Multilevel Competing Risks Models for the Performance Assessment of Transportation Infrastructure

Natural disasters such as hurricanes, earthquakes, storm surges, wildfires among other hazards affect communities and large geographic areas of the United States resulting in negative repercussions on the environment and the economy. The impacts of these hazards on bridges and other civil infrastructure affect the structural integrity and functionality of bridges, highway pavements and overall efficiency of the transportation network. This study focuses on the hazards that affect bridges and pavements, and the complex interactions and correlations among them, to evaluate the performance of civil infrastructure. Hazard scenarios are considered as competing risks impacting the health of bridges and highway pavements. The study derived stochastic distributions characterizing the behavior of bridge elements and pavement segments during natural deterioration process which are compared to their response in the presence of hazards. To achieve the above objective, competing risk models were developed for highway pavement in Florida in the presence of hurricane failures. Also, distributions and competing risk deterioration models for AASHTO Commonly Recognized (CoRe) bridge elements were developed using legacy data for bridges from the Florida Department of Transportation (FDOT). Annual probability of hazard occurrence data sourced from Federal Emergency Management Agency (FEMA – HAZUS) was employed to model hurricane induced pavement and bridge element failure. The expected service lives for highway pavement and bridge elements, transition and sojourn times from one condition state to another were obtained using the Cox Proportional Hazards, cumulative incidence functions, product-limit survival estimates and other survival functions. The method of likelihood estimation, weighting techniques and inference procedures were used to describe risk event data with censoring and truncation scenarios where necessary for the analyses. The cumulative incidence function and the Kaplan – Meier estimates were used to ascertain the effects of the modes of failures of bridge elements and highway pavements at the network levels in the presence of hurricanes. The results showed that three modes of failure (cracking, riding and rutting) are all significant to for pavements. As the roadway pavement section ages, the chance of failure is more likely to be due to cracking than the other competing modes. Based on the road functional classifications, the survival probabilities and the cumulative incidence estimates showed that the cracking defect was predominant on both interstate and non-interstate roadways. It was observed that urban and rural pavements deteriorated by the cracking and riding defects with the rutting failure mode being significant at the end of the service life of the pavement. The research also evaluated the significance of two competing risks events: “natural” crack deterioration of highway pavements in the presence of hurricane failure (Hurricane Categories 1, 2 and 3), for 6702 highway pavement sections using the nonparametric survival probability (Kaplan-Meier estimates) and the cumulative incidence function (CIF). The risks were compared using the Logrank Test (to indicate if the survival probabilities of the risks are significantly different), and the hazard ratio (ratio of hazard rates based on time to failure covariate). From the results, it was observed that the contribution of the Hurricane Category 3 as a competing risk was significantly higher and different from that of crack deterioration. For example, the hazard ratio indicated the effect of Hurricane Category 3 on pavement failure was twice as significant as that of the crack deterioration for the inland urban interstates roadways. Also, the hazard ratio between hurricane category 3 and crack deterioration was about 16 for rural interstates and 18 and 28 for urban non-interstates and rural non-interstates at the coastal locations respectively. The hazard ratios and CIF plots showed that impact of hurricanes on coastal roadways is more significant compared to how they affect the inland pavements. Finally, it was observed that the “natural” deterioration of bridge channels and hurricane induced channel failures generally yield significantly different impacts based on the logrank chi-square outputs. Also, it was observed the impact of hurricane categories 3 and 2 on bridge channel elements were more significant (based on the hazard ratios) at the coastal bridge locations than in the noncoastal areas, and also generally high for urban bridge channels compared to rural channels.