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.

26 This chapter presents the results from an experiment that was conducted to demonstrate the effects of potentially adverse operational conditions on the measurement of longitudinal road profile by high-speed inertial profilers. The experiment included staged reproductions of common operational con- ditions encountered while conducting network-level profile measurements on urban and low-speed roadways, including operation at low speeds, acceleration and deceleration, stop- and-go operation, profiling from a dead stop, and operation on a curve. The experiment included 30 test conditions com- posed of repeated runs under ideal conditions and multiple iterations of those listed above. All of these runs were per- formed on the low-volume loop at the MnROAD test track near Albertville, Minnesota. The experiment demonstrated the effect on profile of errors in the measurement of the inertial reference caused by (1) accelerometer tilting and changes in accelerometer tilt during longitudinal acceleration, longitudinal deceleration, or lateral acceleration of the host vehicle; (2) limitations in accelerometer resolution, and (3) limitations in real-time and post-processing filtering procedures. Six commercially available high-speed inertial profilers participated in the experiment. Together, these six manufac- turers build most of the equipment in use for measurement of longitudinal profile for network-level pavement manage- ment in the United States. The reported results do not characterize the inherent qual- ity of each profiler manufacturer’s products, or the expected performance of every profiler of a given make. Indeed, the accuracy of the profile measured in a given pass over a pave- ment section is heavily influenced by the quality of the pro- filer’s sensors and electrical components, the manner in which they are mounted to the host vehicle, and the profile compu- tation algorithm. However, accuracy also depends on proper- ties of the host vehicle and the manner in which the profiler is operated. In particular, the sensitivity to the staged distur- bances in this experiment depended heavily on the dynamics of the profiler host vehicle (center of gravity height, suspen- sion compliance and kinematics, etc.), and the result from individual passes depended on the lateral placement of the vehicle within the lane. 3.1 Background A previous NCHRP study explored measurement errors associated with speed of operation (Karamihas et al. 1999). The report for the previous study discussed three sources of measurement error: (1) accelerometer misalignment during deceleration, (2) travel over areas of localized roughness that resulted in sensors reaching the limits of their range, and (3) errors in longitudinal distance measurement caused by changes in longitudinal tire slip during braking and accelera- tion. The remainder of this sub-section describes some of the findings from that study. Road profile measurement quality degrades when inertial profilers operate at reduced speeds, change speeds rapidly, or come to a stop. An inertial profiler uses an accelerometer mounted above each height sensor to monitor its vertical movement and establish an inertial reference. Above some limit wavelength, the amplitude of the accelerometer signal is so low that it is masked by sensor noise. The wavelength limit gets shorter at lower speeds (Sayers 1986), and below a specific speed a portion of the waveband that affects rough- ness measurements like the IRI is affected. A further problem is caused by misalignment of the accel- erometer during longitudinal acceleration of the host vehicle. This occurs during braking or heavy acceleration. Figure 31 shows a conceptual schematic of an accelerometer from the side view during deceleration of the host vehicle. The vertical acceleration measured by the transducer is: ( ) ( ) ( )= + θ − + θcos (2)A A g g A sinmeas z x where g is acceleration due to gravity, θ is the vehicle body pitch angle, Az is the vertical acceleration of the point on the C H A P T E R 3 Evaluation of Existing Inertial Profilers for Use on Urban and Low-Speed Roadways

27 vehicle body where the accelerometer is mounted, and Ax is the rearward acceleration. In perfect operation, the pitch angle (θ) is zero, and the total measurement is equal to Az. During typical braking, θ is small, and the sum of the first two terms in the equation above is near Az. However, the third term is roughly proportional to the square of the longitudinal acceleration, since sin(θ) grows in proportion Ax for typical braking events. NCHRP Report 434 provides examples of this for various levels of braking (Karamihas et al. 1999). Profiling through a dead stop contaminates the acceler- ometer readings, because it includes a combination of decel- eration, a complete stop, acceleration, and operating below the valid speed range of the profiler. While the profiler is moving very slowly or at rest, the accelerometer may mea- sure a small bias due to misalignment from true vertical. A slight bias in the accelerometer signal grows in proportion to the square of time when it is integrated twice to get vertical position (Huft 1984) and may grow into a very large overall change in elevation. Starodub (2003) studied the effect of accelerometer per- formance on road profiles collected at various speeds and on roads with significant horizontal curvature. The research concluded that proper specification of the accelerometer and careful mounting could improve measurement quality in challenging environments (e.g., low-speed, high verti- cal curvature), but measurement of acceleration along axes other than vertical could lead to further improvement in data quality. Subsequently, Gagarin et al. (2011) explored geo metric conditions that may cause profile measurement errors at various speeds. The report described mechanisms (also described previously in this section) related to acceler- ometer alignment that confound longitudinal profile mea- surement, but did not list specific combinations of road conditions and measurements speeds to avoid. Walker and Becker (2006) provide examples of profile measurement errors during stop-and-go operation. For the combination of roads and profilers used in the study, the lowest valid speed of operation was found to be 12 mi/hr (19.3 km/hr). Walker successfully demonstrated a method of distorting the relationship between time and distance during very low-speed operation (and stops) to prevent a sudden change in elevation from appearing in the profile. Measurement of profile in urban environments may also challenge the performance of height sensors. The needed range for height sensors depends on the amount of move- ment from the nominal position that is expected in typical operation. The amount of movement of the height sensor depends on the roughness of the road, the speed of operation, the profiler mounting position, and the dynamic response of the vehicle. Figure 32 shows a set of profile measurements Figure 31. Accelerometer tilting during braking (Sayers and Karamihas 1998). Figure 32. Profile measurement exceeding height sensor range (Karamihas et al. 1999).

