Computational Technique for Detecting Errors in Network-Level Pavement Condition Data

Abstract

Errors in pavement condition data can affect not only the assessment of current and predicted performance of the network but also the quality of pavement maintenance and rehabilitation plans. The large size of pavement condition databases makes manual error detection methods difficult to implement. This paper provides a new computational technique for detecting errors in network-level pavement condition data sets. The technique integrates conventional statistical methods and heuristics. The statistical methods are used to identify outliers in uniform performance families, and the heuristics are used for delineating potential errors from extreme, yet valid, behavior within these outliers. The new technique was validated with field data from Texas on pavement cracking. Of 9,166 audited pavement sections, 681 sections having potentially erroneous cracking data were detected. These potential errors are randomly distributed across the Texas roadway network and are not related to a specific geographic region, district, or climatic zone. Pavement engineers can use the presented technique to identify potential errors in large pavement condition data sets and to design more effective programs that ensure the quality of these data.