Measuring, Characterizing, and Reporting Pavement Roughness of Low-Speed and Urban Roads (2019)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

67 This chapter summarizes the research that was performed and the primary findings, provides recommendations for implementation of the findings, and offers suggestions for future research. 5.1 Summary of the Research and Primary Findings 5.1.1 Measurement Accuracy An experiment was conducted to demonstrate the effects of potentially adverse operational conditions on the measure- ment of longitudinal road profile by high-speed inertial pro- filers. The experiment included production inertial profilers manufactured by six different vendors. The testing program replicated common operational conditions encountered by profilers on low-speed and urban roadways, such as operation at low speeds, acceleration and deceleration, stop-and-go oper- ation, profiling from a dead stop, and operation on a curve. The results of this experiment confirmed that braking, accelerating, operating on a curve, and coming to a stop while operating an inertial profiler cause profile measurement errors. This is because the accelerometers used in inertial pro- filers are rigidly mounted to the host vehicle chassis and do not remain consistently aligned with the true vertical direction. Two operational conditions introduced errors into profiles that appeared as localized roughness and affected the IRI: 1. As the brakes were released at the end of braking events, artificial curvature appeared in the measured profiles. Braking caused the vehicle to pitch forward, and the result- ing misalignment caused the accelerometers to errone- ously detect a portion of the horizontal acceleration. This caused measurement error throughout each braking event. The errors affected roughness most at the end of braking events because of the combination of host vehicle pitch misalignment and the rapid change in deceleration. 2. Collection of profile through a stop introduced an artificial change in elevation at the location of the stop. At a stop, inertial profiler accelerometers typically experience only a small bias caused by misalignment with the direction of gravity and other error sources. However, the gradual change in elevation with time that resulted after integrat- ing the accelerometer signal twice was concentrated at one location, because the profiler was not moving. Application of a high-pass filter to the measured profile spread out the influence of these errors. The severity of local- ized roughness caused by braking increased with the severity of deceleration and the severity of localized roughness caused by stop-and-go operation increased as the length of time at the stop increased. For both braking and stop-and-go operation, the error in profile measurement and the severity of artificial roughness in the profile was greater for profilers mounted to vehicles with a higher propensity for pitch and roll. Operational conditions that included transitions in hori- zontal acceleration affected the long-wavelength content in measured profiles. This included the onset of braking, releasing the brakes, transition to coasting, heavy changes in throttle position, and operation through changes in horizon- tal road curvature. Operation at low speed also degraded the measurement of long-wavelength content. Lateral tracking errors affected the reproducibility of measurements during the experiment, which confounded the interpretation of the results. 5.1.1.1 Constant Speed Operation Effect on IRI: • Position records from a GPS data logging system showed that lateral tracking of the profilers strongly affected their repeatability and accuracy. Over the range of speeds from 25 mi/hr (40 km/hr) to 60 mi/hr (97 km/hr), inconsistency C H A P T E R 5 Summary, Findings, and Recommendations for Future Research

68 in lateral tracking affected agreement in profile and rough- ness more than travel speed for five of the six profilers. • Travel speed affected longitudinal distance measurement and measured IRI in turn. In the absence of intermittent calibration of the distance measurement instrument to account for changes in travel speed, systematic differences in measurement of longitudinal distance of up to 1 percent occurred over the range of speeds from 10 mi/hr (16 km/hr) to 60 mi/hr (97 km/hr). • One of the profilers measured much lower IRI at 10 mi/hr (16 km/hr) than at other speeds, and another showed greater variation at speeds of 20 mi/hr (32 km/hr) and below than at higher speeds. However, three of the profilers were able to measure the IRI with the same level of accuracy and repeatability over the entire range of speeds from 10 mi/hr (16 km/hr) to 60 mi/hr (97 km/hr). • One of the profilers showed a systematic increase in IRI with travel speed over the entire range from 10 mi/hr (16 km/hr) to 60 mi/hr (97 km/hr). This appears to be due to a signal processing issue that has since been addressed. Effect on profile: • Speed affected the repeatability and reproducibility of long- wavelength profile content (8–67 m; 26.2–220 ft). Repeat- ability was higher for repeated passes above 20 mi/hr (32 km/hr) than for measurements below 20 mi/hr (32 km/hr). Reproducibility was highest for comparisons of runs among the higher speeds and lowest for compari- sons of runs among lower speeds. 