28 made at various speeds over a rough patch on an urban street. Profiler A measures the profile at 40 mi/hr (64 km/hr) in this area with no obvious errors. Profiler B measures the profile at 20 mi/hr (32 km/hr) and produced a similar trace. How- ever, at 35 mi/hr (56 km/hr), profiler B produced a trace with an artificial dip. This occurred where the profiler moved downward, and drew so close to the road surface that the height sensor exceeded its range. In that area of the profile, the height sensor does not adequately compensate for vertical motion of the profiler host vehicle. Longitudinal distance measurement by inertial profilers may also be compromised during urban and low-speed oper- ation because most profilers rely on measurement of wheel rotation to infer travel distance. First, when the measure- ment wheel on a profiler encounters sharp obstacles, such as potholes and other pavement distress, they exhibit a high level of variation in longitudinal “slip,” and the relationship between wheel rotation and forward vehicle motion is not consistent. Second, the relationship between wheel rotation and forward motion depends on speed, even for a free-rolling tire (Schuring et al. 1974). Lastly, the common wheel speed measurement system used in automobiles is based on tooth- wheel inductive pick-ups. Their output is not reliable and becomes erratic at speeds below 5 mi/hr (8 km/hr), so an operation that includes profiler stop will introduce potential errors in distance measurement. 3.2 Field Experiment 3.2.1 Test Sections The testing took place on the low-volume loop at the MnROAD research facility near Albertville, Minnesota, and at a pavement segment on a nearby urban road. At MnROAD, a tangent section and a section located on a curve were established for testing. Most of the testing took place on the tangent section. The tangent section was 1,502 ft (458 m) long. This length was needed to observe the effects on measure- ment of long-wavelength content during operation at very low speeds and long transient effects that occurred during acceleration and deceleration, as well as the residual effects of high-pass filter settling behavior. Most of the test section was surfaced with asphalt concrete (AC), but it included a transition to PCC about 110 ft (33.5 m) upstream of the section end. The curved section was a jointed PCC pavement with a joint spacing of 20 ft (6.1 m). It was 2,100 ft (640 m) long and included a 75-degree change in direction to the right on a 200-ft (61-m) radius curve, followed by a short transition to a 270-ft (82-m) radius loop running counterclockwise. The start and end locations of the test section were marked with temporary reflective tape. Diamond-shaped marks were painted 18 in (45.7 cm) to the right of the left wheel path of interest to help drivers maintain the lateral position of the profiler accurately and consistently on both test sections. On the tangent section, cones were placed at the right lane edge at several landmarks along the section as guides for staging specialized speed profiles. Depending on the desired speed profile, the cones were placed at loca- tions where the driver was expected to change their speed or come to a stop. Each profiler collected data on Pine Street in Monticello heading north from 6th Street to 4th Street in the outer lane. This segment had several drainage inlets along the curb, abutting driveways, and a railroad crossing. 3.2.2 Profilers Table 1 lists the mounting location of profiler sensors on its host vehicle, the height sensor type, and the lateral sensor spacing for each profiler. In the discussion of results, the term “raw profiles” refers to the profiles as they were submitted for analysis. The effects of various speed and acceleration conditions on the raw pro- files depended on the type of high-pass filtering applied on the data that were submitted. Table 2 lists the type of high- pass filtering applied by each profiler. Profiler 4 applied a cascaded filter that applied the same sub-filters running in each direction to cancel phase shift. Profiler 6 produced three sets of data, including sets with two different cut-off wavelengths and one set with no high-pass Profiler Mounting Profiler Designation Profiler 1 Profiler 2 Profiler 3 Profiler 4 Profiler 5 Profiler 6 Location front mid-wheelbase front front rear front Height Sensor Type single point line single point single point line line Lateral Sensor Spacing (in) 175.3 175.3 172.7 167.6 167.6 174.0 Table 1. Participating profilers. Profiler Designation Profiler 1 Profiler 2 Profiler 3 Profiler 4 Profiler 5 Profiler 6 Filter Type none none Butterworth, 3rd order Butterworth, 6th order moving average, anti-smoothing cotangent cotangent none Cut-off Wavelength (ft) none none 300 300 229.7 200 300 none Table 2. High-pass filtering.

29 filtering. The data from Profiler 1 and Profiler 2 were submit- ted without high-pass filtering. Note that the profiles sub- mitted “without filtering” might have included procedures for suppressing drift. 3.2.3 Reference Measurements Reference profile measurements were collected on both wheel paths on the curved and tangent sections using a SurPRO 3500. The distance between the wheel paths was 69 in (175.3 cm). The overall IRI of the tangent section was 112 in/mi (1.76 m/km) in the left wheel path and 111 in/mi (1.75 m/km) in the right wheel path. On the curved section, the measurements indicated that the IRI of the left wheel path was 163 in/mi (2.57 m/km) and the IRI of the right wheel path was 158 in/mi (2.50 m/km). Reference measurements and measurements by the inertial profilers were performed in the afternoon on the curved section. Visual inspection of the profile plots showed that the PCC slabs within the section were curled downward during all of the testing. However, the roughness exhibited changes during the experiment with changes in environmental conditions. 3.2.4 Speed/Location Measurement A custom-built GPS data logging system was mounted on each profiler to collect independent measurements of the profiler’s speed and position during all of the runs performed at MnROAD. The system included corrections provided by the Minnesota DOT’s network of continuously operating ref- erence stations (CORS) via cell modem. The system output a record of position and velocity at 20 Hz. When conducting measurements, the GPS antenna and cell antenna were fixed to a high point on the vehicle (e.g., the roof) with magnetic mounts. Figure 33 shows an example of the set up. The technician who observed the testing recorded the posi- tion of the GPS antenna on each host vehicle relative to the projection of the profiler’s laser footprint onto the ground. This, in conjunction with the lateral sensor spacing, provided a way to transform the raw measurements of antenna posi- tion to positions of the tracks followed by the left- and right- side profiler sensors. 3.2.5 Test Conditions The experiment included the following test conditions on the tangent section: • Constant Speed: Pass over the section at various constant speeds using cruise control if possible. • Coasting: Enter the section traveling at a constant speed. Disengage the throttle or cruise control at a designated land- mark and coast to the end of the section. • Braking: Enter the section traveling at a constant speed. Apply the brakes at a designated landmark, decelerate at a targeted level to a targeted final speed, and continue at the final speed to the end of the section. • Throttling: Enter the section traveling at a constant speed. Apply the throttle at a designated landmark, accelerate to a targeted final speed, and continue at the final speed to the end of the section. • Stop-and-Go: Enter the section traveling at a constant speed. Apply the brakes at a designated landmark, slow at a targeted deceleration level to a stop as close as possible to a second landmark, remain stopped for a designated length of time, accelerate to the initial speed, and continue to the end of the section. • Dead Stop: Stop at a designated landmark on the section (before initiating profile data collection), initiate data Figure 33. GPS system setup.

30 collection, accelerate to a target final speed, and continue at the final speed to the end of the section. • On the curved section, data collection was performed at several constant speeds using cruise control if it was pos- sible to set the cruise control to the desired speed. Table 3 lists the number of conditions of each type of test, including the designated speed, acceleration levels, decel- eration levels, and stopping intervals where applicable. Each driver was asked to perform three valid passes for each test condition. For the Dead Stop and Throttling tests, “normal” and “heavy” acceleration was driver and vehicle dependent. How- ever, practice runs early in the experiment showed that spe- cific acceleration levels were difficult to achieve accurately. In the Braking and Stop-and-Go tests, guide cones were placed where braking should start and end to achieve the requested change in speed and average deceleration level. In many cases, more than three passes were needed to get an acceptable set of three runs. The GPS data logger described above was moni- tored to provide feedback to the driver in real time or just after a given pass to help make adjustments for the next pass. On the curved section, travel at a forward speed of 20, 30, and 40 mi/hr (32, 48, and 64 km/hr) corresponded to lateral acceleration of about 0.1, 0.22, and 0.4 g, respectively. 3.3 Data Processing 3.3.1 GPS Data The GPS data logging system recorded the time, position, velocity, and diagnostics with real-time corrections at a rate of 20 Hz. The purpose of using the GPS data logging sys- tem was to obtain the path that was traversed by the left and the right sensor of the profiler and to obtain the speed pro- file of the profiler host vehicle. As described in Appendix B, additional processing was applied to data from each run to characterize the speed profile; the lateral tracking error; the location of the onset of braking, forward acceleration, lateral acceleration, or stops; the relative strength of acceleration and deceleration pulses; or the duration of stops. 3.3.2 Profile Data The profile data were processed to examine the following: • Agreement between measurements by the inertial profilers operating at various speeds and reference profile measure- ments, and • Agreement between repeat measurements made under the same conditions, or comparison of profiles measured under ideal conditions (i.e., constant speed) to subsequent passes that include operator-induced disturbances. Analysis of the measured profiles emphasized agreement in profile for calculation of the IRI and the spatial distribu- tion of roughness along a test section. Results are presented in terms of (1) comparison of segment-wide IRI values, (2) comparison of short-interval roughness profiles using a base length of 25 ft (7.62 m), (3) cross correlation of band- pass filtered profiles, and (4) visual comparison of raw and filtered profile traces. Sayers and Karamihas (1996a) and Karamihas and Senn (2009, 2010) demonstrated the use of these methods for profile comparison. 3.4 Results 3.4.1 Transverse Profile Variations For experiments conducted on the tangent section, this section presents results for segment-wide IRI values and cross-correlation analysis in the left wheel path over a range Test Type Number Test Conditions Constant Speed 9 60, 50, 45, 40, 30, 25, 20, 15, and 10 mi/hr Coast 1 initial speed 45 mi/hr Braking 6 braking from 45–20 mi/hr at 0.1, 0.2, and 0.3 g braking from 30–15 mi/hr at 0.1, 0.2 and 0.3 g Throttling 1 normal and heavy acceleration from 20–45 mi/hr Stop-and-Go 4 braking at 0.1 g to stop from 30 mi/hr; stop for 5 sec; accelerate to 30 mi/hr braking at 0.2 g to stop from 30 mi/hr; stop for 5 sec; accelerate to 30 mi/hr braking at 0.2 g to stop from 45 mi/hr; stop for 5 sec; accelerate to 45 mi/hr braking at 0.2 g to stop from 45 mi/hr; stop for 1 sec; accelerate to 45 mi/hr Dead Stop 2 normal and heavy acceleration to 45 mi/hr Operation on a Curve 3 20, 30, and 40 mi/hr 1 mi/hr = 1.609 km/hr Table 3. Test conditions.

