Bridge Element Data Collection and Use (2022)

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.

7   History of Bridge Element Data The aftermath of two dramatic and fatal bridge failures—first the Silver Bridge in West Virginia in 1967 and later the Schoharie Creek Bridge in New York in 1983 (Swenson and Ingraffea 1991, Small et al. 1999)—and the increasing gap between the available funds and needs of the national bridge network stimulated increasing research to develop BMSs in the mid-1980s (Bektaş 2011). Subsequently, the Intermodal Surface Transportation Efficiency Act of 1991 required states to develop and implement BMSs to improve management of the national bridge network. Although the National Highway System Designation Act of 1995 made BMS development optional, many states decided to continue implementing them (Sanford, Herabat, and McNeil 1999). While some states set out to craft their own tools, many decided to implement available systems. Under a research project directed by the FHWA, California DOT, and other state transportation represen- tatives, the Pontis Bridge Management System (Pontis BMS) was developed in 1989 (Cambridge Systematics 2005a), which later became the most popular bridge management tool in the United States (Thompson and Shepard 1994). NCHRP then developed the BRIDGIT BMS, but it has not become as popular as Pontis (Hawk 1998). The Pontis BMS has evolved through changes of the modeling framework over the years, with a significant change in its prioritization algorithm based on an NCHRP project study report published in 2007 (Patidar et al. 2007). Following this change, the Pontis BMS was renamed as the AASHTOWare BrM software. The foundation of the BMS modeling approach lies in the principles of operations research and economic analysis. Typical main inputs for BMS preservation modeling are bridge element condition data, cost models, and deterioration models. In most BMSs, bridges are represented as a set of structural elements (Figure 2-1). The bridge is divided into individual elements (or sections) that consist of the same material. These homogeneous elements thus are expected to deteriorate in the same manner. This analytical approach of subdividing a structure into its ele- ments facilitates the development of associated deterioration and cost models. Bridge deterioration is modeled with a probabilistic approach in most BMSs, recognizing the uncertainty and variability in deterioration rates. The most common deterioration model used in BMSs is the Markovian deterioration model, which can be expressed as a matrix of transition probabilities. BMS modeling also includes feasible actions and associated costs applicable to each element and CS. These models work together through a prioritization algorithm to identify the most cost-effective treatments or the treatments that maximize the BMS prioritization objective function. This function ideally is a mathematical representation of the agency bridge manage- ment vision and objectives. As discussed in Chapter 1 in the background section, bridge element data collection in the United States started in the late 1990s with the development and implementation of BMSs. C H A P T E R 2 Literature Review

8 Bridge Element Data Collection and Use AASHTO published the CoRe Structural Elements Guide (AASHTO 1997) to define and standardize bridge element data inspection and collection. Subsequently, the AASHTO Sub- committee on Bridges and Structures approved a new element-level bridge inspection manual (AASHTO 2011), which was later updated (AASHTO 2013). The new MBEI (AASHTO 2013, 2019) replaced the CoRe Structural Elements Guide. The new element inspection methodology incorporated the experience of state DOTs in inspecting elements over 20 years, thereby improving the guidelines. The bridge elements in the new manual are classified into three categories: NBEs, BMEs, and ADEs. NBEs are the primary structural components of bridges necessary to determine the over- all condition and safety of primary load-carrying members (e.g., girders, trusses, columns) (AASHTO 2013). The NBEs are a refinement of NBI bridge components (deck, superstructure, and substructure), as defined in the FHWA Recording and Coding Guide (FHWA 1995). BMEs include bridge components (e.g., joints, wearing surfaces, protective coating systems, and deck and slab protection systems) that are typically managed by BMSs. ADEs provide flexibility for agencies that need to define custom elements, which may be sub-elements of NBEs or BMEs or may be independent elements (Figure 2-2). Element identification numbers 800 and higher in the AASHTO MBEI are reserved for ADEs. The number of CSs is now standardized as four for NBEs and BMEs whereas CoRe elements previously had three, four, or five CSs. Also, the new manual introduces a new defect-based inspection methodology and separates wearing surfaces and protection systems from structural elements. Element-level inspection methodology subdivides the bridge components (deck, super- structure, and substructure or culvert) into more granular elements, and it quantifies the type, severity, and extent of defects (FHWA 2016). The condition of each element is inspected and reported according to a CS, which is a quantitative measure of deterioration (Cambridge System- atics 2005a). The amount of each element is quantified for its size (area or length) and the number of elements. During each inspection, the total quantity of an element is distributed among four CSs, depending on the type and severity of defects that are present. A rating of CS1 represents a like- new condition while CS4 indicates a severe condition that warrants structural review, repair, or replacement. Element CSs as defined in the AASHTO MBEI are shown in Figure 2-3. Photos with examples of CSs 2–4 are presented in Figure 2-4. Figure 2-1. Schema of major elements in a bridge.

