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.

9 This chapter describes features that are commonly observed in urban and low-speed road profiles. Like high-speed limited- access freeways, urban and low-speed roadways include roughness caused by construction defects, pavement distress, and environmental degradation. However, the longitudinal profiles of many urban and low-speed roadways differ from typical limited-access freeways in two ways. First, urban and low-speed pavements are often built without as much opportunity for grade control. In some cases, rural roadways designed for low-speed operation tra- verse a landscape where manipulating the grade is not prac- tical or not permitted. Urban paving requires compatibility with existing features such as intersections, or cross streets, or matching the elevation along adjacent lanes, curbs, and driveways. Second, longitudinal profiles of urban and low-speed pave- ments often include roughness at specific locations associated with roadway design elements, hardware in the roadway that does not functionally support the safety and maintenance of the roadway infrastructure, and repairs for excavations. This includes several categories that may affect roughness: • Built-in roadway features: These are aspects of the road design that satisfy pavement-related engineering require- ments, including longitudinal grades for drainage, drain- age inlets, crowned intersections, specialized surfaces (e.g., cobblestones), etc. • Built-in features for right-of-way access: These are aspects of the roadway design required to accommodate right-of- way access for other infrastructure, such as railroad cross- ings, textured crosswalks, trolley tracks, overpasses, and underpasses. • Utility access: These are alterations made to the pavement for access to underground infrastructure, such as sewer lines, gas lines, electrical or communication lines, etc. • Excavation of in-service pavement: These are alterations made to the pavement for repair of underground utilities and require patching of the pavement surface. Some of these features cause built-in roughness because they impose constraints on the design profile that are in con- flict with the engineering requirement of a flat and smooth roadway. These include crowned intersections and grade changes for drainage or compatibility with other right-of- way elements (e.g., railway crossings, intersecting roads, etc.). Built-in roughness also occurs at built-in features installed on the roadway (e.g., utility covers) with allowances for sur- face imperfections, or when an installation is built to within tolerance, but is not perfectly smooth. For example, guide- lines for highway-railroad grade crossing geometric design includes criteria for sight distance and vehicle ground clear- ance, but no explicit requirement for ride quality of the intersecting roadway (Eck and Kang 1991; Wooldridge et al. 2000; Ogden 2007). Roughness also occurs at built-in features that are not installed to within tolerance or when the presence of the fea- ture causes accelerated pavement deterioration. Causes of accelerated wear that affect surface roughness include struc- tural weakness at interfaces between materials (Wilde et al. 2002), poor protection from water penetration at interfaces between materials, and settlement of backfill material at util- ity cuts (Schaefer et al. 2005). Maintenance of existing built-in roadway features may also exacerbate roughness. For exam- ple, adding ballast to the track structure at a railway crossing often requires the track to be raised, which may reduce its compatibility with the roadway (Sobanjo 2006; Ogden 2007). This chapter presents typical examples of the effects on profile and roughness of built-in features from urban and low-speed roadways in New Jersey and in Philadelphia County, Pennsylvania. Roughness is quantified in terms of the influence on the IRI and, in the majority of cases, in terms of localized contributions to the IRI. In addition to causing roughness, many built-in surface features present a challenge to road profile measurement and interpretation. This chapter discusses profile measurement and interpretation issues that were encountered in the exami- nation of roughness at built-in road features. C H A P T E R 2 Features on Urban and Low-Speed Roadways