31 running from 262.5–1378.0 ft (80–420 m) from the start of the section and in the right wheel path over a range running from 65.6–1378.0 ft (20–420 m) from the start of the section. In the left wheel path, the segment from 262.5–1378.0 ft (80–420 m) was relatively free of transversely inconsistent pavement features, and modest lateral tracking errors did not affect the results. Outside of these limits, “hit or miss” fea- tures appeared inconsistently in the measured profiles, such as filled core holes and transverse cracks with longitudinal wander. In the right wheel path, the tangent section grew progres- sively rough between the right wheel path and the right lane edge, particularly at two locations. The first location had four narrow dips up to 0.4 in (1 cm) deep that were located between 459 ft and 525 ft (140 m and 160 m) from the start of the section. The second location had four narrow dips up to 0.3 in (0.75 cm) deep that were located between 968 ft and 1,033 ft (295 m and 315 m) from the start of the section. The increase in roughness to the right of the right wheel path affected the observed accuracy and consistency in profile measurement because each of the inertial profilers tracked to the right of the designated path. GPS data show a typical bias of 4–12 in (100–300 mm) rightward of the wheel path of interest. This may have been caused by the placement of the diamond- shaped guide marks, which was rightward of the center of the driver’s seat when the profilers were properly positioned. The GPS records showed that the IRI within the two ranges described above increased with increasing “rightward” track- ing error. Figures 34 and 35 provide an example. The figures show the short-interval roughness profiles over the ranges that include the dips. The figures include the following three passes by Profiler 6, which were collected as part of the constant- speed experiment: (1) the second pass at 30 mi/hr (48 km/hr), (2) the third pass at 30 mi/hr (48 km/hr), and (3) the third pass at 40 mi/hr (64 km/hr). In Figure 34, the roughness profiles from all three passes by Profiler 6 agree very well with each other and the reference profile outside of the range with the dips. The profiler tracked over a path to the right of the reference device, and measured higher roughness within the range with the dips in all three passes. The second pass at 30 mi/hr (48 km/hr) and the third pass at 40 mi/hr (64 km/hr), which tracked approximately 4.3 in (11 cm) rightward of the intended location, agree very well. However, the third pass at 30 mi/hr (48 km/hr), which tracked approximately 9.4 in (24 cm) right of the intended location, measured much more roughness in this area. Figure 34. Roughness profiles of transversely inconsistent localized roughness. Figure 35. Roughness profiles of transversely inconsistent localized roughness.

32 Figure 35 shows the roughness profile over the second transversely inconsistent area. Here, the third pass at 30 mi/hr (48 km/hr) and the third pass at 40 mi/hr (64 km/hr), which tracked approximately 9.1 in (23 cm) right of the intended location, agree very well. The second pass at 30 mi/hr (48 km/hr), which tracked approximately 7.1 in (18 cm) right of the intended location, measured less roughness in this area and was in closer agreement with the reference profile. 3.4.2 Constant Speed Operation This set of runs was intended to discern the valid speed range of each profiler for constant-speed operation by observ- ing the change in IRI and change in long-wavelength profile content with travel speed. Speed profiles measured by the GPS data logging system were inspected to verify the average travel speed for each pass over the test section and to confirm the absence of longitudinal acceleration. 3.4.2.1 IRI Agreement Figure 36 shows the percentage disagreement in overall IRI over the segments of interest described in the previous sec- tion between each profiler and the reference measurement at various measurement speeds. In three passes, the profile data from the reference device for the segment of interest produced IRI values of 117.8, 119.3, and 117.5 in/mi (1.860, 1.884, and 1.854 m/km) in the left wheel path and 111.5, Figure 36. Agreement in IRI between inertial profilers and the reference values versus travel speed.

33 110.6, and 112.0 in/mi (1.761, 1.746, and 1.767 m/km) in the right wheel path. Cross correlation of IRI-filtered profiles produced an average agreement score among the possible combination of the three passes of 0.990 for the left profiles and 0.987 for the right profiles. Inspection of filtered profile traces, short-interval rough- ness profiles, and slope spectral density plots revealed the following: • Profiler 2: Inconsistency in measurement of longitudinal distance caused variation in IRI values calculated for Pro- filer 2. For example, in the profiles measured at 15 mi/hr (24 km/hr), the start and end tape for the first, second, and third passes appeared 1513, 1512, and 1479 ft (461.0, 460.8, and 450.8 m) apart, respectively. The “compressed” profile from the third pass registered the highest IRI value. With the exception of the third pass at 15 mi/hr (24 km/hr), Profiler 2 measured the section with increasingly smaller overall length as speed increased. This phenomenon was observed in other profilers. However, some of the other profilers re-calibrated the longitudinal distance measure- ment system frequently or used the section endpoints to enforce a consistent calibration in every pass. • Profiler 3: Measurements on both sides of the lane by Profiler 3 at 10 mi/hr (16 km/hr) produced profiles with roughness that was uniformly reduced across the spectrum compared to measurements at other speeds. The reason for the reduction at 10 mi/hr (16 km/hr) is unclear. • Profiler 4: Some roughness variation measured by Profiler 4 was isolated near a wavelength of 6.2 ft (1.9 m), but it was highest at 10 mi/hr (16 km/hr) and 60 mi/hr (97 km/hr). • Profiler 5: Much of the change in IRI with speed by Pro- filer 5 occurred in a range of wavelengths of 10 ft (3 m) and below. Further investigation using raw sensor signals revealed that a signal timing issue that has since been addressed caused the speed-dependency for IRI values. • Profiler 6: Profiler 6 did not measure the left wheel path, and results are presented for the right wheel path. The unit tracked rightward of the wheel path measured by the reference device. The upward bias in IRI corresponds to the rightward bias in lateral position, and some of the scatter is caused by changes in tracking position. 3.4.2.2 Long-Wavelength Content Inspection of the raw profile traces and profile traces with additional filtering applied showed that host vehicle travel speed affected the long-wavelength content, particularly at low speeds [i.e., speeds below 25 mi/hr (40 km/hr)]. For example, Figure 37 shows a portion of three raw profile traces from the left wheel path produced by Profiler 4 while traveling at 30 mi/hr (48 km/hr), and Figure 38 shows three raw profile traces over the same area produced by Profiler 4 at 10 mi/hr (16 km/hr). The profiles measured at 30 mi/hr (48 km/hr) are much more consistent than those measured at 10 mi/hr (16 km/hr). The inconsistency among the measurements conducted at speeds below 25 mi/hr (40 km/hr) may have been caused by changes in the application of the throttle by the driver. Application and release of the throttle under manual control causes a change in pitch orientation of profilers as they travel over a test section, which tilts the sensitive axis of the accel- erometer. This in turn introduces low-frequency errors into the accelerometer signal due to the projection of a compo- nent of horizontal acceleration onto the sensitive axis of the accelerometer. The reduction in vertical acceleration experi- enced by the profiler and the resulting reduction in the valid portion of the accelerometer signal relative to electrical noise and measurement resolution may have also contributed to measurement error at low speed. Differences in high-pass filtering applied by each profiler strongly affected what could be observed about the long-wavelength content. Table 4 presents cross-correlation values for comparison of slope profiles in the long wavelength range measured by Profiler 4 in the left wheel path. Long wavelength content was isolated using a high-pass filter with a cut-off wavelength of 220 ft (67 m) and a low-pass filter with a cut-off wavelength of 26.2 ft (8 m). This table presents the average agreement Figure 37. Raw profile traces measured by Profiler 4 at 30 mi/hr (48 km/hr).