5.1.1.2 Coasting Transition from operation with cruise control to coasting did not cause any discernable errors in profile measurement or IRI. However, the net change in travel speed affected measurement of longitudinal distance. For a coast from 45 mi/hr (72 km/hr) on an upward grade of 0.5 percent, up to 6.6 ft (2 m) of longitu- dinal misalignment accumulated over 771 ft (235 m) of travel. 5.1.1.3 Braking Application of the brakes introduced large artificial dis- turbances into measured profiles. Each profiler that applied a high-pass filter spread out the effect of braking beyond the range where braking occurred. The shape, severity, and range of the errors in profile depended on the type of high-pass filtering applied by each profiler. The following observations were made: • The errors in profile caused by braking affect the IRI most at the end of a braking event, near the location where the brakes were released. • Braking with peak deceleration of 0.26 g or more nearly always caused profile measurement error that registered as an area of localized roughness. • Braking with peak deceleration from 0.16–0.26 g affected the measured IRI somewhat, but caused profile measure- ment error that registered as localized roughness in fewer than half of the test runs. In this range of severity, braking was more likely to cause localized roughness in profilers carried by vehicles with a high center of gravity. • Braking with a peak deceleration of 0.16 g or less rarely caused an increase in the measured roughness that raised the IRI of a 0.1-mi (160.9-m) section by more than 3 in/mi (0.05 m/km), or caused an area of artificial localized roughness. • Braking caused a rapid accumulation of bias toward under- estimated longitudinal distance. For the braking events included in the experiment, a reduced estimate of travel distance of 2–4 ft (0.6–1.2 m) was common. 5.1.1.4 Application of the Throttle Application of the throttle for normal and aggressive acceleration—from 20 mi/hr (32 km/hr) to 45 mi/hr (72 km/hr)—affected the long-wavelength content that is typically visible in raw profile plots. However, no discernable errors in the measurement of IRI were observed. 5.1.1.5 Stop-and-Go Operation Stopping during a profile measurement caused a large artificial disturbance in the profile at the location of the stop. Profiles submitted without high-pass filtering typically included a large step change in elevation at the stop. Profiles submitted with high-pass filtering included erroneous con- tent over a wide range, often with a severe slope change at the location of the stop. Some of the profilers applied special provisions for miti- gating errors near the stop. In most cases, the severity of the error in measured IRI was reduced, but not sufficiently enough to avoid the appearance of severe localized rough- ness. One profiler, which used a proprietary “stop-and-go operation” feature, was an exception. The profiler with stop- and-go operation registered an artificial area of localized roughness at the stop in only some of the stop-and-go runs, with a maximum severity of 190 in/mi (3 m/km). In most of the stop-and-go runs by the other profilers, severe localized roughness with a peak value of at least 380 in/mi (6 m/km)— and in some cases many times greater than that—appeared at the location of the stop. Examination of the profiles showed that an area of up to 155 ft (47.3 m) upstream of the stop and up to 248 ft (75.6 m) downstream of the stop should be marked as invalid

69 for computation of IRI, depending on the type of high-pass filter applied to the profile. 5.1.1.6 Dead Stop The testing program included runs where profile was col- lected from a dead stop. However, all six profilers included provisions for preventing the collection of profile until a suf- ficient speed was reached or a sufficient distance was trav- eled. Limited observations showed that initiating profile data collection shortly after a stop caused less error in IRI than including the erroneous content at the stop. 5.1.1.7 Operation on a Curve Operation on a curve at various speeds by the inertial pro- filers caused inconsistency in the raw profile traces. However, the overall IRI values did not include an obvious bias. Test- ing on a curve showed that measurement of longitudinal dis- tance using only one wheel caused a large bias in longitudinal distance measurement, depending on whether the wheel was on the inner or outer side of the vehicle. 