Estimating Life Expectancies of Highway Assets, Volume 1: Guidebook (2012)

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.

20 Life expectancy models can be simple or sophisticated, with various options for policy sensi- tivity, accuracy, and precision. The selection of models will depend on how the information will be used. For example • For asset valuation, such as the basic GASB 34 approach, agencies may decide to use straight- line depreciation to convert asset age directly to dollars of value. Total asset lifespan in this case might be determined from a table of accounting conventions, with the remaining life of a given asset determined directly from its age (Figure 3-1, left side). • For relatively low-value assets whose condition is not routinely monitored (e.g., roadside reflectors), their lifespan might be determined from the manufacturer’s recommendations or from agency experience and then applied to a whole population of features. All of the features in the population are replaced at the same time in a single project, even if certain assets in the group were already replaced earlier due to premature failure (Figure 3-1, right side). • For higher value assets which are custom-made and whose condition is monitored by periodic manual or automated processes (e.g., signs and pavement markings), their condition may be translated directly to life expectancy using simple deterioration models. Replacement is triggered when the condition passes a performance threshold (Figure 3-2, left side). There may be more than one performance measure that could trigger replacement (e.g., pavement cracking and rutting). • For large constructed facilities, condition and performance may be input to a lifecycle pres- ervation optimization model and/or long-range decision-making process to plan preventive maintenance actions, repairs, rehabilitation, and replacement. Life expectancy is policy- sensitive and may vary based on the maintenance policies and programming decisions made in the intermediate period before the end of the asset’s life. The definition of end-of-life may itself be dependent on the agency’s policy, program, and project decisions (Figure 3-2, right side). In order to adopt the more policy-sensitive life expectancy methods, it is not sufficient to perform a more elaborate calculation. In addition, it is necessary for an agency to • Gather and manage data on asset condition and performance on a regular basis. • In some cases, gather and manage data on asset repair and replacement activities. • Develop warrants and feasibility criteria for maintenance, repair, rehabilitation, and replacement. • Develop data on crew and/or contractor capabilities, as well as materials and equipment, to support life-extension activities. • Develop planning processes that can forecast and program life-extension activities at the most favorable time. • Earn stakeholder confidence that the life-extension activities are cost-effective enough so that an appropriate budget level is established for them. C h a p t e r 3 Establish the Framework: How to Design Life Expectancy Models

establish the Framework: how to Design Life expectancy Models 21 This is why the concept of agency maturity, introduced in Chapter 1, is so important for selecting appropriate methods for calculating life expectancy. Agencies that are higher on the asset management maturity scale tend to conduct condition and performance monitoring on a wider range of facilities and tend to have end-of-life definitions that are more often policy- sensitive. This is just another way of saying that they are proactive in their decision-making and have more alternatives available for extending asset life instead of automatic replacement. 3.1 Defining Performance Measures One of the reasons for implementing life expectancy analysis is to use it as an outcome mea- sure of infrastructure health or of preservation work accomplished. Often an informal justifica- tion given for preservation activity is “to extend asset life.” Whether this argument is under- standable or verifiable may depend on context. Consider, for example, the following scenarios: 1. The asset is at the end of its normal life expectancy. It is in poor condition or performing at a level that is below agency standards. Replacement is a justifiable alternative. There is also a repair or rehabilitation alternative that is less expensive than replacement and that will alleviate the current deficiencies for a period of time before replacement once again must be considered. 2. The asset is at the end of its life expectancy. It is in serviceable condition and functions according to agency standards. There is some risk that the asset might fail suddenly and cause an interruption to traffic. 3. The asset was procured with a 10-year life expectancy but is already performing below standard after 5 years. It can be repaired or rehabilitated, which may correct the deficiency and provide 2 to 3 years of additional life. Subsequent repair may or may not be able to offer further life extension. Figure 3-1. Pre-determined interval-based life expectancy. Figure 3-2. Condition or performance-based life expectancy.

