Dynamic Performance Models for Transportation Infrastructure Management

Chih-Yuan Chu

doi:https://doi.org/10.21985/N2X709

Work

Dynamic Performance Models for Transportation Infrastructure Management

Public Deposited

Download PDF

Download All Files (.zip)

The abundance and breadth of condition data generated by advanced inspection technologies have increased the frequency and accuracy of data collection, which is not exploited by the existing performance models. Moreover, simultaneity of deterioration and maintenance causes difficulty of estimating effectiveness of maintenance activities for existing performance models. To address these problems, we present state-space specifications of time series models as a framework to formulate dynamic performance models for transportation facilities, and to estimate them using panel data sets. A dynamic performance model, which considers serial dependence, has two important implications. First, it is capable of estimating major types of effects of maintenance activity as exogenous variables. As a result, the study supports maintenance decision by developing infrastructure performance models that drive any maintenance optimization model adopting time series for prediction. Second, the capability of updating forecasts in response to new information makes the periodic inspection beneficial. Moreover, the dissertation is the first to extend the state-of-the-art latent performance approach into dynamic modeling since state-space specification also accounts for the uncertainty in the deterioration and measurement processes when multiple inspection technologies are used. From the methodological point of view, the framework provides a flexible approach to simultaneously capture the effect of serial dependence and of exogenous factors, while controlling for individual heterogeneity when pooling data across the facilities that comprise the panel. The ensuing performance models capture effects that are not identifiable in either pure time series or pure cross-section data. Furthermore, we consider three specifications of models, which differ in the assumptions regarding the structure of the underlying mechanisms generating the data sequences. Thus, the framework is applicable to different transportation infrastructure systems. The methodology is validated with three case studies. The results indicate that serial dependence is indeed significant. We also compare the three specifications to assess the poolability of pavement condition data. The results provide evidence that unobserved heterogeneities among the facilities are present in the panel. Finally, the dissertation conducts a comprehensive, quantitative comparison of the representative performance models in the literature and the estimated dynamic models in the dissertation

Last modified