5.1.2 Characterization An experiment was conducted to correlate measured road roughness to objective measurements of vibration experi- enced by a vehicle driver. Three instrumented vehicles were tested on 29 urban and low-speed test sections. The vehicles included a mid-sized sedan, an SUV, and a full-sized van. The instrumentation provided simultaneous measurements of road profile with a wide-footprint laser, vehicle speed and position, vehicle chassis acceleration, and acceleration at sev- eral vehicle/driver interfaces. The testing included multiple passes over each test section at each of two speeds. The analysis produced standard metrics for quantifying driver discomfort from measured accelerations at driver/ vehicle interfaces, including RMS weighted acceleration at the seat/buttock interface, RMS weighted acceleration at the floor/foot interface, a vibration level aggregated from all three directions at the seat/buttock interface, and an aggregated total vibration experienced at the floor/foot interface, seat/ buttock interface, and seat/back interface. Driver discomfort was correlated to the IRI, RN, and several adaptations of the IRI algorithm using the Golden Car model that included variations on the simulated travel speed, index accumulation type, and predicted response type (i.e., sprung mass acceleration in place of suspension stroke). The experimental results produced observations with practical implications regarding the need to record localized roughness and relationship between vehicle response and a road roughness index at lower travel speeds. 5.1.2.1 Localized Roughness The vibration experienced by the driver in all three vehi- cles on the majority of the test sections included content that classified as “transient.” The presence of transient vibration implies that the overall vibration level is not a sufficient rep- resentation of user perception of comfort, and that some esti- mate of the severity of transient vibration is also required. The magnitude of peak values within a short-interval rough- ness profile at or near each instance of transient vibration correlated favorably with a transient vibration magnitude. In particular, peak values from roughness profiles averaged over 20 ft (6.1 m) through 35 ft (10.7 m) correlated best. 5.1.2.2 Roughness Index The IRI exhibited an acceptable level of correlation with standard measures of user discomfort for all three vehicles. For example, a linear fit between RMS acceleration at the floor/foot interface and IRI from the left wheel path produced R2 values of 0.80, 0.78, and 0.80 for the mid-sized sedan, SUV, and full-sized van, respectively. RN calculated from both wheel paths exhibited comparable, but higher, correlation overall, with R2 values of 0.79, 0.85, and 0.84. Testing was performed at speeds ranging from 25 mi/hr (40 km/hr) to 55 mi/hr (88 km/hr), with more than half of the test runs at 30 mi/hr (48 km/hr) through 35 mi/hr (56 km/hr). Testing at lower speeds shifted the sensitivity of the test vehicles toward shorter-wavelength content within the profiles. Reducing the speed used in the IRI simulation algorithm below the standard value of 49.7 mi/hr (80 km/hr) improved correlation. For example, using a fixed speed of 35 mi/hr (56 mi/hr) produced R2 values of 0.87, 0.86, and 0.83. This shows that a shift toward shorter wavelengths improved the relevance of the Golden Car model used by the IRI for travel at lower speeds. Use of the specific travel speed from each run in place of a fixed speed in the Golden Car model produced similar agreement (R2 values of 0.87, 0.84, and 0.83). Adjusting the IRI algorithm to output temporal intensity of roughness (i.e., inches of response per second of travel) instead of spatial density of roughness (i.e., inches of response per distance traveled) improved correlation further. A tem- poral Golden Car index, which simulated velocity across the suspension using the travel speed from each run, produced R2 values of 0.90, 0.87, and 0.88 for the mid-sized sedan, SUV, and full-sized van, respectively. This adaptation of the IRI algorithm is dubbed Golden Car Average Rectified Veloc- ity (GCARVV). It is equivalent to Reference Average Rectified Velocity (RARV), which served as a basis for the IRI before it was standardized (Gillespie et al. 1980). Further adjustments to the Golden Car model to specifi- cally predict vertical acceleration at the sprung mass, as well