34 score for comparison of passes at each combination of speeds. Diagonal entries provide the average agreement score for comparisons of repeated runs at the same speed. The other entries provide the average agreement score for comparisons of repeated runs at different speeds. Agreement scores for repeatability at a given speed (i.e., diagonal entries) are greater for measurements above 20 mi/hr (32 km/hr) than for measurements below 20 mi/hr (32 km/hr). Reproducibility is highest for comparisons among the higher speeds, and lowest for comparisons among lower speeds. These observations held in nearly every case for the other five profilers (see Appendix B). 3.4.3 Coasting These tests were intended to quantify the sensitivity of iner- tial profilers to the onset of host vehicle deceleration during a transition from operation at a constant speed to coasting. Table 5 summarizes the passes performed by each pro- filer. In each run, the driver initiated coasting by disengaging cruise control while traveling at speeds of 43.7–46.2 mi/hr (70.3–74.4 km/hr). With one exception, coasting began 726.0–801.3 ft (221.3–244.2 m) from the start of the section. Each profiler host vehicle exhibited a unique level of decel- eration resulting from an upward grade of about 0.5 percent on the test section; host vehicle rolling losses (aerodynamic drag, tire rolling losses, drivetrain and engine losses, etc.); and environmental conditions (wind speed and direction, ambient air pressure, etc.). Figure 39 compares IRI values from passes with coast- ing to those measured at constant speeds from 35–45 mi/hr (56–72 km/hr). These values were calculated using the pro- files from the left wheel path of the tangent section from 262.5–1378.0 ft (80–420 m) from the starting point. Inspec- tion of filtered elevation profile plots and roughness pro- files confirmed that coasting most likely did not cause the variation in measured roughness. Comparison of profiles measured with coasting to those measured at constant speed showed no discernable change in profile at the location of onset of coasting. Short-interval roughness profile plots from passes with coasting included no additional localized roughness compared to passes at constant but comparable speeds. 3.4.4 Braking These runs were intended to quantify the effects of host vehicle braking on the measured profile. Passes with staged braking events were performed for deceleration from 45 mi/hr (72 km/hr) to 20 mi/hr (32 km/hr) and from 30 mi/hr (48 km/hr) to 15 mi/hr (24 km/hr) with target average deceleration values of 0.1, 0.2, and 0.3 g. This sec- tion examines the difference in measured profile and rough- ness of runs with braking to those at constant speed. Figure 38. Raw profile traces measured by Profiler 4 at 10 mi/hr (16 km/hr). 10 mph 15 mph 20 mph 25 mph 30 mph 40 mph 45 mph 50 mph 60 mph 60 mph 0.963 0.971 0.983 0.983 0.976 0.972 0.974 0.974 0.990 50 mph 0.942 0.954 0.976 0.984 0.986 0.987 0.992 0.992 45 mph 0.942 0.955 0.978 0.986 0.989 0.990 0.992 40 mph 0.941 0.955 0.980 0.985 0.992 0.991 30 mph 0.946 0.961 0.984 0.991 0.996 25 mph 0.953 0.966 0.988 0.991 20 mph 0.954 0.969 0.986 15 mph 0.958 0.964 10 mph 0.948 Table 4. Agreement scores, long waveband, Profiler 4, left wheel path.

35 3.4.4.1 Characterization of Speed Profiles The strength of the disturbance caused by braking was quantified for each run using speed profiles derived from data recorded by the GPS data logging system. Figure 40 shows a sample speed profile that was recorded during a pass by Profiler 4 with a target deceleration of 0.2 g from an initial speed of 30 mi/hr (48 km/hr) to 15 mi/hr (24 km/hr). Like the trace in Figure 40, each pass with braking included four regions, which correspond to four successive modes of opera- tion: (1) entry at a constant speed, (2) braking, (3) coasting (i.e., during the transition from using the brakes to using the accelerator), and (4) travel to the end of the section with manual control of the accelerator. The speed profiles provided the limits of each range shown in Figure 40 (i.e., the longitudinal position of each transition), and the high and low speed during each mode of operation. Post-processing of data collected by the GPS data logging system also provided the peak acceleration (or deceleration) averaged over a 1-second interval. In a few of the passes, the driver began to gently accelerate within the last 328 ft (100 m) of the test section. Passes were rejected if acceleration above 0.1 g was recorded over any 1-second interval within the 492 ft (150 m) of travel past the end of the prescribed braking event. In this discussion, passes are identified primarily by two characteristics: (1) the average deceleration during the braking event, and (2) the highest deceleration observed during any 1-second interval within the braking event (i.e., a “peak” value). Figure 41 shows the longitudinal acceleration profile for the same pass presented in Figure 40. The accel- eration trace is displayed after smoothing using a 1-second moving average. This braking pulse represents a case with relatively uniform deceleration. For this pass, the average deceleration during braking was −0.21 g, and the peak decel- eration was −0.23 g. Often, the details of the acceleration profile from two passes with similar peak and average deceleration differed. For example, acceleration was heaviest at the onset of braking in some passes, and heaviest near the end of the braking event in others. In many passes, the driver elected to coast for only a short distance after the end of braking before transitioning to the throttle. Coasting over distances of up to 131 ft (40 m) was common, and some passes included coasting over distances of up to 279 ft (85 m). 3.4.4.2 Effects on Raw Profiles An artificial change in curvature appeared in the unfiltered profiles submitted without high-pass filtering over the range where braking occurred, which was relieved (e.g., changed direction or reduced) in the area after the end of braking. These observations were clearest at deceleration levels of 0.15 g and above. Profiler Passes Initial Speed (mi/hr) Coasting Onset Location (ft) Average Decel. (g) Final Speed (mi/hr) Profiler 1 2 44.7 729.8–730.5 0.037 33.6 Profiler 2 2 45.1 726.0–727.3 0.048 30.2–30.9 Profiler 3 2 44.7–45.3 790.7–801.3 0.046 31.8–32.9 Profiler 3 1 44.7 242.2 0.036 25.1 Profiler 4 3 43.7–44.3 743.0–775.7 0.041–0.042 30.7–32.1 Profiler 5 3 44.9–45.9 743.0–748.3 0.042–0.044 32.3–33.7 Profiler 6 3 45.5–46.2 759.4–799.6 0.036–0.043 34.9–36.2 Table 5. Range of speeds, deceleration, and coasting onset location. Figure 39. Comparison of IRI values in passes with coasting and at constant speed.