22 estimating Life expectancies of highway assets 4. The asset consists of separate and distinct components, and each component has its associ- ated set of preservation actions that may and may not influence the life of other compo- nents. For example, consider a bridge having 25 years of remaining life that is functioning well, but the protective steel coating is deteriorating. If allowed to remain as it is, the steel elements of the bridge might last only 10 years. If the coating is replaced, the bridge is likely to realize its full 25 remaining years. The non-steel elements of the bridge, such as concrete piers and abutments, might have 25 years of life remaining, or more, even if no maintenance is performed. Scenario 1 is the easiest to understand and measure. If an asset is not performing up to stan- dard, for example, a guiderail that cannot withstand a required impact force or a sign whose retroreflectivity is below standard, then the potential justification for immediate replacement is understood. If an alternative is available that is less expensive than replacement, but offers fewer years of life than replacement, then its justification might be made based on funding availability or lifecycle cost analysis (Figure 3-3). Scenario 2 is more difficult to measure, because it expresses a risk of failure rather than observed failure. The asset might remain in satisfactory service for many years or it might fail the next day (Figure 3-4). With sufficient historical data from the manufacturer or from the agency’s internal records, the probability of failure might be quantified as a function of age and any potential preventive maintenance actions could be identified. The optimal replacement time then can be determined from a probabilistic analysis of its lifecycle cost. Figure 3-3. Extended life expectancy as a measure of project benefit. Figure 3-4. Failure at an uncertain time.

establish the Framework: how to Design Life expectancy Models 23 Situations such as Scenario 2 are considered for assets where sudden failure would be cata- strophic (e.g., a high-mast light pole might fracture and fall onto vehicles in traffic) or where mobilization costs to respond to isolated failures are high, relative to the cost of a replacement asset or component (e.g., traffic signal lamps or pavement markings). For Scenarios 1 and 2, the fact that an asset has already reached its life expectancy, makes it easier to use life expectancy as a performance measure for certain audiences and purposes. Compared to lifecycle cost, life expectancy may be easier for laypeople to understand. For elected officials, the ability to postpone expenditures to a point in time longer than the election cycle may appear to be a very tangible and relevant decision criterion. Scenario 3 is also more difficult to measure. In this case it is necessary to estimate remaining life for the asset under one or more scenarios of repair as well as a replacement asset. Each of these measurements has uncertainty. It is possible to use remaining life as a performance measure to justify investments; but, given that there are multiple estimates of remaining life, depending on current or future actions, and given that all such estimates are difficult to verify, the credibility and comprehensibility of asset life estimates may be jeopardized. In such cases, lifecycle cost becomes a more manageable performance measure to use instead of life expectancy. Scenario 4 presents even more complications that make it difficult to use life expectancy as a performance measure. In this case, there is a possibility of replacement of just a portion of an asset and other preventive maintenance or corrective actions may exist. Even in a do- nothing scenario, life expectancy is uncertain in Scenario 4. Future deterioration and future agency decisions have many sources of uncertainty, such as weather, traffic, and future bud- gets (Figure 3-5). In Scenarios 3 and 4, it is always useful to quantify life expectancy because this measure sets a time window within which any repair or rehabilitation actions may be considered and in which any benefits of such actions must be realized. However, life expectancy in this case is not used as a performance measure to quantify the benefits of the work. Instead, it is an intermediate result in an analysis where lifecycle cost and other more direct measures of performance (e.g., safety, resilience, travel speed, reliability, and comfort) are to be optimized. In contrast, for assets that have short or very predictable lifespans, life expectancy can be used not only as a measure of benefit, but even as a measure of current economic condition. If the average age of traffic signal controllers in a highway agency is 13 years, and the life expectancy for those assets is 15 years (i.e., 2 additional years), then this describes a relatively adverse eco- nomic situation where higher than normal replacement needs can be expected in the near future, compared to an inventory that is only, say, 5 years old. Figure 3-5. Portions of an asset with shorter life expectancy.

