Automated Pavement Condition Surveys (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.

5 This chapter presents the findings of the literature review relevant to automated pavement condition surveys. Automated collection of pavement condition has been conducted in the United States for almost 40 years. In 2004, McGhee documented the state of automated pavement condition data collection within the United States. Since that time, tremendous technological advances have occurred and have been adopted for automated pavement condition surveys. Topics covered in this chapter include pavement condition assessment, survey methods, collection equipment, national reporting requirements, quality management plans, and collection of nonpavement assets. Specific focus is given to documents published after 2004. Current technology allows analysis using 2D and 3D representations of the pavement surface, which can be used to automatically identify and quantify distress. Pavement Condition Assessment Pavement condition assessment is essential for defining current network pavement condition, predicting future pavement condition (using performance prediction models), and determining the timing and most appropriate pavement preservation and rehabilitation treatments or strategies. During the pavement condition survey, the distress type, severity, and extent are determined. Typical pavement condition surveys include the following: • Roughness (ride)—Pavement roughness is a measure of how much the roadway surface deviates from a flat line and represents the ride quality of the vehicle (and thereby the user). Roughness affects not only user ride comfort, but also vehicle operating costs (Chatti and Zaabar 2012, Robbins and Tran 2016). Ride quality is an important roadway characteristic for the traveling public. Islam and Buttlar (2012) showed that rougher roads result in higher vehicle operating costs, and smoother roads have longer performance life. Figure 1 illustrates a schematic of the International Roughness Index (IRI) calculation. • Rutting—Rutting is a surface depression in the wheel paths of asphalt pavement (the topmost surface is constructed with asphalt materials) under repeated truck traffic loading. Rutting can occur due to mix-related issues and weaker subgrade soil conditions. Figure 2a illustrates rutting from a pavement cross section, and Figure 2b illustrates rutting at an asphalt pavement intersection. • Faulting—Faulting occurs on jointed plain concrete pavement (JPCP) typically when the leave-side of a slab is higher than the approach-side of the next slab. Faulting is measured because of its profound effect on pavement roughness and ride quality, making it an ideal trigger value for maintenance and rehabilitation activities. Figure 3 illustrates an example of faulted JPCP. • Surface distress—This is a visible defect on the pavement surface. Pavement surface dis- tress is quantified according to distress type, severity, and extent (or quantity). A number C H A P T E R 2 Literature Review

Figure 1. IRI (Sayers and Karamihas 1998). b. Example (image courtesy of Washington State DOT). a. Schematic (Vermont Agency of Transportation 2017). Figure 2. Rut depth. Figure 3. Example of faulting (image courtesy of Washington State DOT).

Literature Review 7 of identification manuals have been developed and include, but are not limited to the following: – Distress Identification Manual for the Long-Term Pavement Performance Program (Miller and Bellinger 2014). – ASTM D6433, Standard Practice for Roads and Parking Lots Pavement Condition Index Sur- veys (2018c). – Highway Performance Monitoring System (HPMS) Field Manual (FHWA 2016). Summary of Previous Syntheses This synthesis represents the fifth National Cooperative Highway Research Program (NCHRP) syntheses conducted in relation to pavement condition assessment over the last four decades. This effort represents not only the interest in pavement condition assessment, but also the significant technology changes that have occurred. Following are some of the major findings of the previous four syntheses. NCHRP Synthesis of Highway Practice 76: Collection and Use of Pavement Condition Data Hicks and Mahoney (1981) conducted a survey of nine agencies (seven SHAs; one MoT; and the U.S. Air Force, Engineering Research and Development Center). The agencies were asked to identify the types of data collected for assessing roadway conditions; included in the survey were ride (eight agencies), surface distress (eight agencies), deflection (one agency), and skid (seven agencies). In relation to ride, three agencies reported using a Mays meter, two agencies reported using a Cox meter, one agency used the Portland Cement Association meter, one agency used a subjective rating, one agency used a “unique mobile vehicle response profiler,” and one agency used a laser profilometer. Surface condition assessment (e.g., cracking, rutting, raveling, patching) was collected regularly by all but one agency, but was usually collected subjectively, and photo logging was not common for the responding agencies. NCHRP Synthesis of Highway Practice 126: Equipment for Obtaining Pavement Condition and Traffic Loading Data Epps and Monismith (1986) conducted an agency survey on pavement condition and traffic load data collection equipment. Thirty-seven of the responding agencies indicated collecting pavement condition data. Eight agencies indicated using a photo logger for assessing surface distress, and two agencies indicated using a profilometer to quantify pavement surface roughness on the network level. The majority of the agencies (36 agencies) indicated the use of ride meters (e.g., Mays meter, Cox & Sons road meter, Soiltest road meter) for assessing pavement surface roughness. During this time, several companies developed (or were developing) video-imaging processes to identify crack counts using digitized equipment. NCHRP Synthesis of Highway Practice 203: Current Practices in Determining Pavement Condition This survey of agency practice indicated that nearly all of the 60 responding agencies collected pavement condition data (Gramling 1994). Although a number of data collection standards had been developed (e.g., Smith et al. 1979, Lytton et al. 1985, Darter et al. 1985, Strategic Highway Research Program 1993), the survey responses showed a lack of standardization of data collection types and methods. Pavement surface condition data collection methods included windshield

