Roughometer were usually lowest, even in the best of cases when it was compared to the BI trailer. This BPR Roughometer was the most fragile of the RTRRMSs, and experienced constant breakdowns. It was not operated at high speeds on the rougher surfaces. The Mays Meter systems were operated at the 80 km/h test speed on all surface types, although the operators did not run them on the four roughest test sections at this speed. As a result, the highest levels of vehicle excitation occurred at 50 km/h, with ARV levels being as high as 390 mm/sec. This roughness limit could also be measured with most of the other RTRRMSs, even though none of the others were operated at speeds over 50 km/h on iost of the sections. The BI trailer was never operated at the 80 km/h test speed, but was able to run on the roughest section at 50 km/h. Given the ARV results from the Mays Meters, the 50 km/h speed limit cannot be attributable to the level of vehicle excitation. 27 т � � � с {�fi р� ° �_ (.,,� ° � ! � LrJ ° . lJ.t {�1 J ■ s.� й ° а N � а .._ - л -а м - i7d r�'- а � _ 6 50 1@6 15С� �6�I �5од 366 З50 МА`rS MiETER #2 - ММ!S ��Г--�F�йСЕS � Сг� Т� GГ� ТЕ ��EEI�E _ ��.� -�� �t� . т � 1 � � • rл м � � � i � с� . � � . � � ¢ ,9 ¢ Ф . � а� т . 0 5б 166 156 26б 2S6 366 350 МА`�S METER �2 - MM1S �LJ�Г-г�"АСЕ� � СА ТЕ �F: ТЕ S�'EEDS а �t� пу ,_.�,�с� � Ф ;t ° tt} � ° Е М � 1 � � ш с�и �_ Q ° �" СС © .�� F-- у н � _ CiJ и ° Ф , 0 1 бд 26kЭ 3�36 �#бмЭ Nй�SRp - MM�S �U�:FACE� � �G� Т� С� ТЕ �F''^`_Е�ыЕ � �� � = � ,.�_лt� Figure 7. Comparison of RTRRMS Measures ot ARV Taken at Identical Speeds, Over а Raпge of Speed апд Surface Ccnditioпs. ?g The vehicle equipped with the NAASRA meter and a BI unit was not operated at 80 km/h except on 11 of the asphaltic concrete sections. Nearly all of the measurements obtained from the two roadmeters were nearly identical (when scaled to "m/km"), and were compatible with those of the other RTRRMSs. The exception to this was the case of the data taken at 80 km/h. The BI and NAASRA data did not agree as well as for the other speeds. Correlations with the Mays Meters and a simulated RTRRMS (using profile measurements as input) were higher for the NAASRA meter than for the BI meter. 6. Effect of Different Roadmeter Instruments -- Figure 8 shows,-, that when the results of th(P., BI and NAASRA roadmeters were rescaled to the same physical units of A7,S (m/km = slope x 1000), the two gave readings that were virtually interchangeable. There were no PCA meters in the IRRE, but a similar correlation experiment performed in the United States showed that PCA meters can also be used to measure ARS and ARV by eliminating the complicated PCA data reduction process [7]. Because different manufacturers of roadmeters recommend different measurement practices, there is often a tendency to assume that the same brand of roadmeter instrument must be used in all vehicles for good agreement. Yet the theoretical understanding and the practical evidence obtained in recent years show that the choice of roadmeter instrument is not of primary importance. Instead, the critical factor is the methodology adopted to obtain and analyze the roughness data. 7. Effect of Individual Wheel Track_Roughness -- Theoretically, the measures obtained from two-track vehicles such as an automobile are best correlated to the measures obtained from single-track trailers if the trailers are towed over each wheel track and the ARS numerics averaged. The empirical correlations between these averages and the measures from the two-track RTRRMSs were excellent, being as good as the correlations between different two-track systems. In addition to the average, a difference can be calculated from the two trailer measures. The difference measures were found to be uncorrelated to the measures of the two-track vehicles. 29 1 æu One Vehicle. 