36 High-pass filtering modified the spurious content caused by braking. Figures 42 through 45 compare profiles from a pass with braking to three passes at constant speed by the four profilers that applied high-pass filtering (Profilers 3, 4, 5, and 6). The plots show three profiles collected at a constant speed of 30 mi/hr (48 km/hr) compared to a pass by the same profiler with braking from 30 mi/hr (48 km/hr) to 15 mi/hr (24 km/hr) with an average deceleration of 0.28 g and a peak deceleration in the range from 0.32–0.35 g. Each plot denotes the locations of the start and end of braking. The high-pass filtering applied by all four profilers spread out the effect of braking on the profiles outside of the areas where the brakes were applied. In each case, the braking occurred over a distance of approximately 82 ft (25 m), but the effect on the elevation profile appeared over a distance of at least 427 ft (130 m) that started before the location where braking was applied and ended at a location past the release of the brake. For three profilers (Profilers 4, 5, and 6) braking affected the profile ahead of the start of braking and after the end. For Profiler 3, most of the effect on the profile appeared in the range past the location where braking ended. Each profiler applied a high-pass filter of a different type, and each filter spread out the effect of the braking pulse dif- ferently. The filtering applied by Profiler 4 and Profiler 5 spread out the effects of the braking symmetrically. Profiler 3 and Profiler 6 used recursive filters applied in the forward direction, and they shifted the influence of the braking pulses with a forward bias. Note that Figures 42 through 45 show the traces from the wheel path where the vertical range of the artificial content caused by braking was greatest, and very little artificial content was present on the right wheel path profiles from Profiler 3. 3.4.4.3 Effects on Short-Interval Roughness Profiles This section examines the effects of braking on IRI using short-interval roughness profiles. Figure 46 shows profiles measured by Profiler 4 with and without braking after appli- cation of additional high-pass filtering to eliminate content at wavelengths above 100 ft (30.48 m). This additional filter- ing helps to emphasize the content that affects the IRI. The figure compares the profile from a pass at a constant speed of 15 mi/hr (24 km/hr) to a pass with braking from 30 mi/hr (48 km/hr) to 15 mi/hr (24 km/hr) with average deceleration of 0.32 g and a peak deceleration of 0.36 g. With some of the long wavelength content removed, the change in profile is more localized near the range where braking occurred. Figure 47 compares short-interval roughness profiles for runs by Profiler 4 with braking of various deceleration levels where speed decreased from 30 mi/hr (48 km/hr) to 15 mi/hr (24 km/hr) to a pass at a constant speed of 15 mi/hr (24 km/hr). Figure 40. Speed profile for a pass with braking. Figure 41. Longitudinal acceleration profile for a pass with braking.

37 Figure 42. Raw profile trace with braking at 0.28 g, Profiler 4. Figure 43. Raw profile trace with braking at 0.28 g, Profiler 5. Figure 44. Raw profile trace with braking at 0.28 g, Profiler 6. Figure 45. Raw profile trace with braking at 0.28 g, Profiler 3.

38 The figure shows that the effects of braking on the IRI are localized. The braking with an average deceleration of 0.32 g began at 644.7 ft (196.5 m) and ended at 723.4 ft (220.5 m) (illustrated in Figure 46). Additional roughness appears in the vicinity of the end of braking, but not in the location of the start of braking. The braking terminated at the same location within 8.5 ft (2.6 m) in all three of the passes, with braking shown in Figure 47. In each pass, additional roughness appears at a location centered just downstream of the end of braking, and the amount of roughness grows with the severity of the deceleration. Collectively, the short-interval roughness profiles for passes with braking did not exhibit a systematic relationship to the severity of braking. In many cases, two passes by the same unit with braking of equal or similar severity registered dif- ferently on the short-interval roughness profile. This is due in part to the details of the speed profile and the profiler host vehicle’s response to it. The precise location where the braking started and ended also varied between passes. The erroneous content in the profile caused by braking is super- imposed on the actual profile. Typically, much of that con- tent causes localized increases in slope or curvature of the profile. However, some of the erroneous content cancels out or reduces the severity of existing features within the profile. Short-interval roughness profiles were inspected for every pass performed with braking and compared to runs performed at a constant speed. Four cases were observed: • Additional roughness appeared in a localized area. • The distribution of roughness changed, such that the rough- ness profile was increased over some range and decreased over another with no net change in average roughness. • Roughness decreased in a localized area. • No discernable change appeared in the roughness profile. Table 6 summarizes the observations from all six profilers for the 64 passes with braking from 30 mi/hr (48 km/hr) to 15 mi/hr (24 km/hr). The table sorts passes into groups by peak and average deceleration. For each group, the table pro- vides the total number of passes, the number of passes for which the short-interval roughness profile was deemed to be altered (cases 1, 2, and 3, above), and the number of passes that included a localized increase in the short-interval rough- ness profile of 63 in/mi (1 m/km) or greater. Note that when the influence of an increase in localized roughness of 63 in/mi Figure 46. High-pass filtered traces from passes with and without braking. Figure 47. Short-interval roughness profiles for passes with braking.

39 (1 m/km) is spread out over a 0.1-mi (160.9-m) long section, it increases the overall IRI by 3 in/mi (0.05 m/km). Figure 47 showed the short-interval roughness profile for one pass with an increase of 63 in/mi (1 m/km) and two others with a larger increase. When changes were observed in the short-interval rough- ness profiles, they consistently appeared near the end of brak- ing. In cases with an observed localized roughness increase, the peak change appeared no more than 4.6 ft (1.4 m) in advance of the location where braking ended and no more than 43.3 ft (13.2 m) past the location where braking ended. In most cases with altered roughness, the change in the short-interval rough- ness profile appeared no more than 37.4 ft (11.4 m) in advance of the location where braking ended and no more than 82.3 ft (25.1 m) past the location where braking ended. Tables 7 and 8 summarize the observations for the 75 passes with braking from 45 mi/hr (72 km/hr) to 20 mi/hr (32 km/hr). Table 7 groups the passes by average deceleration and Table 8 groups the passes by peak deceleration. When changes were observed in the short-interval roughness profiles, they con- sistently appeared near the location where braking ended. In cases with an observed localized roughness increase, the peak change appeared no more than 35.4 ft (10.8 m) in advance of the location where braking ended and no more than 50.9 ft (15.5 m) past the location where braking ended. In most cases with altered roughness, the change in the short-interval roughness profile appeared no more than 59.7 ft (18.2 m) in advance of the location where braking ended and no more than 103.3 ft (31.5 m) past the location where braking ended. For inertial profilers that are in current use, it is necessary to mark an area of profile in the vicinity of braking as invalid, depending on the severity of braking. The boundaries of the marked area would depend on the high-pass filtering applied by each profiler and would have to be determined by testing. As shown in Figures 42–45, this range would be quite long for engineering purposes that require valid elevation profile for wavelengths up to 300 ft (91.44 m). Based on the results presented in this section, a smaller area for exclusion can be identified for valid measurement of IRI. 3.4.5 Throttling These runs were intended to quantify the effects of forward host vehicle acceleration on measured profile. Passes with staged events were performed for acceleration from 20 mi/hr (32 km/hr) to 45 mi/hr (72 km/hr). Drivers were asked to use a “normal” level of acceleration in some passes and heavy acceleration in others. Typically, acceleration was highest when the throttle was first applied, and it reduced near the final speed. Figure 48 shows the speed profile for a pass with acceleration by Pro- filer 5. In this pass, the acceleration began at about 0.21 g and reduced to about 0.05 g by the end of the event with an average level of 0.12 g. This example represents a typical distribution of acceleration. The most aggressive acceleration pulse for all runs collected in this experiment began at 0.27 g and averaged 0.17 g. Like the runs with braking, the staged events with accel- eration caused changes to the long-wavelength content in the raw profiles. In addition, the effect of acceleration on the profile spread out over a large area and spread out dif- ferently depending on the type of filtering applied by each profiler. Although the raw profile plots showed the effect of acceleration, comparison of short-interval roughness profiles Number of Passes Deceleration Range (g) Total Altered Localized Peak Average Roughness Roughness Increase 0.09–0.16 0.05–0.13 23 4 1 0.18–0.25 0.16–0.20 16 12 5 0.26–0.43 0.22–0.40 25 23 16 Table 6. Effects on short-interval roughness profiles, braking from 30–15 mi/hr (48–24 km/hr). Number of Passes Average Deceleration (g) Total Altered Roughness Localized Roughness Increase 0.04–0.16 26 5 0 0.17–0.23 26 11 4 0.24–0.35 23 16 12 Table 7. Effects of average deceleration on short-interval roughness profiles. Number of Passes Peak Deceleration (g) Total Altered Roughness Localized Roughness Increase 0.07–0.21 30 6 0 0.22–0.28 21 11 4 0.29–0.48 24 15 12 Table 8. Effects of peak deceleration on short-interval roughness profiles.