8 Automated Pavement Condition Surveys surveys (18 agencies), shoulder surveys (nine agencies), walking surveys (10 agencies), auto- mated surveys using sensors (eight agencies), and a combination of methods (12 agencies). NCHRP Synthesis of Highway Practice 334: Automated Pavement Distress Collection Techniques McGhee (2004) documented the state of automated pavement distress collection techniques. At the time of the synthesis, automated pavement distress collection was just beginning to have widespread use in the United States and Canada. McGhee (2004) discussed agencies’ use of downward-facing camera images to determine pavement surface distresses. These images were typically obtained by photographing, on film or videotape, analog images of the pavement surface, which provided high-quality images, but were not easy to convert into digital format. On the other hand, digital photos could be collected, but there were quality and resolution issues. Agencies characterized pavement roughness using IRI, predominantly measured using laser sensors, although acoustic and infrared sensors were also used. Rut-depth measurements were typically performed using three or five sensors, although several agencies indicated using up to 37 sensors for characterizing transverse profile. With several different data items being collected, a new trend in the industry was initiated for integrating data collection into a single vehicle. Data collection vehicles (DCVs) included equipment for automatically collecting images, profile measurements, linear reference, and a satellite-based Global Positioning System (GPS). The majority of the DCVs were owned and operated by data collection vendors, although some had been sold directly to agencies. Finally, McGhee (2004) noted that many agencies had difficulty achieving high-quality images and trouble keeping up with the rapidly advancing digital image technology. There was no consensus on the appropriate way of measuring rut depth, and many agencies did not have confidence in faulting data and opted instead to use manual methods of data collection. A relatively new technology at the time of NCHRP Synthesis 334 was 3D laser imaging, which was primarily used to determine IRI and rut depth. Table 1 provides a summary of the four NCHRP syntheses completed to date on pavement condition surveys. As stated previously, the number of agencies transitioning from manual to automated methods steadily increased from 1981 to 2004. In 1986, 82% of agencies (36 of 44) indicated using a ride meter to determine roughness, while only 5% (2 of 44) reported using a Condition Number of Responding Agencies NCHRP Synthesis 76 (1981): 9 agencies NCHRP Synthesis 126 (1986): 44 agencies NCHRP Synthesis 203 (1994): 60 agencies NCHRP Synthesis 334 (2004): 56 agencies IRI (or ride) Ride meter, 7 Profilometer, 1 Ride meter, 36 Profilometer, 2 Ride meter, 22 Profilometer, 49 Laser, 44 Acoustic, 3 Infrared, 4 Rutting Included in distress rating or not specified Included in distress rating or not specified Straightedge, 16 Sensor-based, 14 Visual estimate, 9 3 Sensors, 16 5 Sensors, 16 5+ Sensors, 14 Faulting Included in distress rating or not specified Included in distress rating or not specified Included in distress rating or not specified Manual survey, 11 Sensor, 23 Distress Manual survey, 81 Manual survey, 26 Photo, 10 Manual survey, 37 Automated, 8 Combination, 12 Manual survey, 23 Automated, 33 Table 1. Summary of pavement condition assessment technology (Hicks and Mahoney 1981, Epps and Monismith 1986, Gramling 1994, and McGhee 2004).

Literature Review 9 profilometer. All agencies reported collecting rut depth and faulting as part of the distress survey, and 23% of the agencies (10 of 44) indicated using photo equipment for distress identification. By 1991, 82% (49 of 60), 23% (14 of 60), and 0% of agencies reported using sensor-based measurements for determining IRI, rut depth, and faulting, respectively; 13% (8 of 60) reported using automated distress detection, and 20% (12 of 60) reported using a combination of manual and automated distress detection. By 2004, 91% (51 of 56) and 82% (49 of 60) of agencies reported using sensors for measuring IRI and rutting, respectively; 41% (23 of 56) reported using sensors for fault measurements; and 59% (33 of 56) reported using automated methods for quantifying distress. Figure 4 illustrates this information and shows the progression of agencies using automated methods from 1986 to 2004. Data Collection and Analysis Standards Quality pavement condition data are critical elements of pavement management systems, and it is important to have a process to consistently gather data in the same manner. Additionally, established requirements are needed for the data collection equipment and how the collected data is processed. AASHTO recently released several specifications related to automated pave- ment conditions surveys, discussing collection pavement surface images and quantifying cracks from these images. Table 2 summarizes AASHTO and ASTM specifications related to collecting and assessing pavement condition. Pavement Condition Survey Methods There are two primary methods for conducting pavement condition assessment: manual condition surveys and automated condition surveys. Manual Condition Surveys Manual condition surveys are conducted by walking or traveling at a slow speed and noting the existing surface distress. Manual surveys may be limited to selected roadway segments (i.e., samples) or span the entire lane area (i.e., 100% survey). The majority of U.S. highway, 5% 0% 0% 23% 82% 23% 0% 33% 91% 82% 41% 59% 0% 20% 40% 60% 80% 100% IRI Rutting Faulting Distress PE RC EN T O F AG EN CI ES 1986 1991 2004 Figure 4. Progression of agencies using automated methods.

