The development of performance prediction models for Virginia's interstate highway system.

35681409Final report.PDFTech Reporthttp://ntl.bts.gov/lib/36000/36900/36983/96-R8.pdfVTRC 96-R7VTRC 96-R8VirginiaVirginia Transportation Research CouncilSadek, Adel WFreeman, Thomas EDemetsky, Michael JVirginia Transportation Research CouncilUS Transportation CollectionPerformance prediction models are a key component of any well-designed pavement management system. In this study, data compiled from the condition surveys conducted annually on Virginia's pavement network were used to develop prediction models for modeling the interstate system. The study is being reported in two volumes. Volume I describes the task of preparing the data base for model development. At the onset, several problems challenged the modeling effort: a data base containing non-homogeneous sections unsuitable for use in modeling, a user-unfriendly system incapable of efficient data manipulation, and missing and incorrect data. A methodology was devised to address these limitations, involving the development of a number of computer programs to process, merge and screen the data files. In addition, missing data items were secured from external sources and added to the data base. The problems encountered during this phase of the study suggested some desirable pavement management system features that would make prediction model development easier and more accurate. The second volume describes the development and evaluation of the performance prediction models. An exploratory data analysis was first conducted to examine the data distribution, and to reveal the underlying relationships among the variables. "Robust" regression techniques were used to identify outlying observations that could adversely affect the regression analysis results. Stepwise regression was then used to select the significant predictors of deterioration. Different model forms were examined to identify the most suitable for fitting the data. The models were evaluated by checking their goodness-of-fit statistics and conducting a series of sensitivity analyses. To further assess the models' accuracy, their predictions were compared against field-observed values. An analysis-of-variance (ANOVA) test was also conducted to compare between the accuracy of two model forms and two model adjustment procedures. In general, the developed models provided an adequate fit and generated predictions that conformed with accepted engineering judgment. Comparisons with field observations showed their accuracy to be quite reasonable even for long-range predictions. Finally, the ANOVA results indicated that no significant differences existed between the two model forms tested or between the two adjustment procedures