Literature Review 9   Bridges and elements in different external environments deteriorate differently. For example, bridges subject to higher traffic volumes may deteriorate faster, and so may external girders exposed to more road salt or water. This difference in deterioration attributable to different environments is facilitated by the environmental factors (service environments) in the element- level inspection methodology. The AASHTO MBEI (AASHTO 2011, 2013, 2019) defines four environments as benign, low, moderate, and severe (from the least severe to the most harsh, respectively). MAP-21 requires each state and appropriate federal agency to report bridge element-level data for NHS bridges. As a result, state transportation agencies have revised their element inspection manuals and condition databases to adopt the new elements. As of October 1, 2014, state DOTs were required to begin collecting element-level data, in accordance with 23 U.S.C. 144(d)(2); for the April 1, 2015, NBI data submission, agencies included element-level data for part of their inventories. State DOTs are required to submit data on NBEs and on some BMEs (e.g., bridge rails, joints, bearings, wearing surfaces, and protective coatings), based on the guidelines in the FHWA specification for NBI bridge elements (FHWA 2013). Figure 2-2. Examples of ADEs from Michigan DOT (Michigan DOT 2015). Figure 2-3. Bridge element condition states (AASHTO 2013).

10 Bridge Element Data Collection and Use Outside of the United States, existing BMSs also typically utilize an element data framework. Shortly after development of the Pontis BMS in the United States, Stantec developed the first BMS in Canada for the Ontario Ministry of Transportation (Thompson et al. 1999), adopting a four-CS element data framework. A 2014 survey by the International Association for Bridge Maintenance and Safety (IABMAS) Bridge Management Committee reported findings on 25  inter national BMSs that all used element-level condition data (Mirzaei et al. 2014). The reported BMSs were from 18 countries: Australia, Canada, Denmark, Finland, Germany, Ire- land, Italy, Japan, Korea, Latvia, Netherlands, Norway, Poland, Spain, Sweden, Switzerland, United States, and Vietnam (Mirzaei et al. 2014). The number of CSs ranged from three to nine, but 15 BMSs in the survey used four CSs, the most common number (Mirzaei et al. 2014). Bridge Element Data Quality The majority of bridge element data are collected during routine inspections, which are per- formed periodically during the service life of a bridge, every 24 months for most bridges (Phares et al. 2004, Gattulli and Chiaramonte 2005, Washer et al. 2020). A recent NCHRP study by Washer et al. (2019) surveyed 36 DOTs on a variety of topics regarding the quality of element- level bridge inspection data in 2015, leading to research deliverables that aim to increase the quality of bridge element data. Washer et al. (2019) reported a number of findings. About 80% of the DOT respondents reported that they began collecting element-level data before 2010 and that they had element-level Condition State 2 Condition State 3 Condition State 4 Condition State 4Condition State 3Condition State 2 Figure 2-4. Examples of reinforced concrete condition state (Michigan DOT 2015).