40 revealed no effects on the IRI near the onset or termination of acceleration. This indicates that the staged acceleration events did not affect profile content in the range of wavelengths that affect the IRI. 3.4.6 Stop-and-Go Operation These runs were intended to demonstrate errors in profile that occur when the host vehicle comes to a complete stop (e.g., due to stopped traffic, at traffic signals, etc.) and to iden- tify the area of profile contaminated by the stop, examine the effect of deceleration level to the stop, and examine the effect of the time spent during the stop. Passes with staged stops were performed for stops from 45 mi/hr (72 km/hr) with a target deceleration level of 0.2 g and from 30 mi/hr (48 km/hr) with target deceleration values of 0.1 g and 0.2 g. Stops from 45 mi/hr (72 km/hr) were per- formed with a requested stop time of 1 second and 5 seconds, and all stops from 30 mi/hr (48 km/hr) were performed with a requested stop time of 5 seconds. This section compares measured profile and IRI of runs with stops to those at con- stant speed. Figure 49 shows a sample speed profile that was recorded during a pass by Profiler 6 with a target deceleration of 0.2 g from a target initial speed of 30 mi/hr (48 km/hr), a target stop time of 5 seconds, and acceleration back to a target speed of 30 mi/hr (48 km/hr) after the stop. Figure 50 shows the corresponding acceleration profile. Each pass with a stop included five regions, which correspond to five successive modes of operation: (1) entry at a constant speed, (2) braking until the vehicle stops, (3) remaining stopped, (4) accelera- tion from a stop to the target speed, and (5) travel to the end of the section. In some passes, the end of the section was reached before the target speed was achieved. The speed and acceleration profiles provided several sta- tistics to facilitate the analysis of the measured profiles: the locations of the start of braking, the location of the stop, the duration of the stop, the level of deceleration during braking, and the level of acceleration after the stop. The level of accel- eration and deceleration were characterized by the average value and the highest value observed during any 1-second interval (i.e., a “peak” value). For the pass shown in Figures 49 and 50, the braking level averaged −0.19 g with a peak value of −0.24 g, and the acceleration level averaged 0.14 g with a peak value of 0.30 g. The stop occurred 739.2 ft (225.3 m) from the start of the section with a duration of 7.1 seconds. Inspection of profiles and IRI distributions showed that erroneous content at or near the location of the stop accounted for most of the roughness introduced by measurement error during passes with stop-and-go events. This section dem- onstrates the manner in which measurement errors appeared in the profile data and the effect of those errors on the short- interval roughness profiles. Figure 48. Speed profile for a pass with forward acceleration. Figure 49. Speed profile for a pass with a stop.

41 3.4.6.1 Effects on Raw Profiles The time spent without movement during the stop intro- duced artificial content into the profiles beyond the effects for braking and acceleration. The accelerometer is typically not oriented precisely vertically after the profiler host vehicle comes to rest. Misalignment of the accelerometers with the direction of gravity causes a small bias in their readings which, when integrated twice, grows in proportion to the duration of the stop squared. Although there is no change in the longi- tudinal position during the stop, a step change in elevation is introduced at the stop as a result. This section examines the effect on profile of stops from 45 mi/hr (72 km/hr) with dura- tions of approximately 1 second. Profilers 1, 2, and 6 submitted profiles without high-pass filtering. Raw profile traces from all three units included arti- ficial curvature in the vicinity of the stop. Figure 51 shows a profile provided by Profiler 2 without high-pass filtering for a stop at a distance of 716.9 ft (218.5 m) and a duration of 1.2 seconds. The disturbance introduced into the profile by the stop-and-go event includes curvature near the stop and a step change in elevation at the stop. For Profiler 2, the mag- nitude of the step change varied, and was typically larger for cases with longer stop durations. For the pass shown in Fig- ure 51, no step change appeared in the right wheel path profile, and it agreed well with passes collected at a constant speed of 45 mi/hr (72 km/hr). The reason why the disturbance to the profile caused by the stop only occurred in the left wheel path profile is not known. Profiles from Profiler 1 and Profiler 6 included “sample and hold” areas surrounding the location of the stop where the profile elevation was held at a constant value over some range where the travel speed of the profiler was low. (The speed thresholds used by these profilers to enable the “sample and hold” are not known.) Figure 52 shows an example from Pro- filer 1 for a pass with a stop at a distance of 743.4 ft (226.6 m) and duration of 0.9 seconds. A flat area is present in this profile close to 743.4 ft (226.6 m), which corresponds to the “sample and hold” area. Although this practice eliminated the step change in elevation at the stop, the sudden change in slope at the borders of the artificially flat area introduced roughness into the profile that affected the IRI. High-pass filtering modified the spurious content differ- ently depending on the filter type. Figures 53 and 54 show profiles submitted by Profiler 4 and Profiler 5, respectively, that were high-pass filtered. The high-pass filtering applied by these profilers spread out the effect of the stop. For the pass in Figure 53, the stop occurred at a distance of 675.5 ft (205.9 m) and duration of 1.5 seconds. For the pass in Fig- ure 54, the stop occurred at a distance of 753.3 ft (229.6 m) and duration of 4.5 seconds. For both profilers, the high-pass Figure 50. Acceleration profile for a pass with a stop. Figure 51. Elevation profile measured by Profiler 2 with a stop.

42 filtering spread out the effect of the step change in elevation at the stop over the same distance in each direction and in proportion to the filter cut-off or base length. Profiles sub- mitted by Profiler 6 with high-pass filtering spread out the influence of stops in a similar fashion. The software in Profiler 3 included a proprietary “stop- and-go” processing feature, which is meant to reduce the mea- surement error caused by non-emergency stops. Figure 55 compares a profile measured by Profiler 3 at a constant speed of 45 mi/hr (72 km/hr) to a pass with a stop from 45 mi/hr (72 km/hr) at a distance of 733.6 ft (223.6 m) and duration of 1.4 seconds. The profile measured with a stop includes a small artificial disturbance where the stop occurred, but no elevation discontinuity. The stop also affected long- wavelength content in the profile. Profiler 3 applied a third- order Butterworth filter with a cut-off wavelength of 300 ft (91.44 m) as part of the measurement process. This is a recur- sive filter, and it spread out the influence of the stop on the elevation profile downstream. 3.4.6.2 Effects on Short-Interval Roughness Profiles Figure 56 shows the short-interval roughness profiles for the pass by Profiler 2 shown in Figure 51. A narrow area of severe localized roughness appears in the left roughness pro- file in the vicinity of the stop with a peak value of 3,510 in/mi (55.4 m/km) at a distance of 727.2 ft (221.7 m). Most of the artificial roughness is caused by the step change in elevation at the stop, with a rapid reduction in the area downstream. No localized roughness appeared on the right side. The reason why the localized roughness appeared on the left side and not on the right side for this pass is unknown. How- ever, localized roughness was present on the right wheel path profiles for other passes with a stop by Profiler 2. Figure 57 compares the left wheel path short-interval roughness profile from the same pass to a pass collected at a constant speed of 45 mi/hr (72 km/hr). The vertical scale of the plot is truncated to help show the location where the pass with a stop departs from the pass at constant speed in advance of the stop, and where they begin to agree again downstream. The roughness profiles agree up to a distance of 689.0 ft (210 m), which is 27.9 ft (8.5 m) upstream of the stop. With the exception of a longitudinal shift caused by the braking associated with the stop, the traces agree after a dis- tance of 839.9 ft (256 m), which is 123.0 ft (37.5 m) past the stop. Note that the 25-ft (7.62-m) moving average used for plotting the short-interval roughness profile spread out the influence of the stop by an additional 12.5 ft (3.81 m) on each side. Figure 52. Elevation profile measured by Profiler 1 with a stop. Figure 53. Elevation profile measured by Profiler 4 with a stop.

