Reformulated Pavement Remaining Service Life Framework

Chapter 6. Performance Curves

INTRODUCTION

Performance curve"s are used as a means to predict the time when a pavement's condition will reach a construction trigger threshold. While the results of current pavement condition states can be used to respond to conditions, long-term planning of future needs and optimization requires the prediction of future pavement changes.

A key consideration in the development of performance curves is grouping pavement types into categories of pavement families. A pavement family is a group of pavement structures constructed with similar structural materials, construction methods, pavement components, and experience loading conditions and are expected to have a common set of distress mechanisms. The number of pavement families that should be used depends on the diversity in types of pavement structures with an agency's jurisdiction and the amount of time history data available for each defined family of pavements.

The best practice is to base performance curves on the analysis of pavement performance history observations. This requires the availability of uniform long-term time-series data on pavement condition that are linked to measured pavement features that permits application of mechanistic-based performance models. When empirical data are not available to formulate proper statistical models of future performance, expected performance curves based on the best available information can be used as a surrogate starting point to judge the relative performance of pavements. For example, the knowledge of engineers with long-term experience in a region can be used as a surrogate starting point. While the use of expert subject opinion can be used as a starting point for creating performance curves, the curves should be updated over time with field measurements to improve their accuracy and applicability.

MODELS BASED ON DESIGN EQUATIONS

Performance curves used for pavement design can be different than those used for pavement management. The performance curves used for pavement management can be based on empirical survivor statistics, which lack explanatory terms for differences in factors considered during pavement design. Pavement design performance curves require models that provide a measure of predicted future performance as a function of controllable design factors based on uncontrollable design requirements. Controllable design factors include man-made aspects of pavement structures, while uncontrollable design requirements include changes in climate, future traffic loads, and earthquake events. While pavement performance curves used for pavement design can be used for pavement management, pavement performance curves developed from pavement management data may not be appropriate for pavement design use since they may not be sensitive to controlled pavement design factors. A simple example is that a statistical analysis of time until pavement reconstruction is not sensitive to pavement thickness because all of the pavements included in the analysis had approximately the same thickness.

Models based on the 1972 AAHSTO Interim Guide for Design of Pavement Structures equations and subsequent updates have been used by some agencies.⁽⁸⁾ Pavement condition is expressed in terms of PSR or PSI, and pavement structure capacity as SN for flexible pavements and slab thickness for PCC pavements. Using this system, resurfacing or reconstruction is indicated by the level of PSR. When the predicted pavement PSR in an analysis cycle drops below a minimum tolerable condition based on highway functional classification, then resurfacing was indicated. Reconstruction was triggered if the PSR dropped below the lower reconstruction threshold if the section was not previously selected for resurfacing.

While the legacy AASHTO pavement design equations served the industry well by providing a basis for the design of the interstate system and other pavements in the United States, their use in pavement management has some limitations. Because PSR/PSI is used as a measure of pavement condition, it is not a good measure of what future construction treatments are required since it is most sensitive to pavement roughness. Updates to the 1972 AAHSTO Interim Guide for Design of Pavement Structures equations were made over the years by adding more terms to allow greater calibration to local conditions.⁽⁸⁾ These updates did not change the fundamental nature of using PSR/PSI as the primary measure of pavement condition.

FHWA has developed a set of performance curves based on the MEPDG for use in the Highway Economic Requirements System (HERS), National Pavement Cost Model (NAPCOM), and pavement health track (PHT) analysis tools. ⁽⁷⁾ Simplified models based on the use of the default level 3 MEPDG inputs along with the HPMS data are used to predict changes in the multiple pavement condition measures adjusted for current and past observed levels. The following distress prediction models are included in the tool:

• New hot mix asphalt (HMA) and HMA overlay of existing HMA.
  • Alligator cracking.
  • Rutting.
  • Transverse cracking.
  • Smoothness–IRI.
    
    
• New jointed plain concrete pavement (JPCP).
  • Transverse cracking.
  • Transverse joint faulting.
  • Smoothness–IRI.
    
    
• Composite–HMA overlay of existing PCC.
  • Reflection cracking.
  • Smoothness–IRI.
    
    

The use of specific types of pavement distresses as construction triggers at the network level allows for greater flexibility in assigning and determining the cost of future construction needs. This permits the application of automated rational decision tree logic to the selection of appropriate treatments based on multiple aspects of pavement condition.

The use of complex pavement performance prediction models, such as those used in the adaptation of the complex MEPDG pavement design models, is an example of how detailed research findings can be adapted to the management of pavement assets. The key to implementing these complex models, which are based on hourly changes in climate, traffic, and material properties, is to simplify them to the level commensurate with the intensity of data collection and sensitivity of business decisions.

EMPIRICAL MODELS

Empirical models are based on observations of events. All pavement mechanistic models based on the theoretical response of pavement structures to wheel loads and environmental factors need to be adjusted to fit field observations. The numbers of factors that have the potential to influence the performance of a pavement are greater than can be explained by available mechanistic models. While advancements are being made in pavement measurement technology, there will always be the need in pavement engineering practice to calibrate theoretical considerations to field observations.

The following list summarizes empirical approaches used to model pavement performance:

• Survivor curves: The various approaches used to develop survivor curves based on life table, Kaplan-Meier, or failure time theory approaches (see appendix B of this report) are based on the prevailing definition of pavement death as defined by each agency. These legacy survivor curves have been replaced with more modern engineering change of state models, which can be used to estimate future construction needs not properly reflected by the variability in past definitions of pavement death.

• Failure Cox Proportional Hazard (PH) model: This statistical model form is recommended as the basis for future developments in empirical performance curves for pavement management applications (see appendix B). The statistical approach appears suited to pavement performance management data. The Ohio Department of Transportation (ODOT) has used this approach to create a prediction model based on their pavement condition rating scale.

• Numerical regression models: Numerical regression models are used to find the best numerical functional form that describes the variation in the relationship between variables being compared. Regression models only describe the numerical relationship in the available observation data and are most applicable to that inference space. Extrapolation beyond the inference space should include an estimate of the magnitude of variability in the prediction.

• Bayesian statistical updates: Updating prior models with new data can be based on Bayesian statistical concepts. A key component in Bayesian statistics is the weight given between prior and posterior observations.

• Neural network models: Neural networks are models inspired by the way biological nervous systems, such as the brain, process information that have the capability to fit the variations in observed data that is not possible by using traditional numerical regression techniques. Research to extract information from neural analysis similar to that inferred by numerical regression-based model forms is being pursued.

Some measure of variability distribution for the function and its relevant factors should be contained in all predictions of future pavement condition states. These measures can be used to accumulate future risk probabilities in the prediction based on the variability in the input parameters and prediction models. The FHWA LCC models are an excellent example of how variability can be propagated into the final decision tool.

AGENCY TIME-BASED RULES

The simplest performance curve is a time-based rule for future construction treatment. This type of rule does not require investment in field measurement devices and custom computer programs to predict future events. However, this approach does not provide a basis for optimization of utilization of constrained agency resources.

An example of a time-based rule is that an agency decides that some type of corrective construction treatment will be applied on its rural interstates every 8 years. The network is segmented into contiguous projects, and a construction rotation sequence is established. An appropriate correction strategy is selected and applied during the target construction year. Thus, the performance curve for the time until the next construction event is the time remaining until the next target construction year.