10 This chapter presents a limited set of examples of rough- ness at built-in features present on urban and low-speed roadways in support of the technical discussion. Appendix A provides a broader set of examples. 2.1 Data Sources The study of urban and low-speed roadway features pri- marily used three data sources: (1) right-of-way images from the Pennsylvania Department of Transportation (DOT) pavement 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 seg- ments. Straight-line diagrams show the locations of intersec- tions, traffic signals, overpasses, railroad crossings, etc., as a function of distance along a road section. The images from Philadelphia County included a wind- shield view every 21 ft (6.4 m) along all state routes and many Local Federal Aid roads. Each image identified the profiler location and speed. Review of these images helped identify road segments for further analysis that included roughness unique to urban and low-speed roadways, including exam- ples of built-in features with the potential to cause roughness. Review of the images also provided examples of adverse oper- ational conditions encountered when collecting profile data in congested urban environments, such as slow traffic flow and traffic stops. This helped determine the test conditions to replicate in the experimental profiler evaluation described in Chapter 3. Profile data from 2013 and 2014 in Philadelphia County included 63 road segments, which were typically 2,000– 3,500 ft (610–1,067 m) long. The routes included two inter- states, 3 U.S. routes, 6 other state routes, and 15 local streets. Profiles were provided with synchronized right-of-way images that were collected simultaneously. Most of the pro- files were recorded at a 1-in (25.4-mm) interval, high-pass filtered with an anti-smoothing moving average using a base length of 300 ft (91.44 m) and low-pass filtered with a cut-off near 1 ft (0.3 m). Profiles of the interstates and four other routes were recorded at a 0.73-in (18.5-mm) interval and provided with minimal high-pass filtering (i.e., a cut-off wavelength of at least 1,000 ft) and no low-pass filtering other than anti-aliasing filters native to the individual sensors. Data from New Jersey included right-of-way images, straight-line diagrams, and three repeated profile measure- ments over each lane. These data covered portions of ten state highways in urban areas that were 1.24–7.28 mi (2.0–11.7 km) long. Right-of-way images were recorded every 20 ft (6.1 m). Profiles were recorded at a 2-in (50.8-mm) interval and were provided with no low-pass filtering other than anti-aliasing filters native to the individual sensors. The profiles were high- pass filtered with a third-order Butterworth filter with a cut- off wavelength of 300 ft (91.44 m). 2.2 Built-in Roughness 2.2.1 Roughness Profiles AASHTO R54-14 defines localized roughness as any 25-ft (7.62-m) segment of roadway that contributes dispropor- tionately to the overall roughness index value. This definition is based on a roughness profile, which provides continuous report of IRI values for a given base length (Sayers 1990). Roughness profiles with a sufficiently short base length pro- vide a detailed view of the way features that contribute to the IRI are spatially distributed. Several researchers have used roughness profiles or calcu- lations of IRI over short intervals to quantify roughness on short urban streets and at built-in road features. La Torre et al. (2002) recommended a 164-ft (50-m) base length for relating roughness to user perception in congested urban areas where the length of streets is short. Reggin et al. (2008) used a 66-ft (20-m) interval to characterize roughness at grade changes present on a roadway for drainage, railway crossings, and rutted intersections. Williams (2003) calculated IRI on 10-ft (3-m) segments at railroad crossings and classified the rough- ness using the short-interval value, comparing it to the rough- ness of the 0.1-mi (0.16-km) segment surrounding the crossing. Rose et al. (2009) used a 25-ft (7.62-m) base length to character- ize roughness in a study of ride quality over railway crossings. Swan and Karamihas (2003) used roughness profiles with a 25-ft (7.62-m) base length to identify construction defects that degrade ride quality on newly paved and resurfaced roads. Several studies of roughness progression on Long-Term Pave- ment Performance (LTPP) Specific Pavement Studies (SPS) sites quantified the development of roughness at distressed areas using a 25-ft (7.62-m) base length (Karamihas 2007; Karamihas and Senn 2009, 2010). The discussion that follows quantifies localized roughness at built-in road and pavement features using short-interval roughness profiles with 25 ft (7.62 m) as the primary choice for the base length. Figure 1 shows a utility cover in the right wheel path of the outside lane that is not flush with the pave- ment surface. Figure 2 shows the elevation profile and a short- interval roughness profile for a segment that includes the utility cover. The surface of the cover in Figure 1 is approxi- mately 1.25 in (32 mm) below the surrounding pavement sur- face. The roughness profile is shown using a 25-ft (7.62-m) base length. Each point in the roughness profile is the IRI of a 25-ft (7.62-m) long segment, which covers a length ranging from 12.5 ft (3.81 m) upstream to 12.5 ft (3.81 m) downstream of that location.

11 The roughness profile rises to a peak value of 890 in/mi (14.05 m/km) at the utility cover. Since this is so much greater than the roughness in the surrounding area, the utility cover is considered a source of localized roughness. In a 0.1-mi (0.16-km) road segment, the 25-ft (7.62-m) interval rep- resents 4.7 percent of the total length. As such, the rough- ness that caused the 890-in/mi (14.05-m/km) peak accounts for about 42.1 in/mi (0.67 m/km) of the average IRI for the segment. 2.2.2 Localized Roughness Localized roughness was found at many built-in features, including crowned intersections, drainage inlets, utility covers, railway crossings, trolley track crossings, textured crosswalks, bridge approaches, bridge joints, and Portland cement con- crete (PCC) pads at bus stops. Localized roughness was also found at patches installed after underground utility work and where metal plates covered an ongoing utility cut. Not all of these features cause localized roughness in every instance. Localized roughness appeared at many utility covers where the cover itself or the patching surrounding the cover was not level with the pavement. However, many cases were observed where the profiler tracked over a utility cover and registered little or no additional roughness above the pre- vailing level for that road segment. This was often true on the newly resurfaced road segments in New Jersey, where the paving process may have included provisions for avoiding roughness at utility covers. Roughness at the locations of many built-in features varied transversely. For hardware on the pavement surface with a narrow footprint relative to the lane width, such as utility covers and drainage inlets, detection of roughness included a “hit or miss” quality depending on whether the sensors in the profiler tracked over them or what profile features on them were captured. For example, Figure 3 shows a utility cover near the left wheel path of the outside lane, but biased toward the center of the lane. Localized roughness was detected at this utility cover in one of three passes by a profiler for construction quality assurance. Figure 4 shows a close-up view of each elevation profile and the corresponding roughness profiles. Pass 1 reg- isters a dip at the utility cover, and a peak value in the rough- ness profile of 263 in/mi (4.15 m/km). Some evidence of the cover appears in passes 2 and 3, but it is not clear whether Figure 1. Utility cover, right wheel path of outside lane. Figure 2. Elevation and roughness profiles over a utility cover.