30 Computation of Profile-Based Numerics Four candidate roughness statistics were tested as a calibration reference for RTRRMSs: 1. QI -- The roughness staudard used in Brazil, developed during the ICR project [5], is based on the RMSVA summary statistic calculated from a measured profile. PRISVA is the Root-Mean-Square (RMS) value of a signal that is the simulated response of a rolling straight- edge, sometimes called a Mid-Chord Deviation (MCD), as shown in Figure 9. When the baselength is very short, the signal approximates the second derivative-the Vertical Acceleration--of the profile [11]. The baselengths used for road roughness characterization result in a signal that only approximates the true VA for very long wavelengths (100 m and longer), so in a sense the name RMSVA is deceptive. QI is defined by the equation: QI = -8.54 + 6.17 x RMSVA (1.0) + 19.38 x RMSVA (2.5) where RMSVA (1.0) and RMSVA (2.5) use baselengths of 1.0 and 2.5 meters and have units: slope x 1,000,000. The baselengths are equivalent to rolling straightedges with chord lengths of 2.0 and 5.0 meters, respectively. Appendix E presents a complete technical description of QI, its development, its wave number sensitivity, and the QI data obtained from the different profile measurements. 2. RARV -- The concept of using a reference RTRRMS has short- comings when applied to a mechanical vehicle-based system that can be overcome by defining the reference as a mathematical description of such a system, and using that "simulation" to calculate the response of the reference from profile measurements. The model, shown in Figure 10, defines that simulati--n and replicates the two essential resonances shared by all vehicle-based RTRRMSs. Because it has just one wheel, it is called a Quarter-Car Simulation (QCS), although the effects of both wheels on an axle are included by averaging the profiles of the two wheel tracks prior to input to the simulation. The model parameters 31 B B 0 MCD 0 Y (x+B) Y (x) Y(x-B y Ø-x MCD = Y(x+B) + y(x-B) Y (x) 2 - 2 x MCD/B2 L N "RMSVA" = (VA)2dx = VA2 Ni= Figure 9. Geometric Interpretation of RMSVA. 32 Sprung Mass MS s + cs5.(2-5u + KsU s~ u s5M m s + m u + Ktzu = Ktz 62.3 1/sec2n Damper Kt/MS 653 1/sec2 Ii/MS =.150 Ce /Ms 6.00 1/sec Unsprung Nu Mass RARV = |fs- d Tire Spring u t 1 = Measurejent time (sec) = 3600 • L/V Quarter-Car Model L = Road length (miles) V = Speed (mph) Figure 10. Quarter-Car Simulation Model. were selected for maximum agreement with RTRRMSs that have stiff shock absorbers, because the use of stiff shock absorbers reduces many of the sensitivities of RTRRMSs to factors other than roughness and test speed [7]. The simulated response is summarized by the ARV, called Reference ARV, or RARV. A further discussion of the RARV statistic is presented in Appendix F, along with computational details and the RARV values obtained for the test sections at the four simulated test speeds. 3. CAPL 25 -- LCPC determines the quality of newly constructed pavements by towing an APL trailer over the section at 21.6 km/h, and calculating the average absolute value of the signal produced by the trailer. The average is taken over sections of road that are 25 m long, hence the name APL 25 Coefficient (CAPL 25). In a sense,'this analysis is not truly profile-based, because it depends on the unique response properties of the APL trailer. However, the properties are claimed to be so consistent from trailer to trailer, due to the elimination of most sources of variation in less sophisticated RTRRMSs, that the APL trailer could conceivably be characterized mathematically and its results predicted from profile measurements by other methods such as rod and level. Correlations between CAPL 25 measures and those of the RTRRMSs were not good, however, due to time constraints and the objec- tives of the Report, no attempt was made to obtain estimates of the CAPL statistic from the other profile signals. Appendix G describes the sensi- tivity of the CAPL 25 to wave number, presents the data provided by LCPC, and discusses the reasons for the poor correlation with the RTRRMSs. 4. APL 72 Wave Band Indices -- LCPC has developed this analysis method to summarize the present condition of roads. Two APL trailers are simultaneously towed at a speed of 72 km/h, with one trailer follow- ing each wheel track. The signals are played into six electronic band-pass filters (three per signal) to separate the original signals into three band-limited signals. The filtered signals are squared and integrated, to obtain mean-square values calculated over road sections that are 200 m in length. The mean-square values for the right- and left-hand wheel tracks are summed, and a table is then used to assign 34 a rating from 1 to 10 for that section of road. The filters separate the profile into three frequency bands, and a separate table is used for