Performance Modeling of Infrastructure Condition Data with Maintenance Intervention

Abstract

Accurate prediction of infrastructure condition is critical to infrastructure management agencies. Reliable and accurate predictions of infrastructure conditions save significant amounts of money for infrastructure management agencies through better planning of maintenance and rehabilitation activities. A good performance model relies heavily on the quality of the data collected and stored in the infrastructure management system. However, records of maintenance treatments are sometimes unavailable in the inventory. As a result, it is difficult to determine if a sudden increase in condition is caused by a maintenance intervention or by random measurement errors. Agencies with this problem usually filter the data set by removing possible outliers, subjecting the data to several quality control checks. For example, the change in condition may be considered a measurement error if the magnitude of the change is less than a given threshold. However, determination of the threshold value is sometimes arbitrary and may lead to bias in the development of performance models. A robust performance modeling technique was developed to address this problem. The proposed methodology captures the infrastructure deterioration rate without filtering of any data points. This method also calculates the probability that a data point is affected by maintenance intervention. Roughness data from the Long-Term Pavement Performance database is used to demonstrate the proposed methodology.