70 as more detailed models intended to reproduce the responses of the specific test vehicles, improved correlation further. However, these were deemed unsuitable for use in pavement management due to a lack of generality. That is, they may better predict vertical acceleration on the three test vehicles, but at the cost of reduced relevance to other vehicles and other responses of interest. 5.1.3 Profile Features The research examined characteristics of urban and low- speed roadways that register as roughness in measured longi- tudinal profiles. The study primarily used three data sources: (1) right-of-way images from the Pennsylvania DOT pave- ment network survey of Philadelphia County in 2012; (2) profile data and right-of-way images from selected road segments on 26 routes in Philadelphia County collected by the Pennsylvania DOT in 2013 and 2014; and (3) profile data, straight-line diagrams, and right-of-way images provided by the New Jersey DOT on ten newly resurfaced urban road segments. Several examples of roughness were observed at specific locations associated with specific hardware or design ele- ments. This includes (1) aspects of the pavement design itself, such as drainage provisions and intersection crossings; (2) aspects of the roadway design required for accommodat- ing right-of-way access, such as railroad crossings; (3) utility access; and (4) excavation for repair of underground utili- ties. Some of these features cause roughness because they impose constraints on the design profile that are in conflict with roadway smoothness. Others cause roughness when they are installed without allowances for surface roughness, not built to within tolerance, or are poorly maintained. The survey of road profile features from urban and low- speed roadways showed that localized roughness accounted for a significant portion of the overall roughness. Localized roughness appeared at locations with surface distress and at built-in features. Many of the built-in features caused local- ized roughness of severity much greater than the segment- wide average. For example, the roughness observed at crowned intersection crossings was often more than three times that of the surrounding roadway. Localized roughness at rail- road crossings contributed up to 60 in/mi (0.95 m/km) to the average roughness of 528-ft (160.9-m) sections that included them. Several utility covers and drainage inlets were detected in longitudinal profiles that contributed very little to their roughness. However, other drainage inlets contrib- uted 15 in/mi (0.24 m/km) or more (each) to the average roughness of 528-ft (160.9-m) sections that contained them, and many utility covers (and the surrounding patching or distress) contributed more than 40 in/mi (0.63 m/km) (each) to the average roughness of 528-ft (160.9-m) sections that contained them. Comparison of road profiles with right-of-way images revealed some of the challenges posed by urban and low- speed roadways to the measurement of profile. Transverse roughness variations caused by distress, such as potholes and longitudinal cracking, or caused by built-in features, such as utility covers and drainage inlets, affected the roughness that was registered in a given pass over some pavement sections. In some cases, modest variations in the lateral tracking of a pro- filer determined whether a feature was detected at all. Height sensor footprint and sampling procedures also affected the roughness measured at some distresses and built-in features. For example, narrow channels or holes caused deep, narrow dips in profiles. In some cases, the dips register dispropor- tionately as roughness relative to the response of a vehicle tire because the tire contact patch envelops them. Common filtering practices that are appropriate for com- putation of IRI may confound the process of identifying the causes of roughness. Low-pass filtering removes details from the profile that help recognize specific built-in features, such as drainage inlets, utility covers, textured pedestrian cross- ings, bridge joints, and railroad crossings. High-pass filter- ing applied to remove very long wavelength features from the profile distorts the appearance of some design features, such as slope breaks that appear at low points for drainage. The research demonstrated an example of spatial location of built-in roadway features within profiles augmented with GPS position measurements. However, the effort to assemble a database of geo-located roadway features from multiple sources showed that, due to security concerns, information about some underground water and electrical utilities could not be obtained for public use from existing databases. In such cases, right-of-way images synchronized with the pro- file measurements provided a reliable way to identify built-in features and to determine whether built-in features or distress were the cause when localized roughness was detected. 5.2 Recommendations 5.2.1 Measurement 5.2.1.1 Operational Conditions This research included experiments to investigate the link between road roughness and ride quality, and for identifica- tion of specialized profile features on urban and low-speed roadways. In both cases, measurement of profile on urban and low-speed roadways included challenging operational conditions that are not typically encountered on high-speed, limited-access freeways, including restrictions on speed or stops caused by traffic and traffic signals, measurement of very rough pavement sections, and measurement of hardware in the road with deep narrow dips or other features that challenge profiler height sensors.

