Evaluation of Concrete Pavement Performance Model Considering Inherent Bias in Performance Data

Abstract

Concrete pavement performance models are evaluated, calibrated, and validated using field data from databases like the Long-Term Pavement Program. Conceptually, performance model predictions should match field data considering a reliability of 50%; that is, there is a 50% probability that the predictions of a certain distress indicator are higher or lower than the field data. However, modern pavements are designed for higher levels of reliability (usually 90% to 95%). Local performance model evaluation for higher levels of reliability requires a high amount of field data that traditional databases lack. Pavement management system (PMS) databases can be a useful resource for high reliability model analysis because of the large amount of data collected locally and regularly. However, when selecting and filtering field databases (of any source), the effect of censored performance data due to rehabilitation, removal from service, or modification of pavement sections is usually ignored. This paper proposes an approach for the use of PMS databases accounting for censored performance data to evaluate the accuracy of performance models’ high reliability predictions. The approach is exemplified using a PMS transverse joint faulting database. Results show that by addressing the “survival issue,” i.e., accounting for the censored performance data, the resulting PMS-based reliability model improves the faulting model accuracy in matching the field data for high reliability levels.