Literature Review 11   inspection manuals that include at least the FHWA-required elements and the defect elements identified in the MBEI. At the time of the survey, the use of photographs to illustrate different CSs was limited in the inspection manuals. Only 14% of the respondents indicated a speci- fied accuracy requirement for CS quantities. Only two respondent agencies had a similar accu- racy requirement for defects (out of 22 agencies that reported collecting element defect data). Responses to a survey question on how the agencies convey the results of QA processes indicated that 14% of the respondents used reports, 23% relied on face-to-face meetings, and 63% used both. In addition, 90% of the respondents also held periodic meetings with inspectors. The NCHRP survey (Washer et al. 2019) also posed specific questions on inspection proce- dures and practices. Responses to several questions on CS quantities indicated variation in the methodology for making a quantity estimate, so some difference in the consistency of results consequently would be expected. In response to a question on what photographs were required as part of the inspection, about 70% of the DOTs reported that the photographs described in the MBEI were demanded; other responses noted that photographs were required for defects that resulted in CS 2, 3, or 4 (or an NBI GCR below 6). For about 16% of the respondent DOTs, the inspection teams decided on the appropriate photographs. A major survey finding suggested that most of the responding agencies do not have accuracy requirements for element-level inspec- tion data. The report (Washer et al. 2019) included a methodology for developing such an accu- racy requirement. The report guidelines also included field test methods for improving quality. The visual guide from this report, intended to improve the quality of element-level inspection data, was incorporated into the revised version of the MBEI, which AASHTO approved in 2018 (AASHTO 2019). In this revision, AASHTO reorganized and added visual standards in the form of photographs for many of the defects identified in the manual, along with grouping ele- ment descriptions, defects, and units of measure by material type. Key guidelines based on the research results included accuracy requirements, performance testing of inspectors, and inspec- tor calibrations. Field tests in Michigan and Indiana that used the visual guide found variability in damage quantities—on the order of more than 50% of the quantity being measured—based on a statistical analysis of the data. Different inspectors also reported different defect elements and different CSs for the same bridge element. The report concluded that additional training and experience in applying the visual guide are likely needed to enable the inspectors to measure improvements in data quality. Routine inspections rely heavily on visual inspections and consequently subjective assess- ments by the inspectors (Phares et al. 2004). Asset management decisions for bridges are typi- cally governed by the NBI GCRs, which are qualitative and foster reactive decision-making (Flanigan, Lynch, and Ettouney 2020). In that regard, condition accuracy and bridge element data reliability are critical for bridge management functions that ultimately aim for the most cost-effective allocation of bridge funds. Cognizant of the pressing need for data quality, FHWA revised National Bridge Inspections Standards (NBIS) in 2004 and required each state and federal agency to implement a QC and QA procedure for bridge inspection programs by January 13, 2006 (National Bridge Inspection Standards, 23 CFR 650 Subpart C). In compliance with the NBIS, the state DOTs have QC and QA processes in place, which are often further customized to specific agency needs and approaches. Case examples in Chapter 4 illustrate some of these processes. Performance Measures Based on Element Data Currently, the most acknowledged performance measure based on bridge element data is the Bridge Health Index (BHI) (Thompson et al. 1998). First used in California, the BHI is a single number (from 0 to 100) that reflects the overall physical condition of the bridge; it is computed

12 Bridge Element Data Collection and Use as a weighted average of the condition of the individual bridge elements (Shepard and Johnson 2001, Cambridge Systematics 2005b). This index reflects a weighted condition distribution of the BMS elements, with weights determined by expert assignments or element failure costs. The BHI is computed (Shepard and Johnson 2001) as 100 (1)= ∗HI CEV TEV (2)∑ ∑=CEV W Q Wee ej ejj Ne (3)∑ ∑=TEV W Qee ejj Ne 1 1 1 (4)= − − − W j Nej e where HI = health index CEV = current element value TEV = total element value We = weight given to element e (based on the repair, replacement or failure cost, or expert opinion) Wej = weight of CS j on element e Qej = quantity of element e in CS j Ne = number of CSs defined for element e (typically four) For the weighting factor, the Pontis BMS used one of two values for each element. That value could be either the element’s failure cost (calculated as a sum of its agency and user failure cost components) or the weight coefficient explicitly assigned to the element multiplied by the cost of the most expensive action defined for that element (typically element replacement) (Cambridge Systematics 2005b). BMSs take different approaches to weighting, with no current nationally recognized index. In BMSs where BHI is utilized as a performance measure, such as the AASHTOWare BrM software, the agencies are responsible for choosing the type of weight- ing method and the relative magnitudes of element importance factors and weights. Early in BHI implementation, the BHI values typically accumulated at the higher end of the 0–100 range; therefore, relative health index values did not always convey a clear notion of relative perfor- mance (Bektaş 2011). Modifications to the BHI have been proposed in the literature. One sug- gestion was use of a nonlinear relationship between the CS weights (termed the health index coefficient) and the CS, which leads to a relatively faster deterioration rate between the CSs, with the element health index values reduced relative to the CSs (Jiang and Rens 2010a, Jiang and Rens 2010b, Sobanjo and Thompson 2011, Inkoom et al. 2017). In addition, Inkoom and Sobanjo (2018) proposed bridge element importance weights based on the element replacement cost, long-term cost, and vulnerability to hazards. The linear and pessimistic trends used in the Inkoom and Sobanjo (2018) study produced lower BHI values than the optimistic trends, and the use of adjustment factors overall led to a significant reduction in the network BHI values. Fereshtehnejad et al. (2017) proposed another condition index and utilized Ohio data. The Ohio bridge condition index (OBCI) was designed to evaluate bridges at the element, compo- nent, bridge, and network levels and reflect the impact of existing defects as well as maintenance, repair, and replacement actions on the condition of the system. OBCI was also defined as a multilevel cost-based performance index, and its calculations relied heavily on implementation costs and structural and serviceability failure costs.