43 Figure 54. Elevation profile measured by Profiler 5 with a stop. Figure 55. Elevation profile measured by Profiler 3 with a stop. Figure 56. Short-interval roughness profiles measured by Profiler 2 with a stop. Figure 57. Short-interval roughness profiles, Profiler 2 with and without a stop.

44 In the 12 passes by Profiler 2 with a stop, localized rough- ness appeared in the short-interval roughness profile with peak values of 3,041–6,273 in/mi (48–99 m/km) on the left side and 0–2,218 in/mi (0–35 m/km) on the right side. On the left side, the range where the short-interval roughness profile was contaminated included up to 46.3 ft (14.1 m) preced- ing the stop and up to 159.9 ft (48.7 m) past the stop for the 12 passes. Table 9 summarizes these observations for passes with stops by each profiler. The passes with stops by each pro- filer included instigation of the stop from two speeds, at two different target deceleration levels, and short-duration and long-duration stops. However, the severity of the localized roughness and the range contaminated by the stops did not relate systematically to those details. As such, the table lists ranges observed for each profiler. The severity of localized roughness was the second lowest for Profiler 1 because of the “sample and hold” procedure of suppressing elevation changes applied at low speed. However, the contaminated range was consistent with the other pro- filers. Profiler 2 exhibited the third lowest levels of localized roughness. It was also noted for Profiler 2 that the peak levels on the right side were much lower than the peak levels on the left side. Overall, the contaminated range for this group of profilers covered a range starting 155.2 ft (47.3 m) before the stop and ending 248.0 ft (75.6 m) past the stop. The error in IRI adjacent to the stop appeared differently in the short-interval roughness profiles from Profiler 3 because of the “stop-and-go” feature, and the severity of local- ized roughness was the lowest. Figure 58 compares a short- interval roughness profile from a pass by Profiler 3 with a stop from 45 mi/hr (72 km/hr) to a pass at a constant speed of 45 mi/hr (72 km/hr). The pass with the stop included brak- ing with an average deceleration of 0.16 g and a stop at a distance of 793.8 ft (242.0 m) with a duration of 2.65 seconds. Among the 13 passes by Profiler 3, this was the case with the greatest level of artificial roughness near the stop. For inertial profilers that are in current use, it is necessary to mark an area of profile in the vicinity of a stop as invalid. The boundaries of the marked area would depend on the high-pass filtering applied by each profiler and would have to be determined by testing. As shown in Figures 51–55, this range would be quite long for engineering purposes that require valid elevation profile for wavelengths up to 300 ft (91.44 m). Based on the results presented in this section, a smaller area for exclusion can be identified for valid measure- ment of IRI. Profiler Side Passes Peak Localized Contaminated Area (ft) Roughness (in/mi) Upstream Downstream Profiler 1 Left 14 570–1,265 −44 77 Right 14 380–635 −44 137 Profiler 2 Left 12 3,040–6,275 −46 160 Right 12 0–2,220 −34 123 Profiler 3 Left 13 0–190 −33 96 Right 13 30–190 −59 93 Profiler 4 Left 13 6,526–41,247 −78 171 Right 13 4,055–37,000 −46 156 Profiler 5 Left 13 26,930–170,945 −152 248 Right 13 10,330–256,610 −155 245 Profiler 6 Right 12 1,900–10,075 −74 153 Table 9. Contamination of profile around stops. Figure 58. Short-interval roughness profiles, Profiler 3, with and without a stop.

45 3.4.7 Operation from a Dead Stop In these tests, the profiler operators were asked to initiate profile data collection while the profiler was stationary, and then accelerate to 45 mi/hr (72 km/hr) and maintain this speed until the end of the section. These runs were intended to examine the combined effect of insufficient speed, accel- eration, and settling of filter transients. The software in many of the profilers did allow data to be recorded until a specific speed was achieved or a specific dis- tance was traveled after data collection was initiated. In some cases, collection from a dead stop was possible using proce- dures within the data collection software that were contrary to the recommended mode of operation. Some of the profilers began recording data from a location at or just downstream of the stop, but the profile measure- ment and computation procedures were running in the back- ground during the approach. Those runs produced results similar to the outcome of the stop-and-go runs, where a large transient appeared in the raw profiles for some distance after the stop, which was caused by the interaction of the high-pass filters with the artificial disturbance measured at the stop. Profiler 4 provided data from a pass where a profile was recorded from an instant just after a dead stop that appears to exclude the influence of profile computation during the stop. Figure 59 shows raw profiles from Profiler 4 for the following three passes: (1) at constant speed of 45 mi/hr (72 km/hr), (2) with a stop at a distance of 699.0 ft (213.1 m), and (3) beginning from a dead stop at 726.7 ft (221.5 m). The passes with stops included similar speed profiles during acceleration to 45 mi/hr (72 km/hr) after the stop. The fig- ure shows that the absence of the step change in elevation at the stop in the profile recorded from a location past the stop greatly reduces the transient superimposed on the profile by the high-pass filter. Figure 60 shows the short-interval roughness profiles for the same three passes. The vertical scale of the plot is trun- cated to illustrate the range where agreement exists between passes. The trace for the stop-and-go run rises to a peak of 12,902 in/mi (203.6 m/km) at 704.6 ft (214.8 m). The trace from the run from a dead stop does not include spurious localized roughness; however, it does not agree as well with the pass at constant speed over the first 100 ft (30 m) after the stop as it does in the area beyond, in part due to insufficient lead-in and in part due to incompatibility in longitudinal dis- tance measurement. The raw profiles provided by Profiler 3 included similar artificial disturbances in long-wavelength content for a distance after the dead stop as those measured through a stop. Figure 59. Elevation profile measured at constant speed, from a dead stop, and through a stop. Figure 60. Roughness profile measured at constant speed, from dead stop, and through a stop.