71 Adapting inertial profilers for these conditions will require equipment hardware and software changes, and verifying the operational limitations of a given design will require testing effort beyond current practice. To help justify the required investment, it is recommended that (1) AASHTO M328 (Standard Specification for Inertial Profiler) specifically list valid operation at very low speed, with braking, with stops, and on very rough pavement as desired qualities of a profiler intended for use on urban and low-speed roadways, and (2) AASHTO R56 (Standard Practice for Certification of Inertial Profiling Systems) include verification testing for operating conditions that occur on urban and low-speed roadways. To the extent possible, these recommendations place requirements on profiler system performance, rather than on component specifications (e.g., accelerometer and height sensor range, resolution, etc.). The primary motivation for a performance-based approach is to encourage the devel- opment of profiler designs that achieve the recommended performance at the lowest cost, with no restrictions that dis- courage innovation. Further, the optimal specifications for profiler sensors, layout, calculation methods, and host vehicle properties are interdependent, and component-based require- ments may not be universally appropriate. 5.2.1.2 Braking, Low-Speed Operation, and Stops Results from the experimental evaluation of production inertial profilers justify the establishment of procedures for identifying conditions that cause invalid profile measure- ment, such as travel below a minimum speed or braking above a specific deceleration level. AASHTO R57 (Standard Practice for Operation Inertial Profiling Systems) recommends that during network-level profile measurement operators mark data collected below the minimum operating speed of the profiler. It is recommended that AASHTO M328 requires inertial profilers to automatically detect operation below a minimum valid operating speed, deceleration above a given threshold, and instances where the profiler comes to a stop; mark the affected location or range within the stored profile as invalid; and issue a live alert to the operator. This enables objective, consistent identification of adverse conditions, and relieves the profiler operating crew from a potential dis- traction. Temporary activation of an audible alert is recom- mended for training inexperienced drivers and operators. Sensitivity to operational conditions depends on a com- bination of sensor mounting position and hardware, sensor specifications and performance, profiler host vehicle prop- erties, and profiler calculation and filtering procedures. As such, this research did not identify specific thresholds for minimum speed and maximum deceleration that are uni- versally appropriate for all inertial profilers. Instead, four types of additional “Dynamic Certification Testing” in AASHTO R56 are recommended for inertial profilers that will operate on urban and low-speed roadways to identify the following: 1. Minimum valid operating speed: This procedure requires passes at a minimum speed for valid operation proposed by the operator for certification. The profiler qualifies as valid for operation at the proposed speed if the measured profiles meet the required thresholds for repeatability and accuracy. 2. Maximum valid operating deceleration: This procedure requires passes with braking at a deceleration level pro- posed by the operator for certification. The profiler quali- fies as valid for deceleration up to the level used in the testing if the measured profiles with braking reproduce profiles measured at constant speed with the same level of repeatability established for regular operation. 3. Invalid range near deceleration: This procedure requires passes with braking at an average deceleration of 0.26 g. The range to mark as invalid is based on the bias in the short-interval roughness profile compared to a pass at constant speed. 4. Invalid range near stops: This procedure requires passes with a stop. The range to mark as invalid is based on the bias in the short-interval roughness profile compared to a pass at constant speed. The recommended additions to dynamic testing in AASHTO R56 treat stop-and-go runs as a distinct type of adverse condition, rather than a combination of deceleration and operation below the valid profiling speed. This is because experimental measurements that included a stop had profile measurement errors that resulted in a larger error in the IRI extending over a broader area than runs with braking or low-speed operation without a stop. The proposed dynamic testing in AASHTO R56 requires profiler operators to select a minimum test speed and maxi- mum deceleration level for testing. Some iteration may be required to identify speed and deceleration levels that are likely to pass the certification. To help avoid unnecessary effort at the certification site, it is recommended that profiler operators attempt the recommended testing in advance of the official certification. It is anticipated that profiler manufacturers will know the limitations of their equipment, particularly after experiencing the recommended testing. Until the specific limitations of a given inertial profiler are established, default values for minimum test speed, maxi- mum deceleration, and the range of invalid profile near stops and severe braking may have to be set using conservative estimates. For the inertial profilers tested in this research, the most conservative observations suggest a minimum valid test

