Design of a Long Term Pavement Monitoring System for the Canadian Strategic Highway Research Program

EXECUTIVE SUMMARY

This report has a number of objectives. In the narrowest sense, it is a report detailing how to measure the life cycle cost of a particular pavement section given the information obtained from long term pavement performance monitoring. In a broader sense, this report sets forth an integrated, comprehensive pavement monitoring, performance prediction, and pavement management system. In particular, we have endeavored to begin with the Long Term Pavement Performance (LTPP) monitoring gameplan of the Strategic Highway Research Program (SHRP) and augment and modify it so that it meets the needs of the Canadian Long Term Pavement Performance (C­LTPP) project of the Canadian Strategic Highway Research Program (C­SHRP). Our objective is to develop for Canada an immediately workable system that can be used to manage Canadian pavement assets effectively beginning immediately and continuing into the future.

In developing a workable gameplan for Canada, we faced two alternatives:

1. Extrapolate the SHRP analytical approach (a "regression" approach) to the Canadian situation,

or

2. Customize an analytical approach specifically to Canadian needs.

With regard to the former alternative, Canadian resources are limited both absolutely and relative to SHRP. In our judgment, application of limited Canadian resources incrementally at the margin of the SHRP­LTPP program would add little to either the Canadian or United States programs. Doing slightly more of what SHRP will already do is a limited benefit/limited limited cost alternative. Almost surely, Canadian resources would have little if any impact at the margin of SHRP.

By contrast, customizing an analytical approach directly to Canadian needs, borrowing from SHRP­LTPP as desirable, represents a more attractive alternative. While there is higher prospective risk in the customization alternative, there is also higher prospective benefit to be realized by C­LTPP's constituents. In view of the virtual certainty of low return to C­SHRP from activities incremental to SHRP, C­LTPP chose the customization alternative, a higher risk but higher reward alternative. This report outlines the customized analytical approach we have developed to support long term pavement performance monitoring for Canada. We believe it to be not only a major contribution to Canada but also a major extension of the SHRP­LTPP analytical framework in the United States.

In designing a Canadian­specific analytical program for C­LTPP data, we began by assessing the analytical strengths and weaknesses of the SHRP­LTPP program from a

Canadian perspective. The program outlined in this report retains certain portions of the SHRP­LTPP analytical design but eliminates key limitations. In developing a customized approach for C­LTPP, it has not been our intent to denigrate or criticize the SHRP­LTPP program; rather, it has been our intent to "high grade" the SHRP program as we develop a customized long term pavement performance analytical program to meet Canadian needs. We acknowledge the contributions SHRP has made, and we believe we have added substantially to the analytical work SHRP has already accomplished.

In designing a Canada­specific analytical framework for long term pavement performance monitoring, we have found the following:

1.0 THE SHRP REGRESSION APPROACH UPON WHICH SHRP­LTPP IS BASED WILL NOT MEET CANADIAN LTPP NEEDS

· C­LTPP sample sizes are small.

· C­LTPP data promises to be statistically difficult, meaning that effective sample sizes will be even smaller than anticipated and extraordinary steps will be required.

· C­LTPP must generate definitive results beginning immediately. C-LTPP cannot wait five, ten, or fifteen years for data to be gathered and results to be obtained.

· C­LTPP wishes to systematically evolve away from conventional practice toward new practice indicated by long term pavement performance monitoring. Canadian pavement decision makers will not be willing to discard or ignore what they already know in favor of C­LTPP data, as the SHRP­LTPP analytical design would imply.

· Our analytical design will eliminate these problems endemic in the SHRP analytical design.

2.0 SMALL SAMPLE SIZE PROBLEMS WILL SYSTEMATICALLY BESET LONG TERM PAVEMENT PERFORMANCE MONITORING

· There are many variables and not enough measurement sites.

· There are bound to be many unmeasured (and unmeasurable) variables.

· There are bound to be substantial differences across sites and across time.

· The information content of the C­LTPP samples will be much smaller than anticipated because of inherent statistical difficulties including autocorrelation, heteroskedasticity, multicollinearity, and omitted variables.

· Classical statistical techniques (e.g., regression analysis) work poorly with small sample sizes:

- Regression systematically ignores prior knowledge.

- Regression will give the wrong answer, in part because it ignores prior knowledge.

- Regression cannot give reliable results for many years to come.

· Bayesian statistical techniques proposed here will circumvent difficulties with regression analysis.

3.0 THE SHRP STATISTICAL DESIGN IS SUSCEPTIBLE TO STATISTICAL BIAS

· Autocorrelated errors are highly likely to be present in the C­LTPP data base.

- Autocorrelation will devalue measurements and exacerbate small sample size problems.

- Sophisticated methods are required to discriminate systematic dynamic trends from random noise or measurement error.

- Use of such methods expends a portion of the data base.

- Initiating measurement part way through the pavement life cycle has extremely low value when autocorrelated errors are present. This statement is true even for pavements brought to a common starting point through rehabilitation.

· Heteroskedastic errors are highly likely to be present in the C­LTPP data base.