10 Automated Pavement Condition Surveys Specification Description Pr ofi le E qu ip m en t AASHTO M 328 (2018a) Hardware and software requirements for low- and high-speed inertial profilers. Hardware requirements for laser sensors, accelerometers, and distance measuring instrument and software requirements for processing, storing, and reporting data. AASHTO R 57 (2018b) Procedure for operating and verifying calibration of inertial profile systems. Procedures used for QC and assurance for network-level data collection. Not recommended for daily paving operations. AASHTO R 56 (2018c) Procedure to certify equipment for measuring longitudinal surface profile mounted on a host vehicle using an inertial reference system. For measuring profile for construction QC, assurance, and acceptance, and network-level data collection. M ea su ri ng P ro fil e AASHTO R 88 (2018h) Method for collecting transverse profile, using automated measurement devices. Used to quantify cross slope, transverse deformation, rutting, water entrapment, and edge drop-off. Replaced AASHTO PP 70-14 (2017a). ASTM E950 (2018a) Method for measuring and recording the profile with an accelerometer-established inertial reference mounted on a profile measuring vehicle. ASTM E2133 (2013b) Method for measuring longitudinal and transverse profile using a rolling inclinometer at walking speed. ASTM E1656 (2016) Procedure for classifying equipment for measuring longitudinal profile, transverse profile, and pavement surface cracking at or near posted speed. AASHTO R 36 (2017b) Method for quantifying jointed concrete pavement faulting from longitudinal profile measurements. AASHTO R 43 (2017c) Method for estimating roughness according to IRI. ASTM E1926 (2015a) Computing IRI from longitudinal profile measurements. AASHTO R 48 (2013a) Method for determining rut depth from transverse profile measurements. Requires a minimum of five transverse profile measurement points. ASTM E1703 (2015b) Method for measuring rut depth using a straight edge and a gauge. AASHTO R 40 (2018d) Method for measuring longitudinal and transverse profile using a rod and level. AASHTO R 41 (2015) Method for measuring longitudinal and transverse profile using a dipstick. AASHTO R 87 (2018e) Method for quantifying pavement deformation parameters (e.g., cross slope and rutting) using transverse profile measurements. Replaced AASHTO PP 69-14 (2017d). ASTM E1845 (2015c) Method for calculating mean profile depth from pavement macrotexture profile, used to predict speed constant (gradient) for wet pavement friction. Cr ac ki ng AASHTO R 55 (2013b) Method for quantifying cracking in asphalt pavement surface using manual or automated data collection methods. AASHTO R 85 (2018f) Quantifying surface cracking on asphalt pavement using automated data collection methods with little to no human review. Replaced AASHTO PP 67-16 (2017f). Im ag e AASHTO R 86 (2018g) Procedure for collecting pavement surface images, full lane width in the travel direction, at or near the posted speed limit using automated data collection methods. Replaced AASHTO PP 68-14 (2017e). Pr ec is io n an d Bi as ASTM C670 (2015d) Guidance for determining single-operator precision (repeatability) and multilaboratory precision (reproducibility) of a test method. ASTM C802 (2014a) Guidance on conducting an inter-laboratory study to develop a precision statement of a test method. ASTM E177 (2014b) Methods for expressing and examples of precision and bias statements. ASTM E691 (2018b) Technique for planning, conducting, analyzing, and assessing the results of an interlaboratory study of a test method. Se ns or D at a Pa ra m et er s Table 2. AASHTO and ASTM specifications for pavement condition assessment.

Literature Review 11 Canadian provincial, and local agencies have developed guidelines or use one of the following methods for conducting manual condition surveys: • ASTM D6433 (2018c), Standard Practice for Roads and Parking Lots Pavement Condition Index Surveys. • Distress Identification Manual for the LTPP (Miller and Bellinger 2014). Automated Condition Surveys Automated condition surveys are conducted using specifically designed vehicles to obtain images and profile data (e.g., IRI, rut depth, faulting) in a single pass at posted speeds. Profile data are typically processed in real time during collection. Surface distresses are determined from downward pavement images and postprocessed using either semiautomated or fully automated methods. Semiautomated methods to determine the presence of surface distress involve some level of human interaction. This typically takes place in an office setting, where a person who is trained in pavement distress identification visually categorizes distress from downward pavement images. One semiautomated method involves the rater classifying the distress (type and severity), visually measuring the extent of each distress, and entering the distress data directly into a database or pavement management system. Another method requires the rater to locate and mark the distress on the computer screen and categorize the distress (type and severity), and the software measures the extent of distress and automatically saves the results in a database. Fully automated methods involve no (or at most very minimal) human interaction for distress detection and classification. In these methods, computer software is used to review the down- ward images and automatically identify and classify distress, measure the extent of each distress, and compile the results into a database. Data Collection Technology Imaging technologies used for capturing pavement surface distress include analog, and 2D and 3D systems. Although analog images are high quality, they are difficult to manipulate and copy, and signal noise can degrade the image quality (Wang and Smadi 2011). Because of these challenges, DCVs have transitioned, over the last several decades, from analog images to include computer-based digital technology (Wang and Smadi 2011, Tsai and Li 2012). 2D systems include area-scan cameras, which scan an area of the pavement surface in a single pass, or line-scan cameras, which scan the pavement surface one line at a time (Figure 5). Area- scan cameras were used to collect pavement images for decades; however, line-scan systems have become the standard (Wang 2011). This transition is primarily due to the ability to capture a high-quality image with a line-scan system without influence from lighting conditions (Wang 2011, Wang and Smadi 2011, Tsai and Li 2012). Line-scan technology includes a highly focused narrow laser beam that laterally scans the pavement surface, a line-scan camera that captures the laser beam reflection, and computational software that generates the 2D surface image (Wang 2011). Surface distress is determined by a human rater viewing the 2D images or post- processed by analysis software. An example of a 2D data collection system is shown in Figure 6. A 3D system collects high-resolution continuous profiles that are used to construct the pavement surface (Figure 7). This occurs when 3D imaging captures the 2D intensity (intensity of reflected light; identifies stripes, cracks, aggregate, etc.) and the 3D range (elevation; identifies cracks, spalls, potholes, etc.) data. Depending on image resolution, Tsai and Wang (2018) determined that 3D systems can successfully detect crack widths greater than 0.08 in. (2 mm),

12 Automated Pavement Condition Surveys a. Area-scan technology b. Line-scan technology c. Area-scan image ~9.8 ft wide x 4.9 ft long (3 m x 1.5 m) d. Line-scan image ~13.1 ft wide x 5.7 ft long (4 m x 1.7 m) Full Frame Frame Transfer Interline Transfer Object/Web Scan area Lens Figure 5. 2D imagery technology (Wang and Smadi 2011). a. 2D schematic b. DCV with 2D system Side view Top view 13 ft (4 m) Figure 6. Example 2D system sensor arrangement (Wang and Smadi 2011).