72 speed of 25 mi/hr (40 km/hr) and maximum deceleration of 0.16 g. Observations from this research also suggest the following default settings for removing areas from the calcu- lation of the IRI: (1) within 155 ft (47.3 m) upstream of the location of a stop, (2) within 255 ft (75.6 m) downstream of the location of a stop, (3) within the area where deceleration of 0.16 g and above is detected, and (4) up to 51 ft (15.5 m) downstream of the location where deceleration passed below 0.16 g. Less restrictive settings may be justified for a particular profiler design using the recommended dynamic testing in AASHTO R56. The four specialized certification tests proposed for AASHTO R56 provide a framework for assessing the sen- sitivity of inertial profilers to difficult operational condi- tions. Motivating the development and implementation of improvements to the profiling technology for use on urban and low-speed roadways depends on broad implementation of the requirements and will take time. Some improvement is possible with careful selection of the profiler host vehicle. Mounting a profiler to a host vehicle that resists changes in pitch and roll orientation is encouraged. Although host vehicle properties were not explicitly examined in this research, characteristics that typically help a vehicle main- tain a consistent orientation include a long wheelbase, a wide track, a low center of gravity, stiff suspensions, suspension anti-dive, and a high suspension roll center. 5.2.1.3 Diagnostic Testing The experimental evaluation of production inertial profil- ers revealed a systematic error in the measurement of the IRI by one of the units that depended on travel speed. The cause, which has since been corrected, affected an aspect of the profiler’s performance that is not captured by essential data quality checks such as the bounce test, height sensor block checks, or certification of the accuracy and repeatability of measured profile at the recommended speed of operation. However, comparison of profile and measured IRI at a high speed to measurements at a low speed revealed the existence of a problem with minimal testing effort and provided the basis for diagnostic analysis. AASHTO R57 suggests that network profilers regularly measure control sections to demonstrate that they are pro- ducing consistent IRI measurements. For network-level and project-level profilers, it is recommended that the procedure in AASHTO R57 include measurement of a pavement sec- tion at two different speeds. The test speeds should be as different as possible within the valid range for the profiler and the safe operating range for the test section. The recom- mended procedure requires consistent measurement of IRI and profile at the two speeds. This testing is recommended as a means to identify measurement errors that affect short- wavelength profile content but typically do not affect the bounce test, such as high frequency vibration of the profiler without proper correction between the height sensors and accelerometers (e.g., vibration of the profile on its mounting system, excessive tire imbalance, etc.) or incompatibility in sensor signal timing. 5.2.2 Characterization 5.2.2.1 Localized Roughness This research showed that transient vibration caused by localized roughness is an important element of the functional performance of urban and low-speed roads. Further, report- ing of average roughness of a given road segment alone does not sufficiently describe user discomfort. It is recommended that, in addition to average roughness, surveys of roughness in urban and low-speed roadways record the location and magnitude of severe peaks in the short-interval roughness profile using a moving average base length of 25 ft (7.6 m). The experimental measurements did not provide an objec- tive basis for setting absolute thresholds for peak values of localized roughness. Agencies may need to establish thresh- olds based on engineering judgment with user expectations in mind. However, in the vehicle response measurement experiment, nearly every case where the peak value of the short-interval roughness profile [using a base length of 25 ft (7.6 m)] was at least 2.5 times the overall average caused transient vibration in the test vehicle. Using a base length of 25 ft (7.6 m), any peak in the short-interval rough- ness profile signifies that transient vibration is likely in pass- ing vehicles. This implies that the specific feature causing the peak influences user discomfort disproportionately. 5.2.2.2 Roughness Index Use of GCARV, which is an IRI-based temporal index, is recommended as a specific scale for estimating the func- tional status of urban and low-speed roads. This option maintains the relevance of the underlying IRI algorithm to a broad range of vehicle types and to a broad range of vehicle responses. However, use of GCARV produces roughness on an unfamiliar scale, and field experience and further research are needed to establish thresholds for various applications. 5.2.3 Reporting 5.2.3.1 Localized Roughness Many of the profiles from urban and low-speed roadways examined in this research had localized roughness—that is, roughness concentrated within a small area that is much greater than the segment-wide average. It is recommended

