Predicting the Remaining Useful Life of Corroding Bridge Girders Using Bayesian Updating

Abstract

This paper developed a model for predicting the temporal failure probability of prestressed concrete (PC) highway bridges. The model updates predictions based on data from nondestructive testing and visual inspections. Chloride-induced corrosion is taken as the main cause for deterioration, and a gamma process describes the reduction in structural capacity. A nonhomogeneous Poisson process models vehicle arrival, in which the vehicle load variability follows a two-peak lognormal mixture distribution. After each inspection, Bayesian inference updates select model parameters and, consequently, the associated temporal structural resistance. Bridge managers can use the updated failure probability predictions to evaluate the remaining useful life of the bridge and determine the maintenance scheme and budget accordingly. A real bridge example illustrated the methodology and justified the probability distributions used for deterioration and vehicle load. Through comparisons with three existing methods, we argued that the proposed model provides more-conservative recommendations, yet existing methods tend to underestimate failure probabilities.