Literature Review 13 and Wang et al. (2015) noted successful detection of crack widths of 0.04 in. (1 mm). 3D data can be used to automatically detect rutting, faulting, raveling, and potholes (Tsai et al. 2012, Tsai et al. 2015, Wang et al. 2015, Tsai and Wang 2015, Tsai and Chatterjee 2017). The laser-based 3D system is insensitive to lighting effects and low-intensity contrasts (e.g., oil stains). As long as there is an elevation difference between the crack and the surrounding surface, software algorithms are able to identify the presence of a crack (Tsai and Li 2012). 3D imagery represents an upgrade over traditional 2D systems. Wang et al. (2011) investigated the differences between data collected with 2D and 3D images. Initial testing showed that some distresses were either misidentified or missed entirely from 2D images on certain pavement surfaces (e.g., open-graded surfaces); however, the combination of 2D and 3D imagery increased the algorithm’s accuracy for crack detection and identification (Wang et al. 2011). Processing the 3D images to obtain pavement distress data requires the use of image analysis software and crack identification algorithms (Wang and Smadi 2011). Most of these programs and algorithms are proprietary and may be significantly different from each other. All programs use two different components: a method for identifying the pavement distress and a method for measuring the distress extent (e.g., length, area). Crack identification works by increasing the contrast of the image, removing the pavement area without any distress, and leaving just the visible pavement distress. The images shown in Figure 8 illustrate the 3D surface on the left side and the resulting crack detection analysis on the right side of each image. Data Management Procedures One aspect of automated pavement condition data collection that is often overlooked is the management of the collected raw data. McGhee (2004) stated that the widespread use of digital cameras meant that the collected images are stored digitally on hard drives instead of physical means, like videotapes and photo logs. Without compression, the pavement digital images of 0.6 mi (1 km) of pavement measured at 13 ft (4 m) width take up about 1 GB of hard-drive space at a resolution of 2,048 pixels wide. Compressing these image files is crucial to save hard-drive space, and the introduction of the JPEG 2000 image format allowed for about 70% reduction in image file size compared to the original JPEG image format. Eighty-percent compression could be achieved using the JPEG file format; however, compression may reduce image quality (Wang and Smadi 2011). 6.6 ft (2 m) 13.1 ft (4 m) 7.2 ft (2.2 m) Figure 7. Example of 3D system sensor arrangement (adapted from Vavrik et al. 2013).

14 Automated Pavement Condition Surveys For sensor data, upward of several gigabytes of transverse and longitudinal profile data can be collected within just a few kilometers, which is also stored on hard drives in the DCV. The collected images and sensor data must be synchronized with date, time, and location reference information. However, with ever-advancing data storage technology, storing-related issues with large data sets are being addressed and mitigated. 3D profiling of the pavement surface has been evaluated by several different entities and has been found to result in the collection of a considerable amount of data. Laurent et al. (2014) found that the 3D system evaluated was able to gather 4,096 points per profile, spanning the width of the entire 13-ft (4-m) lane. With up to 11,200 transverse profiles collected every second, nearly 19 GB of raw data were collected every mile (12 GB per kilometer). One way to reduce the amount of data collected is through the use of data screening algo- rithms. One algorithm developed by Jing et al. (2014) looked into filtering collected data where no cracks were observed. Experiments showed that the algorithm could successfully identify surface cracks while giving minimal false positives. The algorithm was fast enough to process 1023 × 1528 resolution images collected in real time at a speed of 62 mph (100 km/h). Zhao et al. (2018) looked into “big data” management methods and focused on system setups. Using a literature review and agency questions, the authors compiled a list of the three different systems being used: in-house, private-sector contracting, and university contracting. Each system has pros and cons, which each agency needs to consider when it comes to managing their vast collection of pavement condition data (Table 3). Comparison of Automated Collection Methods An important issue to consider when switching to a new technology is trustworthiness. In the case of automated data collection, how well does the automated data compare to data collected from traditional manual methods? Timm and McQueen (2004) evaluated the use of automated a. Low-severity transverse and longitudinal cracking. b. Medium-severity alligator cracking. Figure 8. Example 3D images (courtesy of New Mexico DOT).

Literature Review 15 data collection data in lieu of manual survey methods. Data were collected using manual and automated methods, and then the data were compared. The comparison found most of the distress data did not correlate well between the two data sets. Underwood et al. (2011) concluded that agencies must have precise and clear distress definitions in order for the vendor to properly identify and classify pavement distress. Addi- tionally, constant communication between the agency and vendor is needed to ensure the automated crack identification software is properly calibrated to the local conditions and reflects agency-specific distress definitions. Vavrik et al. (2013) evaluated several different test sections where pavement condition data were collected using a variety of collection methodologies on several different test sites. The results showed that inter- and intra-rater repeatability was relatively high, with good correlations. However, the comparisons between the automated data and the manual data showed that there were widespread differences between the data sets. This was likely due to the lack of vendor understanding of the DOTs distress definitions. Vavrik et al. (2013) stated that data comparisons would be more favorable with additional vendor training and discussed observed limitations with the downward-facing images. The image quality also suffered when the collection vehicle traversed over large bumps or potholes, or the vehicle tilted and caused the images to be taken at a skewed angle. Finally, Vavrik et al. (2013) noted that the image quality can suffer if the collection vehicle drives faster than the data collection can take place. Laurent et al. (2014) used 3D laser data to calculate a wide array of pavement condition indices, such as crack type and severity, mean profile depth (MPD), rutting, and raveling. The 3D lasers were used to gather data on approximately 6,200 mi (10,000 km) of pavement, and the results were compared to manual survey results. The comparison showed that more than 96% of the information within the two data sets had good agreement regarding crack type, and comparisons of multiple runs showed the 3D data to be very repeatable regarding crack detection (Laurent et al. 2014). A study performed by Texas DOT (Serigos et al. 2014) evaluated 3D laser data collection systems in a head-to-head vendor comparison. Data were collected using automated, as well as manual methods, for 20 different test sections. The results showed that automated data sets gave similar distress values compared to those from manual surveys. However, the study also Method Advantages Disadvantages In-house • Agencies best understand data and what analytics are needed. • Most flexibility to customize and make changes. • Personnel and skillsets needed to support. • Must purchase and maintain infrastructure (space, updates). • May be restricted by state information technology contract. Private Sector Contracting • No special technical expertise required by DOT. • Contractor can do analysis for DOT. • Can provide turnkey support. • May be “black box” with limited ability to customize. • Cost likely to be higher. • May need to be highly specified up front. University Contracting • No special technical expertise required by DOT. • Contractor can do analysis for DOT. • Potentially most flexible, if strong cooperative relationship can be created. • Potentially cheaper than private sector. • May be “black box” with limited ability to customize. • Staffing turnover may be a problem. • Responsiveness and support, depending on model. • May not be available. Table 3. Data storage systems advantages and disadvantages (Zhao et al. 2018).