Literature Review 13   Hearn (2015) proposed measures of highway bridge performance based on bridge element data. Hearn (2015) noted that element-level measures recognize relationships between bridge elements (e.g., joints, wearing surfaces, coatings) that affect the exposure of deck, super structure, and substructure. Poor conditions of these elements in turn put decks, superstructures, and substructures at risk. Moreover, risk also increases when actions for preservation are delayed or limited. In this study (Hearn 2015), three levels of performance were included: good, at risk, and poor. Bridges with a good performance level were candidates for bridge preservation; bridges at risk were viewed as candidates for prompt attention to preserve their current condition; and bridges with poor performance required rehabilitation or replacement. The risk from exposure and the risk from delayed preservation were recognized in element-level performance measures. The Hearn (2015) study included examples of element-level performance measures for a popu- lation of state-owned bridges, comparisons with current performance measures of bridges, and extended use of element-level performance measures to estimate annual work scope and fund- ing for preservation. In addition to BHI variants, other performance measures based on bridge element data have also been suggested in the literature. Saydam, Frangopol, and Dong (2013) proposed a method- ology to quantify the lifetime risk associated with bridge superstructures based on bridge ele- ment data. This methodology used a reliability-based approach and computed the probabilities of component and system failures based on element conditions and a Markovian deterioration process. In response to one of the survey questions for the current study, several state DOTs shared custom performance measures that they are currently utilizing. TAMPs submitted to the FHWA in 2019 (FHWA 2019) were also reviewed to identify DOTs that referred to element perfor- mance measures, exemplified by the following: • Delaware DOT. As noted by the Delaware DOT (2019), the state uses a custom deficiency formula and AASHTOWare BrM outputs to generate a list of bridges that require work according to the preservation models (Figure 2-5). AASHTOWare BrM run results are com- piled into the deficiency formula spreadsheet, and bridges are ranked by deficiency points in descending order. For the BMS runs, the Delaware DOT employs a utility function that combines condition (40%), life-cycle costs (30%), mobility (15%), and risk (15%). The condi- tion utility is heavily influenced by element data because element ratings are weighted at 90% while NBI GCRs are weighted at 10%. The Delaware DOT also noted paint, deck overlay, and pile jacketing actions that are triggered by the BHI. • Connecticut DOT. The Connecticut projections of bridge performance were developed in the Deighton dTIMS® infrastructure asset management software BMS, using a snapshot of condition data from May 2019 (Connecticut DOT 2019). The bridge projection analysis is run to optimize a BHI. The BHI consists of weighted condition ratings: 15% deck; 15% super structure; 15% substructure; 10% structural evaluation; 5% deck geometry; 5% under- clearances; 5% waterway adequacy; 4% approach alignment; 2% structure open, posted, or closed; 5% paint and coating; 5% bearings; 5% girders; 5% joints; and 4% wearing surfaces (Figure 2-6). • Utah DOT. The state developed a custom BHI composed of three separate scores (for deck, superstructure, and substructure) that are weighted to underscore the importance of each category in overall bridge health (Utah DOT 2017). The weighting of these categories is as follows: BHI = (Deck Score × 0.40) + (Superstructure Score × 0.35) + (Substructure Score × 0.25) The health of deck elements is weighted higher because such elements contribute to preserv- ing many other areas of the structure. Culverts employ a different BHI scoring system and are

14 Bridge Element Data Collection and Use Figure 2-6. Connecticut DOT BHI components (Connecticut DOT 2019). Figure 2-5. Delaware DOT deficiency formula factors (Delaware DOT 2019).

