|<< Previous||Contents||Next >>|
Highway Quality Compendium
A Brief History of Highway Quality Assurance
by Richard M. Weed
One of the nation's most valuable assets is the network of roads and bridges linking suppliers of goods and services with customers. The nation's wellbeing depends on the highway system's condition, which in turn relates to the quality of construction.
Highway quality assurance has evolved over approximately four decades and encompasses all the programs and procedures for controlling and accepting construction quality. For the most part, the procedures in use today are fair and effective, but that was not always the case. As a former statistical engineer with the New Jersey Department of Transportation (DOT), I spent most of my career in quality assurance; following are some of the more important lessons learned.
The first of these lessons occurred while I was studying for a civil engineering degree. The lesson was taught not by one of my professors, but by a highway inspector who had few academic credentials. I was working in the summers on highway construction for New Jersey DOT when one of the inspectors had an interesting idea: "Let's send two identical samples to the department laboratory to see if they come out the same."
We carefully prepared two samples as nearly alike as possible and sent them to the laboratory. I do not recall the exact results, but they differed considerably more than we had expected. That was my first exposure to the real world of variability, and I sensed that this must be an important aspect of engineering.
Today we understand that there are several possible explanations for differences between tests of identical samples. Maybe the samples were not as identical as we thought; maybe the samples were handled differently during transportation; or maybe the samples were tested by different operators, or on different testing equipment, or on different days. But despite the potential sources of variability, samples of this type are used routinely to make important decisions about the acceptability of the construction items they represent. If this ever-present variability causes substandard work to be erroneously accepted, performance problems will arise that are likely to prove both costly and inconvenient. If satisfactory work is mistakenly rejected, completion of the project is delayed, the contractor is treated unfairly, and the result may be increases in future bid prices. Obviously, we need to minimize both types of mistakes.
Road Test Results
At roughly the same time I became acquainted with the realities of variability, the highway profession was learning a similar lesson from the American Association of State Highway Officials (AASHO) Road Test. This elaborate experiment alerted everyone that highway construction was far more variable than anyone had realized and, in some cases, was of lesser quality than anyone had recognized.
The reports from the AASHO Road Test used statistical measures to describe construction quality, and a few engineers saw that these same measures might offer a better way to specify what was desired than did the materials-and-methods specifications then in use. Not only would a statistical approach afford greater freedom to the construction industry to use its considerable skills and innovative abilities to achieve the desired results, but the approach also would provide a valid, quantitative way for highway agencies to judge the acceptability of the finished product.
The approach also would offer legal advantages, because in some cases, courts of law had not allowed highway agencies to reject defective work over which the agencies had exercised primary control via materials-and-methods specifications. Another advantage would be the creation of valid databases that eventually could improve understanding of the relationships between construction quality and ultimate performance.
This new approach of basing construction specifications on statistical concepts clearly was a win-win situation for all concerned. As engineers gained familiarity with statistical techniques, the use became more frequent and more effective. Growing pains were inevitable, but these early efforts turned out well enough that within a few years many other highway agencies had followed suit.
Analyzing the Risks
One of the most significant realizations from this early work was that the analysis of operating characteristic (OC) curves and of expected payment (EP) curves was an indispensable part of statistical quality assurance. Only through the study of these curves can two critical risks be known and controlled at suitably low levels: the highway agency's risk of accepting defective work, and the contractor's risk of having good work penalized or rejected.
This offers both technical and diplomatic advantages. The correction of faulty specifications in the office before reaching the field greatly increases the likelihood of making good acceptance decisions. Assuring that statistical specifications perform correctly and fairly greatly improves the working relationship between the highway agency and the construction industry.
Statistical Quality Measures
The first specifications of this type applied simple statistical measures, often the mean-or average-of the test values. As more construction data became available for analysis, engineers realized that the mean by itself was not always an adequate predictor of performance. Two lots of material having the same mean might have markedly different levels of variability and, consequently, substantial differences in the amounts of substandard material and in the expected levels of performance.