12 the profiler tracked over the left edge of the cover or detected unevenness of the pavement near the cover. Figure 5 shows a segment of a drawbridge with drainage inlets at the right lane edge. Figure 6 shows the elevation pro- file and a short-interval roughness profile over 600 ft (183 m) of the bridge that includes ten drainage inlets and two bridge joints. The inlet and joint shown in Figure 5 appear near 15,075 ft (4,595 m) in the profile. The elevation profile is high-pass filtered with a cutoff wavelength of 20 ft (6.1 m). This eliminates longer wavelength undulations and reduces the vertical scale of the plot so details of the profile at the drainage inlets and joints are easier to identify. Localized roughness for the data collected by the profiler was not consistent among the ten drainage inlets for multiple reasons. The profiler may have tracked directly over some of the drainage inlets, but may have missed others and collected data on the pavement adjacent to the inlet. The height of each inlet grate relative to the surrounding pavement, as well as other aspects of their roughness, may have been different at each inlet. In some instances, the height sensor passed directly over a gap in the surface grate and registered a deep narrow dip. Narrow dips up to 1.25 inches (32 mm) deep appear in the profile. The peak value of the roughness profile in the area near 15,075 ft (4,595 m) includes contributions from the drainage inlet and the nearby bridge joint. 2.2.3 Compound Events In some cases, built-in features that contribute to localized roughness occur in close proximity to each other, and it is Figure 3. Utility cover, center-left of the outside lane. Figure 4. Elevation and roughness profiles of a road segment with a utility cover. Figure 5. Drainage inlet and bridge joint.

13 hard to identify the individual contribution of each feature to localized roughness. Locations where this phenomenon occur are referred to as a compound event in this report. The following are two examples of locations where a compound event occurs: • Intersections: Many intersections included a combination of several built-in features, including textured pedestrian crossings, crown of the cross street, drainage inlets along the curb, one or more utility covers, PCC pavement pads at a bus stop, and longitudinal grade breaks of pavement profile at curb inlets provided for drainage. Rutting of the cross street pavement, longitudinal cracking associated with a longitudinal construction joint, and trolley tracks also affected the roughness at some intersections. • Bridges: Roughness often appeared at bridge approach and leave areas, directly at the pavement/bridge interfaces, and at joints within the bridge deck. Several bridge profiles included roughness at drainage inlets in the right wheel path. Drawbridge profiles included roughness at metal grates over the area surrounding the location where the two sides of the drawbridge meet. In both instances, the built-in features appear as a group, and the layout of these features depends on the surrounding infrastructure, abutting property, or underlying terrain. Figures 7–11 provide an example of a compound event where several built-in features contribute to the roughness at an urban intersection. Figures 7–9 show portions of three right-of-way images collected as the profiler passed through the intersection. The right side profile of this area includes the following: • A concrete pad for a bus stop (Figure 7), • A textured pedestrian crossing at the leading end of the intersection (Figure 7), • A crowned intersection (Figure 8), • A utility cover within the intersection (Figure 8), • A textured pedestrian crossing at the trailing end of the intersection (Figure 9), • A utility cover that appears after the pedestrian crossing (Figure 9), and • A drainage inlet adjacent to the curb (Figure 9). Figure 10 shows the elevation profile and a short-interval roughness profile for the intersection and the surrounding area. The elevation profile is high-pass filtered to make some of the details associated with built-in features more visible. As a compound event, the built-in features associated with the intersection run from the start of the concrete bus pad at 4,720 ft (1,439 m) to the slope break at the drainage inlet 4,885 ft (1,489 m). The roughness profile with a base length of 25 ft (7.62 m) shows that the roughness in this range is higher than the surrounding area. However, the plot does not isolate the contribution of each feature within the intersec- tion to roughness, since they are so close together. Figure 11 shows a closer view of the elevation profile for the same intersection. In this view, individual features are more Figure 6. Elevation and roughness profiles with drainage inlets and bridge joints.