Literature Review 15   rated (on a scale of 1 to 100) based on inert culvert elements. The Utah DOT uses the BHI as an infrastructure performance measure and publishes it on external-facing dashboards (Fig- ure 2-7). The 2016–2020 BHI scores for Utah show a decreasing trend and values between the high 80s and 70s, with overall higher ratings for NHS, state, and local systems, in that order. • North Dakota DOT. In its TAMP, the North Dakota DOT also refers to use of the BHI as a planning metric (North Dakota DOT 2019). • Oregon DOT. The Oregon DOT employs a remaining service life method, which is an empirically derived equation that incorporates the deck, superstructure, and substructure NBI ratings to capture the overall bridge conditions, along with component health indexes that capture element conditions (Oregon DOT 2019). The component NBI and health index values are combined into the deck, superstructure, and substructure utilities and then incor- porated into one weighted bridge utility that includes loading conditions as follows: ODOT Bridge Utility = (Deck Utility × 0.10) + (Super Utility × 0.30) + (Sub Utility × 0.30) + (Loading Utility × 0.30) Models Based on Element Data The most common element models used in asset decision-making are cost models, deteriora- tion models, and mapping models between element conditions and NBI GCRs. Figure 2-7. Utah DOT BHI performance measure (Dashboard) (Utah DOT 2017).

16 Bridge Element Data Collection and Use In the literature, studies on element cost models are typically deliverables for agency-led research projects (Adams and Barut 1998, Adams and Sianipar 1998, Sobanjo et  al. 2002), focusing on developing agency cost values for maintenance, rehabilitation, and repair (MR&R) of bridges. Such element cost model studies are fewer in number than those on deterioration modeling. During March 4–11, 2021, nine state DOTs gave presentations for the NCHRP U.S. Domestic Scan Program on successful approaches for using BMSs for strategic decision-making in asset management plans (Bektaş et al. ongoing). During their presentations, state DOTs reported on ongoing and planned efforts for deterioration models, cost models, life-cycle cost models, per- formance measures, and treatment efficiency models for bridge elements. The development and use of bridge element models appear to be an area where the state DOTs currently have needs, plans, and initiatives. In 2020, NCHRP published a project report that developed bridge and deck preservation guides for possible adoption by AASHTO (Hearn 2020). The proposed AASHTO guides were based on data collected from representative agencies and included (1) catalogs of bridge ele- ment preservation actions and (2) criteria and selection methodologies for bridge preserva- tion actions, with associated costs and benefits for use in life-cycle cost analyses and possible integration into a BMS. The Hearn (2020) report included detailed information on the context, performance, and cost of bridge preservation actions, mostly considering GCRs, but including element data references as available. Bridge management literature is quite rich with studies on the development and implemen- tation of element deterioration models. An NCHRP report on estimating highway asset life expectancies (Thompson et al. 2012) is a guidebook that provides a methodology for estimating the life expectancies of major types of highway system assets for use in life-cycle cost analyses that support management decision-making. One of the authors of this guidebook recently pub- lished a national bridge element deterioration model for the United States (Thompson 2018) that includes not only algebraic methods developed in research for Florida and Virginia for use in processing element inspection data into transition probability matrixes but also guidance on developing models compatible with MBEI. As discussed earlier in this chapter, Markovian deterioration models are by far the most commonly used deterioration models for bridge elements (Thompson and Johnson 2005). For Markovian models, which are discrete models, uncertainty is often quantified by using a con- stant transition probability from one state to another state in a single year (Thompson et al. 2012), perfectly aligned with the CSs in bridge element data. Markovian models, however, are time-independent (memoryless), so the transition probabilities are independent of how long a bridge element has already stayed in a certain CS. The memoryless property of Markov chains has been criticized compared to models that assume time-dependent distribution of sojourn (waiting) times, such as the Weibull distribution. For Markovian models, the transition probabilities are entirely dependent on the CSs. To adjust in particular the transition probabilities from CS1 to CS2, Sobanjo and Thompson (2011) developed a hybrid Markov model. This model assumed a Weibull distribution for the time spent in CS1 before transitioning to CS2 and applied a Markov chain model developed from regression analyses for the time spent in the remaining CSs. This model is currently used in AASHTOWare BrM software. The state DOTs have previ- ously developed custom element deterioration models; two case examples from Florida and Wisconsin are included in Chapter 4. In addition, an ongoing pooled-fund project led by the Wisconsin DOT—TPF-5(432): Bridge Element Deterioration for Midwest States—aims to develop element deterioration curves based on historic bridge data, operation practices, maintenance activities, and historic design and construction details from Midwestern DOTs.