73 that network-level surveys record areas of localized roughness in addition to segment-wide averages for two reasons. First, for segments with the same average roughness the repair or rehabilitation strategy will be different for a section where a large share of the roughness is concentrated in a few loca- tions. Second, localized roughness may not increase the segment-wide average in proportion to the degradation in ride quality caused by the underlying feature. Addressing areas of localized roughness on road segments with accept- able average roughness may provide a better return on invest- ment for improving the functional performance of a road network. Per AASHTO R 54-14 (Standard Practice for Accepting Pavement Ride Quality When Measured Using Inertial Pro- filing Systems), a roughness profile with a base length of 25 ft (7.62 m) is recommended for identification of localized roughness using the IRI. A standard base length is required to maintain a consistent interpretation of the peak value in a roughness profile. Since the IRI algorithm responds to isolated rough features for some length downstream, a base length less than 25 ft (7.62 m) may fail to capture their severity. Using a base length greater than 25 ft (7.62 m) increases the like- lihood that more than one built-in feature will interact. For the profiles examined in this research, a base length of 25 ft (7.62 m) offered a suitable trade-off between variability and localization. Use of a roughness profile with a base length of 25 ft (7.62 m) has also emerged as a successful practice for smoothness of bridges and bridge approaches (Ohio Depart- ment of Transportation 2012). 5.2.3.2 Identification of Roughness Sources Road profilers offer the potential to provide value beyond the measurement of roughness and enable pavement engi- neers to identify and diagnose roughness sources. In particu- lar, some applications may require an engineer to distinguish between roughness at built-in features and roughness caused by distress or construction imperfections. Although rough- ness at built-in features affects the functional performance of a roadway segment, its implications to pavement structural health are much different than roughness associated with construction defects or caused by surface distress. Provisions for identifying built-in features and other potential sources of roughness are recommended to better leverage profile mea- surement for the following applications: • Project level: Prioritization of efforts to reduce roughness on urban roadways will require engineers to distinguish between built-in features and pavement surface distress, particularly when localized roughness is detected. Know- ing the sources of roughness may help determine the best rehabilitation strategy for improving both ride quality and structural longevity and to set expectations for the reha- bilitated surface. In forensic applications, it may be useful to know the contribution of built-in features to the over- all roughness, particularly when a section has reached a terminal roughness level. • Network-level roughness surveys: Pavement network management engineers may benefit from a statistical assessment of the effect of bridges, railroad crossings, intersections, etc. on the overall functional performance of their network and review the status of a subset of the road- way network that only includes pavement without built-in features. Further, a more detailed breakdown of potential roughness sources for the portion of the pavement net- work with undesirable status (e.g., “fair” or “poor”) may be useful. • Asset management: Identifying built-in roadway features that are present for other infrastructure, such as rail- roads, bridges, electrical and water utilities, and sewers, is needed to support an integrated asset management system (21st Century Infrastructure Commission 2016). Integrated asset management enables the coordinated repair and upgrade of multiple infrastructure systems, so that new road surfaces are not immediately damaged for work on another system. • Construction quality control and quality assurance: The owner/agency and contractor have a shared stake in recog- nizing potential benefits of provisions to reduce or avoid roughness at built-in features relative to their cost and the relative success of those provisions in terms of the rough- ness of the finished surface. The following recommendations for profile measure- ment practices and capture of additional data are offered to support the uses of profile measurements described above, but are not required for the core function of maintaining a database of segment-by-segment roughness measurements. Over the short term, minimizing distortion and maxi- mizing detail within profile measurements will help pave- ment engineers identify and diagnose sources of localized roughness by inspecting profiles. The following practices are suggested: • Apply high-pass filtering to profiles processed for storage with as little phase distortion as possible. • Collect profile using a wide-footprint height sensor, and apply a tire-bridging algorithm to the individual readings within the footprint. • Apply low-pass filtering with removal of no more of the measured waveband than is required for anti-aliasing. • Identify and remove height sensor dropouts using crite- ria that do not introduce artificial upward spikes into the profile.