14 visible, such as the crown of the cross street, disturbances at the concrete strips located on both sides of each pedestrian crossing and the surrounding pavement, and the utility covers. Figure 11 also shows a roughness profile using a base length of 10 ft (3 m). Areas with the most severe roughness stand out in the plot. This provides a way to identify the largest contribu- tors to the roughness throughout the intersection. Figures 10 and 11 show that the choice of base length in roughness profiles presents a trade-off between localiza- tion and ease of interpretation. The effects of features that are in close proximity to each other overlap and interact to a greater extent with a 25-ft (7.62-m) base length than with a 10-ft (3-m) base length. However, the response of the IRI algorithm to disturbances in the profile typically persists for more than 10 ft (3 m), even though the feature exists within a smaller length. In a typical compound event, no perfect choice of base length for a roughness profile is possible. For example, the response of the IRI algorithm to the slope break at the end of the intersection has not fully diminished at the location of the sunken utility cover. Note that ratings of the severity of localized features produced by roughness profiles with different base lengths cannot be interpreted on the same quantitative scale, because a decrease in base length causes an increase in the severity of peak values. 2.2.4 Idealized Profiles Researchers have used idealized geometric representa- tions of built-in roadway and pavement features to estimate their potential contribution to roughness, including grades for drainage (Reggin et al. 2008), grade changes at inter- section approaches (Movassaghi et al. 1993), camber of bridge spans (McGhee 2002), and rutting at intersections (Reggin et al. 2008). For many of the built-in features identified in this chapter, no idealized shape appeared consistently in the profiles. Typi- cally, the roughness of an idealized shape can be thought of as a lower bound on the potential roughness at a built-in fea- ture, because the idealized shape only captures some aspects of the roughness associated with that feature. However, the contribution of localized disturbances to the IRI is not inde- pendent of the roughness around it. As a result, the roughness of a given disturbance within an otherwise smooth profile is merely an estimate of its potential influence on the roughness of a profile with other imperfections. This section presents examples of roughness at idealized profiles of grade breaks, utility covers, and crowned inter- sections using slope breaks, a rectangular disturbance, and a half-sine wave. 2.2.4.1 Slope Breaks Reggin et al. (2008) provided an example of an idealized and measured profile over grades built into an urban street to direct water to catch basins (see Figure 12). In the ideal- ized profile, areas of constant slope connect grade breaks Figure 7. Built-in features preceding a crowned intersection. Figure 8. Crowned intersection crossing. Figure 9. Built-in features following a crowned intersection.

15 Figure 10. Elevation and roughness profiles through a crowned intersection. Figure 11. Elevation and roughness profiles through a crowned intersection, close-up view.

16 with “average distance between them of 83 m” (272 ft) and changes in slope “from 0.66% to 0.82%.” Roughness profiles using a base length of 61 ft (20 m) include peak values of 102–126 in/mi (1.6–2.0 m/km) at the five grade breaks. Using the roughness profiles of the idealized profile, Reggin et al. (2008) deduced that the grade breaks contributed 26.6 in/mi (0.42 m/km) to the average IRI of the measured profile over the 1,640-ft (500-m) long section shown in Figure 12. The grade breaks were sufficiently far apart that the response of the IRI to each instance did not interact with the others. The IRI of a 528-ft (160.9-m) long section with one grade break and no other roughness is 21.0 in/mi (0.33 m/km) per percent change in grade, and a roughness profile with a 25-ft (7.62-m) base length will have a peak value of 240.4 in/mi (3.79 m/km) per percent change in grade. 2.2.4.2 Rectangular Disturbance No prevailing idealized shape appeared consistently at utility covers in the profiles analyzed for this study. This is due to the following: • Utility covers had different dimensions; • The sensors in the profilers did not always track directly over utility covers, and in some cases the collected data detected roughness at a location beside the utility cover; • Many covers were surrounded by patching or the adjacent pavement was distressed; • The true shapes were altered in the measured profiles by low-pass filtering; • The surface of various covers were not always flush with the housing or included some texture; and • Gaps between the cover and the housing or the housing and the surrounding pavement affected the measured profiles. R el at iv e E le va tio n (m m ) Figure 12. Profile with grades for drainage (reproduced with permission from Reggin et al. 2008). Figure 13. Roughness of a rectangular disturbance. The roughness at a rectangular disturbance may represent a lower bound on the potential roughness at a utility cover caused by a difference between the height of the cover and the height of the longitudinal profile around it. Figures 13 and 14 show the IRI of an idealized rectangular disturbance of height “H” and width “W” on a segment of profile that is otherwise perfectly smooth. Figure 13 shows the average IRI of a 528-ft (160.9-m) seg- ment of profile that contains a rectangular disturbance. IRI is normalized by height. For a disturbance with a width of 3 ft (0.9 m), the roughness is 43.2 in/mi (0.68 m/km) per inch of height. That means a rectangular disturbance 3 ft (0.9 m) wide and 0.5 in (12.7 mm) high registers as 21.6 in/mi (0.34 m/km) when its influence is averaged over a 528-ft (160.9-m) section. Figure 14 shows the peak value in a roughness profile with a base length of 25 ft (7.62 m) on a segment of profile that contains a rectangular disturbance but is otherwise perfectly smooth. For a disturbance with a width of 3 ft (0.9 m), the peak roughness is 855 in/mi (13.5 m/km) per inch of height. As such, a rectangular disturbance 3 ft (0.9 m) wide and 0.5 in