24 estimating Life expectancies of highway assets 3.2 Conceptualizing the Analysis The preceding sections described a top-down process that leads to the design of a life expectancy framework. The process starts with an understanding of the agency personnel and stakeholders who need the information and how they will use such information. The process continues with a concept for applications that produce the needed information and reports. This vision is refined using knowledge of the types of assets to be considered and their typical lifespans and typical agency actions. 3.2.1 Defining End-of-Life Life expectancy is the time between a given point in an asset’s life and a later time when the asset must be removed or replaced. Usually the starting point is the manufacturing date, the date when the asset is placed into service, the present date from which remaining life is measured, or the date of some future action or decision. The starting point can usually be determined with some certainty based on the purpose of the analysis. Determination of the ending point, however, often must be carried out with due circumspection. Here are some of the possibilities: • For an asset designed to fail suddenly, the date of failure. This definition would apply to such assets as lamps and motors (Figure 3-6, left side). • For an asset designed to become obsolete at a definite or identifiable time, the date when the obsolescence event takes place. This might apply to equipment whose support is discontinued as of a specified date or guiderails that will become obsolete when a new, stricter standard is adopted (Figure 3-6, right side). This is often referred to as the functional life of the asset. • For assets where obsolescence is directly defined by age, the time when the predefined lifespan runs out. For example, certain customer amenities in highway rest areas might be deemed to be out of style or “worn out” if their age exceeds 6 years (Figure 3-7, left side). • Certain assets whose life might be defined by condition may have their end-of-life defined by age or accumulated utilization instead if their condition is not routinely measured. For example, highway signs might be replaced at a given age, rather than by tracking retro- reflectivity and damage (Figure 3-7, left side). • For assets whose life is defined by utilization, life expectancy is the time when the utilization threshold is reached. This might apply to consumable materials and can apply to structural parts that are subject to metal fatigue (Figure 3-7, right side). • When an asset has a definite failure state but its failure would be catastrophic or the cost of responding to isolated failures would be high, end-of-life might be determined from a prob- ability distribution of lifespan data combined with a lifecycle cost model. When the cost of unexpected failure is high, the optimal replacement interval may be less than the median time to fail (Figure 3-8, left side). Fatigue life is an example. • When an asset does not have a definite failure state or where a condition of failure entails unacceptable safety or risk levels, end-of-life may be determined by defining terminal criteria for condition or other performance characteristics. This approach is typical of pavements and bridges (Figure 3-8, right side) and is often called structural life. • If portions of an asset can be replaced without replacing the entire asset, then it becomes relevant to define end-of-life in terms of the replaceable parts. This is especially true of con- structed facilities and of vehicles (Figure 3-9). • When an agency has methods of correcting end-of-life conditions or preventing them through maintenance activity, end-of-life depends on a calculation of the optimal application of such methods. Given that the lives of transportation assets cannot be extended forever, the end-of-life may be determined by physical characteristics, obsolescence, extreme events, or project inter- relationships that limit further use of corrective or preventive measures. For example, a bridge might be repaired and rehabilitated regularly until finally material degradation and traffic de- mand necessitate replacement by constructing a larger and/or stronger structure (Figure 3-10).

establish the Framework: how to Design Life expectancy Models 25 Figure 3-8. Additional end-of-life criteria. Figure 3-6. End-of-life criteria. Figure 3-7. Additional end-of-life criteria. • In the most general case where an asset has multiple performance measures, where the agency has corrective and preventive alternatives for preservation, and where uncertainty is modeled probabilistically, simulation methods might find the optimal life expectancy. It is often the case that the end of an asset’s life can be defined by more than one of the criteria described above. This is also the case at the network level when more than one of these criteria may combine to determine future replacement expenditure levels. The performance of highway assets relates to (1) operating characteristics, such as level of service (LOS), levels of safety, mobility, or congestion or (2) physical conditions, such as pave- ment serviceability rating (PSR), bridge health index, and road sign retroreflectivity. Interven- tions (e.g., repair, maintenance, rehabilitation, and reconstruction to highway assets) using warrants based on time intervals are easy to implement but may result in unintended delayed or

26 estimating Life expectancies of highway assets accelerated action. A performance-based warrant presents a superior alternative. Such a thresh- old can be set by one of the following ways: • Expert opinion • Historical records • Optimization Road agencies may combine more than one of these approaches. Bilal et al. (2011) proposed a general optimization framework to determine the optimal threshold for highway assets where maximum possible benefits of an intervention are achieved at the minimum possible cost. He determined that the optimal thresholds were sensitive to changes in user cost weight relative to agency cost and the user cost components used. Pavement and bridge management systems can often be used to establish these thresholds through lifecycle cost optimization. A common thread in these definitions is that, in most cases, end-of-life is certain only in the past. When evaluating an asset in service, its end-of-life depends on a decision about the optimal time to replace the asset, given anticipated deterioration and available life-extension actions. This determination is often referred to as economic life. As agencies become more mature in their asset management practices, they become more adept and sophisticated at finding the optimal life expectancy and in deploying life-extension methods. To evaluate the effectiveness of maintenance strategies, the following techniques are recommended: • Segment data. Calibrate models for each level of maintenance activity. • Incorporate as an independent variable. Evaluate the effect of maintenance through param- eters found in model calibration. • Add life extension on top of model prediction. Predict life based on a consistent level of maintenance or without maintenance; then add/subtract life extensions if known. For Markov chains, an improvement in condition state can be used to extrapolate a new life prediction. Figure 3-10. Planning component life based on functional life. Bridge condition Age End-of-life by matching component life spans End-of-life threshold Substructure rehab adds 10 more years, allows full utilization of the third deck Normal substructure life expectancy 50 years Normal deck life expectancy 20 years Figure 3-9. Planning end-of-life by coordinating the lifespans of components.