- Heteroskedasticity will devalue measurements and exacerbate small sample size problems.

- Sophisticated structural methods are required to mitigate heteroskedasticity.

- Heteroskedasticity cannot be eliminated altogether, but it must be mitigated.

· Missing variable bias is likely to occur.

- There are myriad potential contributory variables.

- Omitted contributory variables may be correlated with included variables, misstating results.

· Multicollinearity (inadvertent, independent measurement of the same quantity) could be present.

- Multicollinearity devalues data base.

- Multicollinearity is automatically detected by regression methods, but much of the predictive power of the data is lost.

4.0 UNCERTAINTY IS INTRINSIC NO MATTER HOW MUCH MONITORING IS INITIATED

· No amount of pavement monitoring will eliminate uncertainty. The best C­LTPP can hope for is systematic, targeted reduction of uncertainty. In practice, the Bayesian and classical regression solutions will never converge, even though theoreticians might cite conditions for such convergence. In practice, there will never be enough data to achieve consonance between the Bayesian and classical regression approach.

· Pavement design and maintenance decisions must be made in the face of uncertainty.

· Pavement management systems must be able to deal fundamentally with uncertainty and the benefits of reduction of uncertainty if they are to reap the full benefits of long term pavement performance monitoring.

5.0 REGRESSION COMPLETELY IGNORES PRESENTLY EXISTING INFORMATION, CONVENTIONAL WISDOM, AND PRACTICE

· With regression approaches,

- The only data deemed to be relevant is data that has come from long term monitoring.

- Everything previously known or data gathered by other means is "thrown away."

- Pavement decision makers in the field would be asked to balance C­LTPP information against conventional wisdom and practice completely without guidance.

· We doubt the efficacy of regression approaches with decentralized decision makers in the field.

6.0 FUNDING AGENCIES ARE UNLIKELY TO WAIT PATIENTLY FOR TEN OR FIFTEEN YEARS FOR LONG TERM MONITORING RESULTS

· Long term pavement monitoring will inevitably be viewed as a "scientific expedition."

- Results are not palpable or observable.

- Long term programs with nonobservable results such as C­LTPP are especially expendable under the budget axe.

- The tenor of regression analysis is: "Give us enough time andmoney and we will discover true science." This tenor reinforces the scientific expedition image.

· C­LTPP requires proactive participation from voluntary sponsors.

- Long term pavement monitoring must deliver definitive benefitsimmediately, i.e., in the short run.

-Funding members must perceive benefits, otherwise they will withdraw, further exacerbating the small sample size problem.

7.0 LONG TERM PAVEMENT MONITORING MUST BE TIED TO A PAVEMENT "BOTTOM LINE'

· Otherwise it will not

- assemble the right data,

- manipulate that data in the right way,

- deliver results that will be embodied in real­world pavement decisions.

· Long term pavement data only has value when (and if) it is embedded in real world pavement decisions.

· C­LTPP needs a fully integrated monitoring, statistical, and pavement management system to guide its monitoring activities.

· "How will we know when we get there if we don't know where we're going?" (Lewis Carroll, Alice in Wonderland.)

8.0 C­LTPP RECOGNIZES THAT PAVEMENT DESIGN AND MAINTENANCE ARE FUNDAMENTALLY DECENTRALIZED DECISIONS

· There is no dictator.

· People cannot be compelled to use long term pavement monitoring information in any particular way.

· Long term pavement monitoring results will at best be inserted gradually over time into

- conventional wisdom and practice

- quantitative pavement management procedures

· C­LTPP should strive to develop consistent "signals" that encourage optimum behavior by decentralized pavement decision makers.

9.0 USER COSTS AND BENEFITS LIE AT THE HEART OF PAVEMENT MONITORING AND MANAGEMENT

· Pavement design and maintenance must not be viewed as purely engineering or technical problems.

· They are user service problems.

· Pavement must be viewed as an asset to serve users in the most cost effective fashion, taking into account

- user costs and benefits as well as

- agency costs and benefits.

· You get entirely different answers when you include user costs and benefits than when you ignore them.

· C­LTPP should give strong emphasis to user costs and benefits as well as agency costs and benefits.

This report details an approach to long term pavement performance monitoring that overcomes the foregoing nine difficulties. The approach, based on Bayesian statistics rather than classical multivariate regression analysis,

· Allows definitive results to be delivered from the first day of the program.

· Explicitly deals with the small sample size problem.

· Balances new knowledge gained from pavement monitoring against preexisting conventional knowledge, arriving at the appropriate balance point.

· Provides definitive quantitative information to deterministic as well as probabilistic pavement management systems.

Section 1 of this report outlines some of the key aspects of our system design. Section 2 presents a detailed mathematical exposition of our proposed Bayesian approach. Section 3 indicates how the Bayesian approach can be interfaced with deterministic and probabilistic pavement management systems. Section 4 presents a proposed gameplan for implementing an overall integrated pavement monitoring, performance prediction, design, and maintenance framework that can support decentralized pavement decision making in the field as well as at agency headquarters.

(Continue)

Return to Table of Contents

Return to Main Page