17 (12.7 mm) high registers a peak value in the roughness profile of 427 in/mi (6.7 m/km). 2.2.4.3 Half-Sine Wave No idealized shape appeared consistently at crowned inter- sections in the profiles analyzed for this study. This is due to the following: • Crowned intersections usually included other built-in features, • The profile over the cross street depended on the type of road and its overall width, • Grades for drainage affected the profile of the approach and leave areas, and • The true shapes were altered in the measured profiles by high-pass filtering. The roughness at a disturbance in the shape of a half-sine wave may represent a lower bound on the potential roughness at intersections without drainage inlets or close-proximity curbs at its boundaries. Figures 15 and 16 show the influence of an idealized half-sine wave with height “H” and width “W” on the IRI in the absence of any other roughness. Figure 15 shows the average roughness over a 528-ft (160.9-m) section, and Figure 16 shows the peak value of a roughness profile with a base length of 25 ft (7.62 m). Roughness is normalized by height. As shown in Figures 10 and 11, the presence of other sources of roughness typically increases the roughness above the level predicted for a half-sine wave. 2.2.5 Distributed Roughness Few of the profiles analyzed in this study included areas without localized roughness. As such, no typical characteris- tics of distributed roughness emerged in the analysis. Close- proximity curbs, driveways, and junctions occurred in many of the evaluated roadway segments, which may have caused roughness as a result of complicating the paving process. However, the influence of these items could not be distin- guished from other sources of roughness. Distributed roughness also occurred on segments where the profile straddled trolley tracks in the lane, a segment with a cobblestone surface, and on bridge decks with highly textured surfaces. Each of these built-in features caused a very distinct type of roughness, which in some cases could be clas- sified as areas with a high density of localized roughness. One method of characterizing the roughness of roads is by examining spectral density plots. However, the spectral den- sity plot of a road profile may be misleading if the content is not stationary (i.e., roughness is not evenly distributed over the length of the profile) or not Gaussian (e.g., the profile includes spikes or rapid elevation changes). In particular, the spectral density plot omits important information for pro- files that include a high level of localized roughness. Nearly all of the profiles analyzed for this study included sufficient localized roughness to render the power spectral density plot an insufficient representation of the content. In areas of the profiles without localized roughness, where the spectral Figure 14. Peak localized roughness of a rectangular disturbance. Figure 15. Roughness of a half-sine disturbance. Figure 16. Peak localized roughness of a half-sine disturbance.