each band. The result is that the road is described by three quality indices, for short, medium, and long wavelengths. In theory, the response properties of the APL trailer play no role in determining the three indices, because the frequency response of the trailer is broader than the bands left after filtering. Thus, the same analysis could be applied to signals obtained from other profiling methods. However, because the correlations between the APL 72 indices and the RTRRMSs includdi in the IRRE, were not good, further studies of the APL 72 analyses were not justified within the objectives of the Report. Further details concerning the APL 72 analyses are presented in Appendix G, along with the indices obtained for the test sections in the IRRE. Correlation of Profile-Based Numerics with RTRRMS Numerics Correlations between the candidate roughness standards and the RTRRMSs were calculated to determine the accuracy and minimum complexity needed for calibrating the RTRRMSs to the candidate standards. Details of the analyses are presented in Appendices E, F, and G. The findings are summarized for each of the proposed methods: 1. QI -- The QI roughness scale provides a single roughness rating for any given section of road, and as a consequence, there is a "best" speed that should be used by RTRRMs whose measurements are calibrated to this scale. The best of the four test speeds used in the IRRE is 50 km/h. This finding was not unexpected because QI was originally based on a QCS with a simulation speed of 55 km/h (see Appendix E for details). In general, separate calibrations are sometimes needed for different surface types. (Example calibration curves are shown in Appendix G.) At 50 km/h, the correlations between QI and the ARS measures from the RTRRMSs are very good for three of the surface types, having approximately the same quality as the direct correlations between RTRRMSs. On the sections with surface treatment, however, correlations with QI are significantly lower; for every RTRRMS/speed combination, a separate calibration would be needed for surface treatment roads. 35 Correlations are always high on the asphaltic concrete sections, but other correlations are not as good at speeds other than 50 km/h. The QI scale can be an effective and accurate reference for calibrating RTRRMSs, but the calibration effort needed is substantial because separate transformations are needed for the different surface types. The fact that RTRRMSs should only be operated at a single speed of 50 km/h might be a deterrent for those users of RTRRMSs who have experience with different test speeds. (While RTRRMS data taken at speeds other than 50 km/h can, of course, be correlated with QI, the estimates of QI made from the RTRRMS measures will have greater error.) 2. RARV -- The RARV roughness statistic (the measure obtained from a simulated reference RTRRMS) is not a unique value for a given road surface, but will vary with speed just as a vehicle sees different roughness at different speeds. Not all RARV values could be calculated for the speeds of 20 and 32 km/h due to profiling limitations described below, but the results that were obtained indicate good agreement with the RTRRMS data at these speeds. Correlations between RARV and the ARV measures obtained from the RTRRMSs are usually very good at all speeds and on all surface types, with the one exception of the 80 km/h datz from the surface treatment sections. Overall, correlations between RARV and the RTRRMSs were higher than with the other candidate standards. Figures 11 and 12 show sample calibration equations that are each calculated for just one speed/surface type combination and plotted only over the roughness range covered by the measures (inference space). With most of the RTRRMSs, better accuracy can be obtained by using separate equations for different speed/surface type conditions, although in the best of cases-the BI trailer-a reasonable calibration can be obtained with just two equations: one for paved surfaces at any speed, and one for unpaved surfaces. If the RARV is used as a reference, the measurement speed is still a variable; a given section of road can be assigned a range of roughness values, each corresponding to a different (simulated) measurement speed. For the purpose of standardizing roughness measurement, a logical step 36 æ -i G.,c A =