46 3.4.8 Operation on a Curve These runs were intended to examine errors in profile caused by lateral acceleration and changes in lateral accel- eration on a curve. Passes over a section with a short curve to the right followed by a long curve to the left were con- ducted at 20, 30, and 40 mi/hr (32, 48, and 64 km/hr). At these speeds, the lateral acceleration for the travel over the long curve to the left is approximately 0.1, 0.22, and 0.4 g, respectively. This section compares measured profile and roughness of the runs at different speeds to each other and to reference measurements collected on the right side of the lane. Figure 61 shows a sample acceleration profile that was recorded during a pass by Profiler 1 at 40 mi/hr (64 km/hr). Acceleration leftward is positive. Travel over this section included (1) 191 ft (58 m) of travel along a tangent seg- ment, (2) 415 ft (127 m) of travel over a curve to the right, (3) 1,357 ft (414 m) of travel around the curve to the left, and (4) 146 ft (44 m) of travel on a tangent segment to the end of the section. The speed and acceleration profiles provided several sta- tistics to facilitate the analysis of the measured profiles. These statistics included speed (for verification of the test logs), locations of the three large transitions in lateral accel- eration, and the magnitude of lateral acceleration. In the 55 passes conducted on the curved section, the transition to rightward acceleration occurred in a range from 122–232 ft (37–71 m). The crossover from rightward acceleration to leftward acceleration occurred in a range from 591.8–615.3 ft (180.4–187.6 m), depending on the specific path taken in each pass. For the pass shown in Figure 61, the crossover from rightward acceleration to leftward acceleration occurred at a distance of 605.8 ft (184.7 m). Transition out of left- ward acceleration occurred in the range from 1,917–2,013 ft (584–613 m). The drivers’ tracking behavior affected the location and severity of these transitions. The strength of the rightward and leftward acceleration was characterized by the average acceleration in each direc- tion, and the highest acceleration in each direction observed during any 1-second interval (i.e., “peak” values). For the pass shown in Figure 61, travel over the rightward curve had an average lateral acceleration of −0.31 g with a peak value of −0.44 g, and travel over the leftward curve had an average lateral acceleration of 0.36 g with a peak value of 0.43 g. Inspection of profiles and roughness distribution showed that erroneous content in the raw profiles was heaviest in areas surrounding the transitions in acceleration. However, the erroneous content did not affect the IRI appreciably because it primarily affected the long wavelength content. 3.4.8.1 Effects on Raw Profiles Profilers 1, 2, and 6 submitted profiles without high-pass filtering. The artificial change in curvature appeared in the “unfiltered” profiles at the transitions in lateral acceleration and was heaviest at the crossover from rightward acceleration to leftward acceleration. High-pass filtering modified the spurious content differ- ently depending on the filter type, but all profilers included spurious content surrounding the transitions. Figure 62 shows a typical example. The figure compares profile measurements from Profiler 1 for passes over the curved section at three speeds to the reference measurement. A high-pass filter with a cut-off wavelength of 300 ft (91.44 m) was applied to all four traces for plotting. This was done to display the portion of the waveband typically recorded by high-speed profilers. All of the profilers with high-pass filtering spread the effects of lateral acceleration over a similar area. 3.4.8.2 Effects on Short-Interval Roughness Profiles Figure 63 compares the short-interval roughness pro- files over part of the curved section for the same four passes shown in Figure 62. No spurious localized roughness appears in the profiles from Profiler 1 at the locations near transi- tions in lateral acceleration. The plot shows that Profiler 1 and the reference profiler measure the same overall level of roughness, but the spatial distribution of roughness agrees only modestly, and short-interval roughness profiles were Figure 61. Acceleration profile for a pass over the curved section.

47 not as well repeated at various speeds by Profiler 1 as they were on the tangent section. This was true of all six inertial profilers and was due in part to less consistent lateral track- ing and less consistent measurement of longitudinal distance on the curves. Note that slab curl accounted for a large share of the overall roughness on the curved section. Although all measurements over the curved section occurred in the late afternoon, the level of curl may have changed from day to day. Figure 64 compares the short-interval roughness profiles over a part of the section with relatively constant lateral accel- eration, and no transition in road curvature. Distributions of roughness measured by the reference profiler and Profiler 1 at various speeds agree in this range. 3.4.8.3 Effects on IRI Although the profiles measured on the curved section by the inertial systems included spurious long wavelength con- tent and longitudinal distortion (as described below), the overall IRI values did not include an obvious bias caused by lateral acceleration. Table 10 lists the average right-side IRI values from each unit for three passes over a portion of the curved section at each speed. These values were calculated for a segment from 347.8–875.7 ft (106.0–266.9 m) from the starting point. This is a 0.1-mi (160.9-m) segment surround- ing the transition from the rightward curve to the leftward curve. The three passes by the reference profiler produced an average IRI value of 132.6 in/mi (2.092 m/km) on the right side for this segment. 3.4.9 Longitudinal Distance Measurement All six participating profilers measured longitudinal dis- tance using an encoder attached to a rear tire on the host vehicle, which measures the distance based on the rotation of the tire. This approach requires calibration by traveling over a pavement section with landmarks that are a known distance apart. However, a typical inertial profiler only applies one cal- ibration factor, which corresponds to a specific travel speed, tire inflation pressure, and tire temperature. In addition to the thermal state of the tire, the amount of relative distortion in longitudinal distance measurement during the experiment depended on two competing effects: • The rolling radius of a free-rolling tire increases with rota- tional speed. Without a change in calibration, longitudinal distance measured by observing wheel rotation registers a smaller apparent distance at higher speed for the same amount of travel. • The tire operates with longitudinal slip that increases with longitudinal acceleration. For example, during braking or coasting, the tire experiences a net rearward longitudinal force at the road interface. In this condition the tire con- tact patch elongates relative to the free-rolling condition, Figure 62. Raw profiles measured on the curved section. Figure 63. Short-interval roughness profiles for passes over a curved segment.

48 and the wheel rotates less for the same travel distance. The resulting reduction in tire rotation relative to the free-rolling condition causes it to register a smaller distance for the same amount of travel. Forward acceleration causes a bias in lon- gitudinal distance measurement with the opposite polarity. The reduction in speed during coasting affected longitu- dinal distance measurement. Typically, the longitudinal dis- tribution of roughness in runs with coasting matched those at a constant 45 mi/hr (72 km/hr) speed over the first half of the section, but compressed the profile slightly in the second half of the section with misalignment of up to 6.6 ft (2 m) by the section end. The effect of braking on longitudinal dis- tance measurement is most visible in Figure 46. In this figure, the profiles are aligned before the start of braking, but offset longitudinally after braking ended. A reduced measurement of travel distance of about 2.6 ft (0.8 m) occurred during the time while the brakes were applied. This phenomenon appeared in passes with braking from all six profilers. The magnitude of the error in distance measurement increased with duration and severity of braking. Stop-and-go operation affected longitudinal distance measurement by all six profilers. Comparison of profiles measured with a stop to those measured at constant speed showed that a negative bias in longitudinal distance mea- surement accumulated during braking, and a positive bias accumulated during acceleration after the stop. In most cases the two effects nearly cancelled each other, and the largest distortion in longitudinal distance was confined to the area around the stop. On the curved section, the inertial profilers accumulated longitudinal distance less rapidly compared to passes in the right wheel path by the reference profiler and more rapidly on the leftward curve. That is, the inertial profilers spread out profile features more than the reference profiler while on a curve to the left, and less while on a curve to the right. The correction to longitudinal distance measured by the inertial profilers that would have been required for com- patibility with the right wheel path profiles measured by the reference was different for each profiler. However, the optimal correction was always either less positive or more negative (by 1.7–3.3 percent) for the area with a curve to the left than the area for the curve to the right. For example, Figure 63 shows short-interval roughness profiles from a set of passes with no adjustment to the measurement of longi- tudinal distance, but alignment of the same profiles for the range in Figure 64 required a downward adjustment to the longitudinal distance interval from the reference profiler of 2.4 percent. In contrast, the reference profiler and inertial profilers measured longitudinal distance in the left wheel path much more consistently, or disagreed by the same percentage on the rightward and leftward curves. This suggests that the iner- tial profilers monitored rotation of a wheel on the left side of the vehicle for measurement of longitudinal distance. For applications where the specific location of localized rough- ness is important or location is referenced to the centerline of the road, some correction may be required. As a minimum, recorded landmarks may be needed at segment boundaries for roughness surveys on roads with tight curves. Figure 64. Short-interval roughness profiles at the right-left transition. Average IRI Value (in/mi) Profiler 20 mi/hr 30 mi/hr 40 mi/hr Profiler 1 126.3 120.6 122.0 Profiler 2 122.5 123.1 119.3 Profiler 3 127.2 124.7 128.9 Profiler 4 124.2 124.1 124.2 Profiler 5 118.7 145.9a 132.0 Profiler 6 121.8 120.4 121.7 aRemoval of the third pass produced an average of 131.0 in/mi. Table 10. Average IRI values measured on a transition in horizontal curvature.