18 density plot is more informative, the long-wavelength con- tent accounted for a larger share of the overall roughness than on typical high-speed, limited-access roads. Figure 17 shows the slope spectral density plot for a low- speed roadway without localized roughness. This is a two- lane undivided road segment with a posted speed limit of 40 mi/hr (64 km/hr). The spectral density of slope is shown, rather than elevation, to help illustrate the relative contribu- tion of long-wavelength and short-wavelength content. To account for the content in various portions of the waveband, Sayers (1986) proposed fitting measured spectral density to a function made up of a combination of white noise sources: ( ) ( ) ( ) ′ = π + + π−2 2 (1)2 2G v G v G G v e s a where: G ′(v) = the slope spectral density Ge = the contribution of white noise elevation Gs = the contribution of white noise slope Ga = the contribution of white noise spatial acceleration v = wave number (i.e., the reciprocal of wavelength). Sayers cited pure white noise slope as a “normal” high-speed roadway, which corresponds to Ga = 0 and Ge = 0. Sayers also observed that the ratio Ga/Gs = 0.0325 ft-2 (0.350 m-2) repre- sents “asphalt with long waves,” and designated this ratio as a case in which the long wavelength range accounted for an usually large share of the roughness. Figure 17 shows the result when the Sayers model is fitted to the spectral density function. In this case, Ga/Gs = 0.0766 ft-2 (0.824 m-2), which indicates an even greater share of content in the long wavelength range than the “asphalt with long waves.” This is typical of low-speed roadways that are not in con- gested urban areas and do not contain localized roughness. 2.3 Measurement Issues This section describes aspects of profile measurement practice that may confound the interpretation of urban and low-speed road profiles. A large proportion of the roughness on the urban and low-speed road profiles analyzed for this study occurred at built-in features, and localized roughness at built-in features, patches, and areas of pavement distress were common. The right-of-way images provided with the profiles were instru- mental in identifying the specific sources of roughness. How- ever, identification of specific features solely by viewing the profiles, and in some cases reconciling profile plots with the images, was often difficult. This section discusses aspects of the profile measurement process that may hinder the use of road profile measurements for diagnostic applications on urban and low-speed roadways. 2.3.1 Transverse Variations Features such as utility covers and drainage inlets often appear within or near the wheel path, and the roughness they contribute to a profile depends on the lateral position of the profiler (see Figures 3–6). The “hit or miss” nature of these items may cause a profile measured in only two narrow wheel paths to misrepresent the experience of the public. Some narrow built-in features that run longitudinally interact with profiler lateral wander to introduce roughness Figure 17. Slope spectral density plot of a low-speed roadway.

19 into the longitudinal profile that is disproportionate to their likely effect on ride quality. These include longitudinal con- struction joints within driving lanes and trolley tracks within a lane that run in the same direction as the pavement. 2.3.2 High-Pass Filtering High-pass filters reduce or eliminate the content in a pro- file at frequencies below a particular value and leave content at frequencies above a particular value intact. High-pass fil- ters are often characterized by a cut-off frequency where the content is reduced by 50 percent. Depending on the filter design, there are some ranges of frequencies above the cut- off value where the content is virtually unaffected. In the con- text of road profiling, the cut-off value is usually expressed in terms of wavelength, where longer wavelength content that corresponds to lower frequencies is removed. High-pass filtering is a common practice in the profile measurement process because the sensors in inertial profilers are not sufficiently sensitive to detect the very slow changes in elevation that occur over long stretches of road. High-pass filters are also applied to eliminate grade and gradual grade changes from the profile so features that affect roughness are more visible. AASHTO M 328-14 mandates high-pass filter- ing with a cut-off wavelength that is 300 ft (91.44 m) or lower, with undistorted response for wavelengths of 150 ft (45.72 m) and below. In accordance with these requirements, use of a cut-off wavelength of 300 ft (91.44 m) is common. The con- sequence is the removal of “trends” that would appear in the profile over lengths much greater than the typical high-pass filter cut-off wavelength. The high-pass filtering distorts the profile of built-in road features by removing characteristics with length equal to or greater than the cut-off wavelength. As a result, features asso- ciated with restrictions on grade are difficult to identify. This includes grades for drainage and cases where gradual changes in profile are required for compatibility with other infrastruc- ture, including bridges, railroad crossings, and intersecting roads. For example, Movassaghi et al. (1993) analyzed the contribution to the IRI of elevation design profiles surround- ing intersections, including six types of vertical transition- design curves. All six designs were characterized using a distance of at least 500 ft (152.4 m) before and after the inter- section. Application of a high-pass filter with a cut-off wave- length of 300 ft (91.44 m) removes most or all of the content in the profile associated with the design profiles. Figure 18 shows a facsimile of the profile measured by Reggin et al. (2008) on a road with grades for drainage. Fig- ure 19 shows the traces produced when three different high- pass filters are applied to the profile shown in Figure 18. In all three traces, grade information is removed and the vertical range is smaller than the original elevation profile. Although the slope breaks are still visible within the profile, the distor- tion obscures useful information about the road profile. Applying an anti-smoothing moving average with a cut-off base length of 300 ft (91.44 m) produced the lower trace. Each slope break affects the profile over 150 ft (45.72 m) to either side of it, and a step change in slope occurs in the profile at locations 150 ft (45.72 m) before and after each break. This is best illustrated at the transition to a slope of zero at 1,177 ft (358.7 m) and from a slope of zero at 1,324 ft (403.6 m). The IRI registers roughness at these locations. Applying a third-order Butterworth filter with a cut-off wavelength of 300 ft (91.44 m) produced the middle trace. This filter imposes non-linear phase distortion, which shifts some portions of the content retained by the filter a differ- ent distance than others. In particular, the longer wavelength content that is passed by the filter is shifted downstream. Hu et al. (1979) noted this in a study of heavy truck dynamic loads caused by roughness at pavement-bridge interfaces. Although the location of each slope break is unchanged by the filter, the resulting trace seems to indicate that each Figure 18. Profile with grade breaks.