74 • Replace the 9.84-in (250-mm) moving average in the IRI computation algorithm with a bridging algorithm that eliminates narrow dips in the profile, or apply an equiva- lent filter while measuring the profile. Phase distortion alters the shape of profiles by shifting the components of various features differently. This is addressed in suggested additions of AASHTO M328-14, Section 4.2.2, which specifies profiler high- and low-pass filtering perfor- mance, including phase distortion. Collectively, the suggested practices listed above maximize the detail within the profile that is available for identifying road features without allowing spurious profile content to affect the measured roughness. In some cases, these sugges- tions may be applied using existing measurement equipment or with minor changes to existing equipment. In other cases, they may be considered when new equipment is procured. Over the long term, collection of additional data synchro- nized with profile to support or automate the identification of built-in features is recommended. • Right-of-way images or pavement surface images provide the means to identify the potential cause of localized rough- ness once it is detected within the profile. Collection and storage of images also create a foundation for an automatic feature detection and recognition system. • GPS coordinates provide the means to identify the specific location of an area of localized roughness and leverage information from other geospatial databases to overlay existing knowledge about built-in features that may cause roughness. These items are recommended to support the development of an inventory of built-in roadway and pavement features that may affect roughness. This inventory would exist side- by-side with functional and structural performance measures within a pavement network management database. Collec- tively, the database would enable pavement managers to assess the effect of built-in features on functional performance. Existing geospatial databases currently include several built-in features of interest. Publicly available databases— such as the National Highway-Rail Crossing Inventory, the National Bridge Inventory, and city or county road shape files for identifying intersections—are available as an interim step. Leveraging these databases will require some algorithm development or fusion with other databases. Green et al. (2015) demonstrates this in an effort to deduce the type and geometric extent of intersections from roadway and intersec- tion node points. Not all of the features of interest will appear in public databases. An agency that endeavors to develop a database of built-in roadway and pavement features can augment accessible geospatial databases with in-house data collection if the additional coverage justifies the cost. 5.3 Suggestions for Future Research This research addressed special issues related to mea- surement of profile on urban and low-speed roadways during braking and with stops. This includes recommended enhancements to profiler certification procedures that require additional effort, and these may require well-trained and experienced personnel to implement. Many agencies that certify inertial profilers may lack the resources to implement the proposed testing. As such, pilot implementation of these procedures administered within a national or regional pro- filer certification program is recommended. The profiler cer- tification procedures recommended here would serve as an extension to the procedures envisioned by the FHWA for a regional profiler certification center (Perera and Karamihas 2014). Testing at a regional or national certification pro- gram may help motivate profiler manufacturers to adapt their technology for urban and low-speed settings because the visibility could improve the possibility of a return on their investment. At the time this report was written, profile equipment ven- dors had begun to offer measurement of longitudinal pro- file extracted from three-dimensional surface scans collected from sensors mounted high on their host vehicles, often more than 6 ft (1.8 m) above the ground. These systems use the same measurement concept as conventional high-speed iner- tial profilers. Due to the differences in sensor specifications and physical layout, however, methods from this research are recommended for the evaluation of their performance during braking and stop-and-go operation. The research confirmed that inertial profilers are inherently vulnerable to measurement errors caused by misalignment of the accelerometers. As such, a robust long-term solution will require the development and implementation of a measure- ment concept that does not depend on body-mounted accel- erometers and is capable of valid profile measurement during horizontal acceleration and during stop-and-go operation. This research examined correlation between objective measurements of discomfort experienced by a driver and variations on the IRI calculation algorithm. More research is needed to refine the results of this examination for practical implementation. Specifically, the following: • Use of GCARVV, which is recommended for urban and low-speed roadways, will require a strategy for selection of a specific travel speed for each road segment for which the index is calculated.

75 • Additional research is required to establish threshold values for roughness on the GCARVV scale. The thresholds must correspond to various applications of road rough- ness measurement, such as construction acceptance and triggers for intervention based on functional performance. • Establishment of threshold values for a new roughness scale must account for a broad range of vehicle responses, as well as user expectations on urban and low-speed roadways. This study examined objective measurements of driver comfort based on vibration experienced by the driver, but did not include other important vehicle responses to road roughness (such as suspension stroke, dynamic tire loads, and cargo wear) or user option of ride quality. • The experimental measurements in this research included a broad variety of built-in roughness types and surface distress, but it did not include many test sections with high long-wavelength content (e.g., “wavy” roads). In particu- lar, the experiment did not include parkland roads, roads on uneven terrain, and low-volume roads where provi- sions for grade control were not applied and only very low travel speeds were anticipated. Subsequent research that includes these types of roads may further elucidate the contrast between high-speed roads and low-speed roads. This may have practical implications to construction quality assurance, where the investment in improving the longer-wavelength portion of the roughness that affects the IRI may not be justified on roads with very low travel speeds. This report recommends automated identification of built-in roadway features as a resource for interpreting roughness measurements within the context of pavement asset management and in support of integrated management of public assets. Research to identify the most efficient and robust methods for developing, augmenting, or accessing databases of geo-located asset inventories is encouraged. This may include automated interpretation of pavement images, recognition of known patterns within measured profiles (longitudinal and transverse), and incorporation of existing databases from multiple sources.