Literature Review 17   Another category of bridge element models is mapping models, which predict NBI GCRs based on bridge element data. Since the 1990s, with the start of CoRe element data collection, an associated interest and need arose for a mapping method between the two. BMSNBI software (also known as NBI Translator), which maps the CoRe element condition data to the NBI con- dition ratings, was developed for this purpose for the FHWA (Hearn et al. 1993). BMSNBI was also a built-in module for Pontis BMS, the predecessor of AASHTOWare BrM software. States had the option to report translated NBI GCRs instead of field ratings to FHWA for the annual NBI submissions. However, BMSNBI was used by only a few state agencies for this purpose, mostly because of general skepticism among the states regarding the accuracy of the BMSNBI algorithm (Hale, Hale, and Sharpe 2007; Bektaş 2011). In comparison studies between BMSNBI- generated and field ratings (Herabat and Tangphaisankun 2005, Aldemir-Bektaş and Smadi 2008), low predictive accuracies were reported. However, BMSNBI was conservative by design in mapping high NBI condition ratings because a major objective was to map condition ratings that determined structural deficiency and Highway Bridge Program (HBP) funding eligibility; BMSNBI thus tends to assign lower ratings, especially for ratings above 7. Regardless of these issues, the BMSNBI algorithm was utilized for many purposes, including mapping future NBI GCRs from forecast element condition data or in the National Bridge Investment Analysis System (NBIAS) software tool, which FHWA uses to forecast bridge needs for the Condition and Perfor- mance reports prepared for Congress. Another current use of this mapping focuses on predicting federal bridge condition perfor- mance measures for element-based BMS analysis outputs. The condition of bridges carrying the NHS, including on-ramps and off-ramps connecting to the NHS, are classified as good, fair, or poor based on NBI GCRs and NBI items (58 for deck, 59 for superstructure, 60 for substructure, and 62 for culverts). For national performance measures under the National Highway Perfor- mance Program, the assessment method for determining the classification of a bridge is based on the minimum of condition rating method (i.e., the condition ratings for the lowest rating of a bridge’s three NBI items for deck (58), superstructure (59), and substructure (60). For culverts, the rating of their NBI item (62) determines their classification). Federal bridge performance measures record the percentage of NHS bridges (by deck area) in good condition and the per- centage in poor condition. Alternative algorithms were proposed in the literature. For example, an artificial neural net- work (ANN) model reported improved prediction accuracy (Al-Wazeer, Nutakor, and Harris 2007). Another alternative tool, the NewTranslator, employs a modeling approach similar to that of the BHI (Shepard and Johnson 1999), assigning an NBI rating for an NBI class based on the assigned NBI ratings of CoRe elements and element weights based on expert opinion. The approach produced more accurate results in assigning NBI ratings in the higher range (Sobanjo, Thompson, and Kerr 2008). A study by Bektaş, Carriquiry, and Smadi (2012) proposed a potentially more accurate method based on recursive partitioning and, compared to previous algorithms, improved the percentage of both exact predictions and predictions within one error term. Recursive partitioning was also recently applied to develop classification trees that predict NBI GCRs by analyzing NBE condition data, resulting in preliminary decision trees with predictive accuracy that is sufficient for adoption by transportation agencies (Bektaş, Carriquiry, and Smadi 2013).