20 positive slope break is followed by a swell and each negative slope break is followed by a dip. The anti-smoothing moving average and the third-order Butterworth filter illustrate a common trade-off in digital fil- tering. Relative to the moving average filter, the Butterworth filter has a sharper roll-off and less pass-band ripple. That is, it removes more of the content longer than the cut-off wave- length and removes less of the content shorter than the cut- off wavelength. However, the Butterworth filter shifts profile features, and the moving average does not (Sayers 1995). The upper trace in Figure 19 was produced by applying four Butterworth filters: (1) first order, forward direction; (2) second order, reverse direction; (3) second order, forward direction; and (4) first order, reverse direction. The cut-off wavelength of these filters was adjusted so that, collectively, their cut-off wavelength was 300 ft (91.44 m). Application of the same filters in both directions cancelled the phase shift. This filter eliminated the artifacts associated with phase distor- tion in the Butterworth filter and with amplitude distortion and boundary effects in the moving average. However, like the other high-pass filters, grade information is removed. Since it includes filtering in the reverse direction, this procedure is not suited for real-time display of profiles as they are collected. 2.3.3 Low-Pass Filtering Low-pass filters reduce or eliminate the content in a signal at frequencies above a particular value and leave content at frequencies below a particular value intact. In the context of road profiling, the cut-off value is usually expressed in terms of wavelength, where the content at wavelengths shorter than the cut-off value are removed. AASHTO M 328-14 requires low-pass filtering with a cut- off wavelength of 0.5 ft (0.15 m) or longer, with undistorted response for wavelengths of 1 ft (0.3 m) and above. This removes some details from profiles without removing con- tent essential for calculating the IRI. Low-pass filtering of road profile sensor signals is an essential step in the collection of valid profiles (Sayers and Karamihas 1998). However, low-pass filtering has two poten- tial drawbacks. First, identifying and locating built-in features such as railroad crossings, bridge joints, textured crosswalks, etc. is much more difficult in filtered profiles because sharp corners, negative spikes at narrow dips, and other features are lost. Second, linear filters retain roughness at deep and narrow dips that do not affect vehicle response because the tire bridges over them. Figure 19. Profiles produced by filtering a profile containing grade breaks.

21 Figure 20 shows a profile with a narrow bridge joint at 28,865 ft (8,798 m). This profile was measured by a profiler with a narrow height sensor footprint and with no low-pass filtering applied to the height sensor signals beyond the filters native to the sensor. The profile includes a down- ward spike more than 2 in (5 cm) deep for a single profile point. Although the local depression around the downward spike would affect vehicle response, the downward spike would not. The IRI algorithm applies a moving average filter with a base length of 9.84 in (250 mm) to represent the envelopment of short-duration asperities by vehicle tires. Figure 21 shows the profile at the bridge joint after application of this moving average. The moving average has a smoothing effect, and it reduces the depth of the spike. However, the filter spreads out the spike and does not eliminate it. Figure 21 also shows the profile after application of a “bridging” version of the moving average filter with the same base length. This filter functions very much like the moving average, but it only con- siders content within the base length that is 0.04 in (1 mm) or less beneath a baseline established from the top of the profile. Karamihas (2005) describes this filter in detail. Figure 22 compares the short-interval roughness profile that results from application of the bridging filter to that of the moving average. The peak roughness level at the bridge joint is reduced by a third when bridging is applied. This may be a more accurate representation of the probable effect of the bridge joint on ride quality. 2.3.4 Spikes The analysis included a search for narrow spikes in the profiles. It was anticipated that metal surfaces could cause incorrect height sensor readings if their surface reflectivity was much higher than typical pavement materials. However, very few spikes were detected in the evaluated profiles at metal surfaces that could be attributed to changes in surface reflectivity. Several spikes appeared in the profiles at gaps in the sur- face at drainage inlets, utility cover boundaries, metal grates, and bridge joints. Typically, gaps caused downward spikes. In some cases, upward spikes appeared at gaps in the pave- ment surface. Figure 23 shows a detailed view of the right elevation profile at the finger joint shown in Figure 24. The upward change in elevation surrounding the spike is associ- ated with the profile of the metal joint. The trough in the center is caused by a gap in the surface. The upward spike at 22,579 ft (6,882 m) is an erroneous reading. Figure 20. Bridge joint profile. Figure 21. Bridge joint profiles with smoothing.