establish the Framework: how to Design Life expectancy Models 27 Comparisons of maintenance strategies can then be made on a lifecycle basis using the life expectancy estimate as the analysis period. With increasing use of automation and informa- tion technology, road agency databases on maintenance activities are becoming more enriched, enabling the inclusion of maintenance history in the explanatory variables to better model high- way asset performances. 3.2.2 Intervention Possibilities Many types of transportation assets are candidates, at certain points in their lives, for possible intervention actions that may extend their lives. The economic attractiveness of these actions may depend on their cost and effectiveness. The cost may depend on economies of scale, traf- fic volume (and traffic control measures), availability of equipment and labor, and contractual relationships. Effectiveness may depend on the availability of the materials used in the asset, the current condition of the asset, the weather, and the capabilities of the crew. When an agency has various intervention possibilities at its disposal, it is in a better position to optimize the lifecycle preservation actions for each asset. It is especially helpful to have alterna- tives that provide different increments of life extension at different costs. For example • Routine maintenance activities that prevent the onset of physical deterioration, such as wash- ing and sealing; • Repair and corrective actions that restore damaged protective systems or prevent acceleration of damage, such as painting and patching; and • Rehabilitation actions that replace deteriorated material or components to restore full func- tionality or stop damage progression. Timing plays a significant role in the attractiveness of an intervention for a given situation. For example, for urban highway sidewalk slabs, an agency might find that leveling of the slabs is too expensive to perform routinely as an alternative to replacement. But for a road that is to be widened in 5 years, leveling might be just enough to restore the facility to agency standards as a stop-gap measure. 3.2.3 Modeling Performance and Uncertainty Estimates of life expectancy depend on quantitative models of asset deterioration in terms of condition or performance. “Performance” in this case refers to the ability of an asset or group of assets to satisfy customer or stakeholder expectations. In order to select the right type of model for a given asset type and application, the distinction between continuous measures and discrete measures needs to be made: • A continuous performance measure is one that changes on a smooth scale, which can be bro- ken into meaningful increments of any size. Examples include the International Roughness Index (IRI), sign retroreflectivity, and traffic volume/capacity ratio. The NBI condition ratings do not fall in this category because the interval between two rating levels cannot be meaning- fully broken into smaller intervals; for example, there is no meaning for a rating of 8.5. • A discrete performance measure is one that changes on a step-wise scale, each level having a definition that may be independent of other levels. For example, a lamp is either functional or non-functional; sidewalk sections might be rated in terms of levels of service (e.g., a section at level A may have no tripping hazards of more than 1 inch in height); or bridge elements might be described in terms of condition states (e.g., a steel girder in condition state 2 may have paint that is peeling or chalking without exposure of metal). Figure 3-11 contrasts these types of measures. The mathematical differences between them are important for quantifying these models accurately with historical data.

28 estimating Life expectancies of highway assets When trying to forecast future condition or performance, another important distinction is between deterministic and probabilistic models. Figure 3-11 shows deterministic models, where the performance at any given point in time is assumed to be known with certainty. Figure 3-12 shows these model types when using probabilistic models. In a probabilistic model, at any given time, it is possible to predict more than one performance level. A continuous model, such as the left side of Figure 3-12, generally describes future perfor- mance using a mathematical function for the most likely value and another function to describe the uncertainty surrounding this value. A discrete probabilistic model, such as the right side of Figure 3-12, generally describes each condition state or service level as a probability of that level at each point in time. To keep the math simple, uncertainty in probabilistic continuous models is often quantified using a constant standard deviation or a standard deviation that increases with time. For discrete models, uncertainty is often quantified using a constant transition probability from one state to another state in one year. This type of model is called a Markov model. A common variation on the discrete probabilistic model is the case where there are only two possible states (e.g., operational versus failed), and the probability of each state varies with age. Figure 3-13 shows an example. This model is called a survival probability model. Chapter 4 will show that this type of model is especially useful for the simplest and most common types of life expectancy analyses. If a more sophisticated picture of probabilistic deterioration to non-failed states is required, as when analyzing life-extension possibilities or maintenance strategies, then a multinomial choice model such as an ordered probit model may be useful. In program planning analysis, uncertainty is very important. Figure 3-13 shows an analysis involving a population of signs. Based on median life expectancy for a cohort of signs, it appears that no funding for replacement will be needed during the 10-year program. However, when Figure 3-11. How performance changes over time. Figure 3-12. Probabilistic models of performance.