16 Automated Pavement Condition Surveys concluded that the DOT algorithms needed additional refinement to improve the accuracy of crack identification. In 2015, Serigos et al. conducted a study to evaluate the Texas DOT automated 3D pavement collection system. In Phase 1 of the study, rutting and transverse profile measurements were collected by the Texas DOT device and four other data collection vendors. The results showed that all devices were accurate to ± 0.1875 in. (4.7625 mm), with the accuracy of the Texas DOT device ± 0.0625 in. (1.5875 mm). In Phase 2, automated distress data were collected from the Texas DOT device and three data collection vendors. The comparisons showed that manual postprocessing techniques are better at removing false positives than adding false negatives. Additionally, the Texas DOT device was not as sensitive to crack detection as the vendor devices. In Phase 3 of the study, two data collection vendors collected and conducted semiautomated analysis for two Texas DOT districts. The extent of the data collection included 7,000 mi (11,265 km) and several different types of pavements (Serigos et al. 2015). By comparing the collected data with the pavement condition data in the Texas DOT pavement management system, researchers determined that small differences in measured distresses can cause the pavement segment to fall into a different condition category (e.g., very good, good, fair, poor). Therefore, it was recommended that sections with condition ratings close to two condition categories should be looked at with more scrutiny (Serigos et al. 2015). Wimsatt (2017) evaluated the repeatability and accuracy of automated distress data collec- tion using three data collection vendors. The final result of the study was the development of a specification for automated data collection. Texas DOT ultimately decided not to include the specification in the data collection contract, but instead chose to use it as part of the quality assurance (QA) practices for the vendor-submitted pavement condition data. A study in New Zealand (Henning and Morrow 2017) investigated automated pavement data collection technologies and whether or not they could replace the manual data collection method. The objective was to determine what impact switching from manual to automated data collection would have on the quality of the pavement condition data. The authors concluded that there were differences between the surface distress data; however, those differences might also have been due to challenges in comparing the two data sets on the same road section. Although the computer algorithms needed to be adjusted to better identify distresses, Henning and Morrow (2017) recommended road agencies switch to automated data collection because it allows for complete network coverage and more accurate and repeatable measurements. Utilization of Pavement Condition Data Primary uses of pavement condition data include assessing current pavement network condition; developing pavement performance prediction models to forecast future condition; estimating short- and long-term funding needs for pavement maintenance, preservation, and rehabilitation; prioritizing treatment timing and type (cost–benefit strategy analysis); and determining optimal improvement programs (Zimmerman 2017). For highway agencies, external customers request information related to current pavement condition, forecasted pavement condition, future budget needs (backlog), funded projects, candidate projects, and future funding levels (Figure 9). Because of availability, pavement condition data are also used for the following (AASHTO 2012, Zimmerman 2017): • Establishing criteria and monitoring performance for pavement warranties; • Assessing new materials, treatments and treatment strategies, and pavement design practices;

Literature Review 17 • Supporting asset management; • Providing national reporting requirements; • Calibrating performance prediction models in the AASHTOWare Pavement ME Design™ software; • Identifying potential locations of vehicular crashes using friction data (de León Izeppi et al. 2016, Wu et al. 2014); and • Assessing environmental impacts, specifically, the selection of maintenance and rehabilitation treatments that support reduced energy and greenhouse gas emissions (Faghih-Imani and Amador-Jimenez 2013, Muench and Van Dam 2015). National Reporting Requirements SHAs are required to submit data and information for a number of national programs. These programs include the Fixing America’s Surface Transportation (FAST) Act, HPMS, and the Governmental Accounting Standards Board. Each of these programs are summarized in the following sections. Fixing America’s Surface Transportation (FAST) Act The National Highway Performance Program (NHPP) was established as part of the Moving Ahead for Progress in the 21st Century Act and requires SHAs (and metropolitan planning organizations) to report pavement and bridge conditions on Interstate and non-Interstate road- ways on the National Highway System (NHS). The required pavement performance measures to be reported to the FHWA are summarized in Table 4. Performance metrics must be collected on the full extent of the mainline highway (sampling is not allowed), in at least one direction on the Interstate System and in one direction on the non-Interstate NHS, and in the rightmost travel lane or in one consistent lane for all data; summarized over pavement section lengths of 0.10 mi (0.16 km); and reported on annually for the Interstate System and biennially for the non-Interstate NHS. 5 5 17 10 12 25 19 39 28 29 37 42 28 37 31 39 44 47 18 11 20 24 25 30 0 50 100 150 Future funding levels Candidate projects Funded projects Funding needs (backlog) Forecasted condition Current condition NO. OF AGENCIES External Stakeholders Internal Stakeholders Agency Decision Makers Elected / Appointed Officials 49 responding agencies Figure 9. Reporting pavement condition data (adapted from Zimmerman 2017).