22 The bridging filter discussed above would not remove the influence of the upward spike. The most effective way to iden- tify and eliminate this type of error would require automated inspection of the height sensor signal for anomalous readings or implementation of a sample and hold strategy. 2.3.5 Operational Difficulties The photo logs provided by the Pennsylvania DOT and New Jersey DOT revealed many operational difficulties expe- rienced by profiler operators. Safe and legal operation of the profiler host vehicles in an urban environment required operation at a low speed, the need to slow or stop for traffic, and frequent stops at traffic signals. Many of these observa- tions influenced the selection of test conditions used in the experiment described in Chapter 3. The profiler operators encountered several other hin- drances in addition to traffic flow and stops. The photo logs showed that the profiler drivers needed to change lanes or straddle two lanes in many instances to safely pilot around bicycles, pedestrians, road debris, illegally parked vehicles, and vehicles backing onto the road. These hindrances, as well as debris in the measured lane, were more frequent in congested urban areas. Automated identification of operational difficulties without the use of images or provi- sions for operators to mark areas where operational prob- lems hindered valid operation may help eliminate invalid measurements. 2.4 Feature Identification Section 2.2 and Appendix A provide several examples of localized roughness at built-in road features that were identi- fied using right-of-way images. Successful interpretation of road profile data for management of urban roadways will require identification of the source of roughness, and a means to distinguish localized roughness caused by pavement dete- rioration from localized roughness at built-in features. How- ever, identifying built-in features using a combination of profile viewing and analysis and manual review of photo logs are not practical for network applications. Instead, auto- mated identification of built-in road features using geospatial data will be required. Figures 25–29 illustrate pinpointing of geospatially located roadway features within a road profile using Global Positioning System (GPS) data. Figure 25 shows the left elevation profile and the left rough ness profile for the outside lane of E. Huron Street Figure 22. Roughness profile at a bridge joint. Figure 23. Profile at a finger joint with an upward spike.

23 eastbound in Ann Arbor through the S. State Street intersection. The profile includes three discernable sources of roughness: (1) crowning of the cross street, (2) two utility covers, and (3) two potholes. Figure 25 shows that the highest level of roughness is in the vicinity of the potholes. However, the crowning and the utility covers contribute to roughness, and the roughness profile reaches a peak value near 350 in/mi (5.5 m/km) in the area preceding the potholes. Successful interpretation of the functional performance of this road section and structural performance of the pavement requires identification of the causes of roughness. Figure 26 shows an aerial view of the intersection, and Fig- ure 27 shows an image of the two utility covers. An oval shows the area with the utility covers in the aerial view. Figures 28 and 29 associate the profile measurement with geospatial Figure 24. Image at a finger joint with an upward spike. Figure 25. Elevation and roughness profiles, E. Huron and S. State. Figure 26. Intersection with utility covers. locations of the intersection and the utility covers. Figure 28 shows the centerlines of the intersecting roadways, the two utility cover locations, and the track followed by the profiler over the left wheel path. Figure 29 shows a close-up of the pro- filer path over the utility covers. The profiler tracking location was identified using GPS readings collected simultaneously

24 with profile. This level of accuracy required real-time kine- matic (RTK) correction of the GPS readings. Automated identification of built-in features will require access to geospatially located data from several public and private sources. An investigation of the complexity of acquir- ing spatial information about all the features necessary to support automated identification on selected routes in south- eastern Michigan revealed some difficulties. Figure 30 shows how features are stratified over agency, owner, and the vari- ety of entities that are involved developing an inventory of features of interest. Some of the data inventories containing elements listed in Figure 30 were publicly available, such as the Michigan Geographic Framework Program. It includes roads, railroads, and pipelines. At minimum, the data provide the means to locate intersections and railway crossings. Many of the inventories depicted in Figure 30 were not available for public use. This was demonstrated in an attempt to secure information about built-in features from four local government agencies and the regional gas and electric sup- plier. All of these agencies required the execution of a Non- Disclosure, Data Use Agreement (NDA/DUA). In general, these agreements stipulate that the feature data is confidential and will not be disclosed to any other person or agency or employed beyond the specific use listed in the agreement. For example, the City of Ann Arbor NDA cites that in Michigan the data are deemed classified and are exempt or restricted under the Michigan Freedom of Information Act relative to federal bioterrorism and homeland security laws. The regional utility company refused to provide any data pertain- ing to hardware in the roadway in any form. To overcome these barriers, pavement network data collection may have to be augmented to include identification of built-in features that affect roughness. Figure 27. Utility covers close-up. Figure 28. E. Huron and S. State intersection, plan view. Figure 29. Profile tracking location over utility covers.

25 Figure 30. Agencies with inventories that include built-in road features.