establish the Framework: how to Design Life expectancy Models 29 Figure 3-13. Role of uncertainty in program planning. uncertainty is quantified, it is found that 20% of the cohort will have failed by the end of that 10-year period. This result implies that funding will, in fact, be needed. 3.3 Determining Data Requirements From the analysis of the stakeholders and their information needs, it becomes possible to list the specific types of assets for which it would be useful to have life expectancy information. Then the agency can determine how the condition and performance of each asset type should be measured to enable performance management, definition of the end-of-life, selection of inter- ventions, and modeling of deterioration. For certain asset types, particularly bridges and pavements, the agency is likely to have data collection processes in place. In most cases, the existing data will be sufficient for life expectancy analysis. For other assets, where data are not already available, the agency should investigate whether the gathering of additional data is worth the expense. Given that the value of life expectancy analysis comes from the ability to make better decisions, one way to approach the estimation of the value of data collection, is to try to estimate the cost savings associated with improved decision-making, made possible by addi- tional data. As the previous sections showed, an accurate estimate of remaining life can help an agency to optimize life-extension activities, to find the right level of investment to minimize the lifecycle cost of each asset. Chapter 5 presents quantitative methods to apply life expectancy information in lifecycle cost analysis. By providing judgment-based estimates of model inputs, the analyst can prepare a pro forma lifecycle cost analysis using current decision-making methods and compare them with optimized methods using better data. To the greatest extent possible, the same level-of-service standards and end-of-life definitions should be used for both analyses. The difference in lifecycle costs would then be an estimate of the savings attrib- utable to improved data. To maximize cost savings, the agency should consider several strategies to minimize the cost of data collection: • Limit data collection to a representative, yet random sample of the asset type to be analyzed (Hensing and Rowshan 2005). If it is acceptable for some facilities to “fall through the cracks” and go unmeasured, then a sampling approach can vastly reduce the cost of data collection (Figure 3-14). • Use deterioration models to monitor intermittently the current condition or performance. A common practice among utility companies is to read the electric meter once every 2 or

30 estimating Life expectancies of highway assets 3 months and estimate usage for the intervening months. A similar approach can be used for asset data collection to reduce costs. • Develop models of replacement interval as a function of asset characteristics. In the best case scenario, this might enable a complete avoidance of routine condition surveys for certain types of assets. This is especially useful for cases where asset data collection is relatively expen- sive in comparison to replacement cost. • Increase the data collection interval for assets that are new or for other asset characteristics that are correlated with smaller changes in performance over time. For example, most bridges are inspected on a 2-year interval, but certain types of new structures can qualify for longer intervals—up to 4 years. • Consider the use of automated data collection methods whenever possible. Automated pave- ment surveys using vehicles that can collect useful data on roadside assets as well are very common (Figure 3-15). • Share data collection costs with other agencies to build economies of scale. State DOTs often perform data collection activities for local agencies to keep statewide costs as low as possible. Appropriate use of these data collection strategies can facilitate a meaningful life expectancy analysis, even for relatively minor asset types. Figure 3-14. Example of 10% section sampling. Figure 3-15. Example of automated data collection equipment (Hensing and Rowshan 2005).

establish the Framework: how to Design Life expectancy Models 31 Figure 3-16. Lifecycle cost analysis application used in Florida DOT. 3.4 Mocking Up Tools and Reports For efficient development of asset management applications, it helps to begin with a set of mockups. Spreadsheet software is an effective tool for rapid development and refinement of mockups of new software tools. Mockups can be converted to working prototypes by add- ing formulas to implement analysis equations, such as the calculation of life expectancy or lifecycle cost. Once end-users are satisfied with the mockups, the spreadsheet files can be used as models for the full software application. Figures 3-16 through 3-19 are examples. In each case the mockup evolved into a prototype, and then into the final application. The figures and examples included throughout this guide and in the software available on the TRB website can form the basis for many useful mockups for life expectancy analysis. 3.5 Gaining Buy-in and Building Demand An important reason for developing compelling mockups is the ability to use them to stimulate (1) agency interest in the study product and (2) demand for better information. Outside stakeholders, and even senior managers who are not technically inclined, might not realize the kinds of information that the agency would be empowered to produce using the study product. Even if stakeholders lack the interest or preparation to appreciate the analysis itself, they might find it easy to visualize how they would use a life expectancy report once they see one.

32 estimating Life expectancies of highway assets Figure 3-17. Resource allocation tool published in NCHRP Report 590. Often a successful implementation tactic for asset management tools is to prototype a small set of reports using a very simple version of the analysis, working around the data gaps that may exist. The product may be very rough at first and should be carefully labeled as such. Once managers and stakeholders develop a vision for better asset management, they are more likely to support the data collection and development work necessary to make the vision a reality.

establish the Framework: how to Design Life expectancy Models 33 Figure 3-18. Risk analysis report developed for Minnesota DOT.

34 estimating Life expectancies of highway assets Figure 3-19. Risk analysis report developed for NCHRP Project 24-25.