18 Automated Pavement Condition Surveys For asphalt pavements, the performance metrics are based on IRI, cracking, and rutting; for JPCP, they include IRI, cracking, and faulting; and for continuously reinforced concrete pavement (CRCP), they include IRI and cracking (Table 5). For pavement sections with posted speeds of less than 40 mph (64 km/h), the present serviceability rating (PSR) can be submitted in lieu of IRI, cracking, rutting, and faulting. Data included in the submitted HPMS information will be used as a basis for determining agencies’ overall pavement condition. For asphalt and JPCP, the overall pavement condition is based on good-fair-poor ratings for IRI, rutting, faulting, and cracking percent or PSR; for CRCP, the overall condition is based on IRI and cracking percent or PSR (Table 7). In addition, the NHPP requires agencies to develop and use a pavement condition data quality management program to “collect and maintain standardized data to carry out a performance-based Metric Unit of Measure Test Procedure/Process Percent Pavement Good Fair Poor Asphalt Pavements IRI in./mi (mm/km) AASHTO M 328 (2018a), AASHTO R 43 (2017c) < 95 (6.0) 95–170 (6.0–10.8) > 170 (10.8) Rutting in. (mm) AASHTO R 48 (2013a) (5-point) or automated transverse profile data1 < 0.20 (5) 0.20–0.40 (5–10) > 0.40 (10) Cracking percent area Manual, semiautomated, or fully automated1,2 < 5 5–20 > 20 PSR N/A See Table 6.2 ≥ 4.0 2.0–4.0 ≤ 2.0 Jointed Plain Concrete Pavement IRI in./mi (mm/km) AASHTO M 328 (2018a), AASHTO R 43 (2017c) < 95 (6.0) 95–170 (6.0–10.8) > 170 (10.8) Faulting in. (mm) AASHTO R 36 (2017b) < 0.10 (2.5) 0.10–0.15 (2.5–3.8) > 0.15 (3.8) Cracking percent slabs Manual, semiautomated, or fully automated1,3 < 5 5–15 > 15 PSR N/A See Table 6.2 ≥ 4.0 2.0–4.0 ≤ 2.0 Continuously Reinforced Concrete Pavement IRI in./mi (mm/km) AASHTO M 328 (2018a), AASHTO R 43 (2017c) < 95 (6.0) 95–170 (6.0–10.8) > 170 (10.8) Cracking2 percent area Longitudinal cracking, punchouts, spalling, or other visible defects2 < 5 5–10 > 10 PSR N/A See Table 6.2 ≥ 4.0 2.0–4.0 ≤ 2.0 2 Cracking in wheel path only. 1 In accordance with the HPMS Field Manual (FHWA 2016). 3 Percent slabs with transverse cracking, includes partial slabs with cracking over majority of slab width. Table 5. FAST act pavement performance metrics. Percent of pavements in Poor condition Not to exceed 5% (10% Alaska only) Non-Interstate NHS Percent of pavements in Good condition Agency-established Percent of pavements in Poor condition Agency-established System Measure Minimum Level Interstate Percent of pavements in Good condition Not applicable Table 4. FAST act pavement performance measures.

Literature Review 19 approach” (Federal Register 2017). The required components of the data quality management plan (DQMP), as defined in 23 CFR Part 490.319(c) (Code of Federal Regulations 2017), include methods and processes for the following: • Data collection equipment calibration and certification: – IRI shall be collected and calculated in accordance with AASHTO M 328 and AASHTO R 43. – In accordance with the HPMS Field Manual (FHWA 2016), cracking percent for asphalt pavement can be determined using manual or automated methods and using manual, imaging, or other methods that identify at least 85% of all surface distress for CRCP and JPCP. – Rutting shall be determined in accordance with AASHTO R 48 (2013a) or the automated transverse profile data method in accordance with the HPMS Field Manual (FHWA 2016). – Faulting shall be determined in accordance with AASHTO R 36 (2013c). • Certification process for persons performing manual data collection. • Data QC measures to be conducted before data collection begins and periodically during the data collection program. • Data sampling, review, and checking processes. • Error resolution procedures and data acceptance criteria. PSR Description 4.0–5.0 Only new (or nearly new) superior pavements are likely to be smooth enough and distress free (sufficiently free of cracks and patches) to qualify for this category. Most pavements constructed or resurfaced during the data year would normally be rated in this category. 3.0–4.0 Pavements in this category, although not quite as smooth as those described above, give a first-class ride and exhibit few, if any, visible signs of surface deterioration. Asphalt pavements might be beginning to show evidence of rutting and fine random cracks. Concrete pavements might be beginning to show evidence of slight surface deterioration, such as minor cracks and spalling. 2.0–3.0 The riding qualities of pavements in this category are noticeably inferior to those of new pavements and might be barely tolerable for high-speed traffic. Surface defects of asphalt pavements may include rutting, map cracking, and extensive patching. Concrete pavements in this group may have a few joint failures, faulting and/or cracking, and some pumping. 1.0–2.0 Pavements in this category have deteriorated to such an extent that the speed of free-flow traffic is affected. The asphalt pavement may have large potholes and deep cracks. Distress includes raveling, cracking, and rutting and occurs over 50% of the surface. Concrete pavement distress includes joint spalling, patching, cracking, and scaling and might include pumping and faulting. 0.1–1.0 Pavements in this category are in an extremely deteriorated condition. The facility is passable only at reduced speeds and with considerable ride discomfort. Large potholes and deep cracks exist. Distress occurs over 75% or more of the surface. Table 6. Present serviceability rating description (FHWA 2016). Condition Category Asphalt Pavement JPCP CRCP Good All three metrics with good ratings or PSR ≥ 4.0 All three metrics with good ratings or PSR ≥ 4.0 Both metrics with good ratings or PSR ≥ 4.0 Fair Does not meet good or poor criteria Does not meet good or poor criteria Does not meet good or poor criteria Poor Two or more metrics with poor ratings or PSR ≤ 2.0 Two or more metrics with poor rating or PSR ≤ 2.0 Both metrics with poor ratings or PSR ≤ 2.0 Table 7. FAST act overall pavement condition.