The next step was to look for statistical quality measures that would take variability into account. The moving average was out-it was as insensitive as the mean was to variability. In addition, the moving average was influenced by adjoining lots of material, making any type of risk analysis extremely difficult.
A few agencies tried average absolute deviation, which has never been studied thoroughly as a formal statistical measure and is not well suited for singlesided specifications for which a unique target value cannot be defined. The conformal index also was proposed, but the drawbacks are essentially the same as those of the average absolute deviation.
This left as the logical choices percent defective (PD) and percent within limits (PWL)-which are different representations of the same thing. PD/PWL is a standard statistical measure, extensively studied, known to be an unbiased estimator, capable of handling single-sided and double-sided applications, and with published tables for use. For these reasons, PD/PWL continues to have the strongest intuitive appeal to most writers on statistical quality assurance.
Another key milestone in the development of highway acceptance procedures was the advent of bonus provisions. The earliest statistical specifications either paid full price or assessed some degree of pay reduction, depending on the deficiency in quality. Highway engineers eventually realized that if withholding payment for substandard work made sense, offering some degree of monetary incentive for superior work also made sense. The idea was to encourage and compensate contractors whose attention to quality control produced work that substantially exceeded the specified levels of quality and, as a result, could be expected to provide above-average performance.
Several arguments support an incentive approach. Once OC/EP curve analyses became more common practice, some degree of bonus provision was recognized as necessary for the long-term average pay factor to be 100 percent for work exactly at the level defined as acceptable. The natural variability of statistical measures often produces quality estimates that are either too low or too high. Bonus provisions allow the resulting underpayments or overpayments to balance in a way that turns out to be fair and equitable.
Other benefits of bonus provisions include motivation for higher quality work, improved relations with the construction industry, and the likelihood that better contractors more often will be the successful bidders-because contractors more assured of receiving bonus payments can afford to bid lower. Because of these benefits, a substantial majority of highway agencies now use bonus provisions in one form or another.
A goal in highway specification writing is to relate basic engineering properties-for example, the resilient modulus of pavement-directly to performance, so that specifications only state appropriate levels of appropriate properties. That goal remains elusive, however, and efforts have focused on developing performance-related specifications (PRS) based on mathematical models linking quality characteristics-such as air voids in asphalt concrete or the compressive strength of portland cement concrete-or statistical quality measures, such as PD or PWL, to performance and longevity. Typically, these specifications include pay schedules developed through life-cycle cost analysis.
PRS developmental efforts have produced a dichotomy of approaches. On the one hand, highly complex national studies have produced sophisticated computer programs like HMASPEC and PCCSPEC, based on mechanistic design principles, life-cycle cost analyses, and various decision-making processes. On the other hand, a few state transportation agencies, including New Jersey DOT, are engaged in grassroots efforts to use their own data to create simplified mathematical models with the same underlying scientific principles.
The methods developed by the national studies offer the potential for greater precision and accuracy, but at the expense of considerably greater data requirements and complexity. The grassroots models are more empirical, but their simplicity and ease of being tailored to local conditions make them attractive from a practical standpoint. States that would like to convert statistical specifications to actual PRS will have to decide which of the two profoundly different approaches to take. The optimal approach may lie somewhere between these two extremes.
Simple but Scientific
Much has been accomplished in the field of highway quality assurance, but much remains to be done. A slight variation of the KISS rule has served New Jersey DOT well: Keep It Simple but Scientific. The guidance may be useful to other agencies as they continue to advance the state of the art of PRS.
In other words, start with the simplest approach that makes scientific sense, and switch to something more complex only if there is evidence or data showing that the simple method is not working. As a statistical practitioner always concerned about the accumulation of error in any complex system, I advocate this practical approach for designing any engineering process.
The author, who retired from the New Jersey Department of Transportation in 2002, is a full-time consultant based in Trenton. He is an Emeritus Member of the TRB Management of Quality Assurance Committee.
From TR News, November-December 2005, pp. 30-32. Copyright, Transportation Research Board (TRB), National Research Council, Washington, D.C. Reprinted with permission of TRB.
|<< Previous||Contents||Next >>|