20 Automated Pavement Condition Surveys Agencies are required to ensure and report the quality of all data collected in support of the pavement performance measures summarized in Table 5 regardless of the data collection method. Highway Performance Monitoring System HPMS, originally developed in 1978 as the National Highway Transportation System database, provides data on the extent, condition, performance, use, and operating characteristics of U.S. highways (FHWA 2014). HPMS data are submitted by each SHA to the FHWA for all roadways on the NHS. HPMS is used by FHWA, in support of a 1965 congressional requirement, to report every 2 years to Congress the U.S. highway condition needs. HPMS data provide a rationale for Federal-Aid Highway Program funding levels (and apportioning these funds to the SHAs), assessment of the highway system performance, and determination of the fatality and injury rates and measures FHWA’s and SHAs’ progress in meeting the FHWA’s performance plan objectives and other strategic goals (FHWA 2014). SHAs are responsible for collecting and ensuring quality of the reported HPMS data. In accordance with 23 CFR 460, SHAs are required to provide annually a certification of public road mileage to FHWA by June 1, which is used by the National Highway Traffic Safety Admin- istration to apportion highway safety funds to the SHAs (Code of Federal Regulations 2011). The previous year’s HPMS data must be submitted annually to FHWA by June 15 using the HPMS submittal software (FHWA 2014). Figure 10 provides an overview of the recommended HPMS processing cycle. The HPMS Field Manual provides details related to data collection and reporting requirements. Table 8 summarizes the pavement condition data to be collected and reported as part of HPMS (FHWA 2016). HPMS Data SHA Central Office Traffic Database Pavement Database Traffic Database Traffic Database Certified Public Road Mileage Interstate Pavement and Related Data Non-Interstate Pavement, Non- Pavement Sample and Survey Data Year 1 Jan 1 Jan 1 SHA data collection, aggregation, and quality review Submittal to FHWA, FHWA Review and Certification Apr 15 Jun 1 Jun 15 Performance Assessment Highway Statistics National Highway Datasets Conditions and Performance Report Data Distribution and Reporting Year 2 Figure 10. Recommended HPMS processing cycle (adapted from FHWA 2016).

Literature Review 21 IRI All NHS roadways, excludes minor arterials, minor and major arterials, and locals AASHTO M 328 (2018a), AASHTO R 43 (2017c), AASHTO R 56 (2018c), and AASHTO R 57 (2018b) Data Item1 Functional Classification Testing Protocol PSR All NHS roadways—optional on roadways with posted speed < 40 mph (64 km/h) See Table 6. Rutting All NHS roadways AASHTO R 48 (2013a), AASHTO PP 69 (2017d) (replaced with AASHTO R 87 [2018e]), and AASHTO PP 70 (2017a) (replaced with AASHTO R 88 [2018h]) Faulting All NHS roadways AASHTO R 36 (2013c) (currently AASHTO R 36 [2017b]) Cracking percent All NHS roadways Manual, AASHTO R 55 (2013b), AASHTO PP 67 (2014) (replaced with AASHTO R 85 [2018f]), and AASHTO PP 68 (2017e) (replaced with AASHTO R 86 [2018g]) • Asphalt pavements—percent fatigue and longitudinal cracking in wheel path (all severity levels). See Figure 11. • JPCP—percent of slabs with transverse cracking, includes slabs when crack extends over majority of width. • CRCP—percent area with longitudinal cracking, punchouts, and patching. 1 Weighted average Table 8. HPMS pavement-related data requirements (FHWA 2016). Adjacent lane Survey lane Shoulder or Adjacent lane La ne e dg e La ne e dg e Zo ne 1 Zo ne 2 Zo ne 3 Zo ne 4 Zo ne 5 Inside wheelpath Outside wheelpath Lane Centerline varies varies 39.4 in. (1 m) 39.4 in. (1 m) 29.6 in. (0.75 m) Figure 11. Wheel path definition. Governmental Accounting Standards Board The Governmental Accounting Standards Board (GASB) governs the financial reporting of infrastructure assets. In 1999 GASB approved Statement No. 34, Basic Financial Statements— and Management’s Discussion and Analysis—for State and Local Governments (GASB 1999). GASB 34 requires state and local governments to report the cost of infrastructure assets, allowing each agency to determine its own asset management methodologies, systems, and standards (PB Consult et al. 2004). For pavements and bridges, SHA reporting includes the measure, a description of the measure, the latest value of the measure, and the target value. Examples of SHA pavement performance measures are shown in Table 9.

22 Automated Pavement Condition Surveys Data Quality Management Plans As of April 2016, 23 CFR Part 490 required all U.S. DOTs to have a data quality manage- ment program describing the process to ensure pavement condition data are of high quality and representative of field conditions (Code of Federal Regulations 2017). Each SHA was required to submit their DQMP to FHWA for review and approval by May 20, 2018. FHWA’s Practical Guide for Quality Management of Pavement Condition Data Collection describes the steps for creating a data quality management plan (Pierce et al. 2013). Typical contents of DQMPs include the following: • Deliverables, Protocols, and Quality Standards. This section outlines the pavement condi- tion data items to be collected, for example, surface distresses, IRI, rutting, faulting, distance measurements, and GPS coordinates. This section may also specify, for each data item, the units of measure, the number of significant digits to report, the repeatability of the measurements, and the accuracy of the measurements (in relation to a known value). • Quality Control. This section describes QC activities performed during data collection. Activities may include, for example, equipment calibration; software checks; and data collection on control, verifications, or blind sites. • Acceptance. This section describes activities performed by the agency to determine if the collected data meet agency requirements. Acceptance activities may include review of pavement images, database checks, and comparison of survey results with acceptance review results. In the event that the collected data do not meet the acceptance requirements, reprocessing or re-collection may be required. • Reporting and Tracking. An important part of quality management is the documentation of all quality-related activities. Reporting and tracking allow the agency to – Review quality activities and make sure that any problems or issues have been resolved. – Identify problems requiring a more thorough review and minimize recurring issues from happening in the future. – Update and revise quality management activities. Agency Measure Description Latest Value Target Alabama Distress rating 0–100 scale; based on roughness, cracking, rutting, patching, and raveling 79.7 ≥ 75 Arizona PSR 0–5 scale; based on subjective rating by road users 3.6 ≥ 3.23 Delaware Overall pavement condition 0–5 scale;% in poor condition; based on pavement distress 9.8 ≤ 15 Idaho Roughness and cracking index 0–5 scale;% in poor condition 18 < 2.5 ≤ 18 < 2.5 Kentucky Pavement condition index Percent poor based on pavement smoothness 20.6 ≤ 30 Maine Highway adequacy 0–100 scale; based on pavement condition rating, safety, backlog, average daily traffic, posted speed and shoulder 76.6 60 Nebraska Serviceability index 0–100 scale; based on cracking, patching, roughness, rutting, and faulting 84 ≥ 72 Tennessee Maintenance rating index 0–100 scale; based on pavement, shoulder, roadside elements, drainage, and traffic services 87.75 ≥ 75 Table 9. Examples of SHA GASB 34 performance measures (adapted from PB Consult et al. 2004).

Literature Review 23 Cost, Advantages, and Disadvantages McGhee (2004) conducted an agency survey, and the results showed that the total cost of data collection and processing ranged from about $30 to $125 or more per mile depending on several different factors. The advantages of automated data collection included minimal impact on traffic, a significant increase in safety, more time efficiency, and the possibility of 100% network coverage. In addition, pavement images represented a permanent record of pavement distress, and data collection costs were reduced compared to manual surveys. Vavrik et al. (2013) summarized the benefits and challenges for switching from manual to automated pavement condition surveys (Table 10). Collection and Analysis of Other Appurtenances In a study for Georgia DOT, Tsai et al. (2017) attempted to develop automated procedures to perform traffic sign inventory from video log images and mobile light detection and ranging (LiDAR) data. Previously Georgia DOT used a windshield sign inventory process that required approximately 43 min per sign. The developed automated process required 12 min per sign, a 72% reduction, without compromising the quality of the collected data. Wright et al. (2014) conducted a study to determine if LiDAR could be used to inventory roadside barriers. The goal was to inventory assets as well as measure the heights of the barrier rails. An algorithm was developed and modified using the results of multiple field trials, which showed the algorithm could calculate the height ±2 in. (50 mm) for 80% of the barriers surveyed through automated means. The report concluded that LiDAR could play a vital role in the automated collection of roadway assets and should be further evaluated. Kim et al. (2009) evaluated several data collection vendors to determine their relative perfor- mance in collecting pavement condition data and roadside appurtenances on a 90-mi (145-km) test loop. For roadside appurtenances, various assets such as pavement markings, road geometry parameters (e.g., grade, horizontal curve radius), and roadside assets (e.g., barrier rail, drop Table 10. Advantages and disadvantages of automated data collection (Vavrik et al. 2013). Advantages Disadvantages • Increased rater safety • Improved data accuracy for certain distresses • Enhanced timeliness of data collection and processing • Ability to easily track, review, and reproduce historical data and images • Ability to collect data compatible with HPMS requirements • District access to vendors for ancillary data collection • Consistent, well-defined methods for future automated distress identification • The ability to combine IRI, rutting, and asset collection with pavement distress ratings • Losing the ability to directly correlate with some historical pavement condition data • Becoming tied to technological evolution that forces early equipment replacement • Increased annual collection and processing costs • Difficulties associated with operational change • Loss of control due to dependence on a single vendor • Potential variability of vendor results year to year • Additional initial costs and personnel demands associated with procurement, calibration, and implementation of system • Breakdowns and long repair delays for DOT- purchased equipment • Additional costs associated with modifying the distress ratings, distress manual, decision trees, pavement performance models, and pavement management software

24 Automated Pavement Condition Surveys inlets) were collected and rated. The collection vendors were able to inventory nearly all of the assets encountered on the test loop, but there were discrepancies as to the condition of the assets when compared to manual surveys. However, these differences could be related to unclear or vague definitions as to what is considered a barrier rail in good condition or a drop inlet considered blocked (Kim et al. 2009). In Wyoming, Andreen et al. (2010) showed that single-car accident rates increase as road side-slopes become steeper, presenting a need for the Wyoming DOT to measure slope values on all state-managed roads. Several test sections were identified, and manual and automated slope measurements were calculated and compared. The results showed the automated slope values were not as accurate as the manual measurements. A more in-depth analysis of the automated results showed that the measurement was affected by the presence of vegetation within the shoulder.