Integrating Semiparametric and Parametric Models in Survival Analysis of Bridge Element Deterioration

Abstract

The central challenge in developing deterioration models for infrastructure facilities based on field data, is the right selection of base models along with the specific nature of the available condition data. A parametric model is extensively used in infrastructure deterioration modeling. However, the pre-assumed hazard shape of parametric models is questionable because of the lack of explanations of the underlying assumption. Cox proportional hazard models have the advantage that they make no advance assumption about the shape of the hazard function and support multivariate analysis; and are thus proposed to solve the problems stemming from the predefined distributions of the parametric models. However, the basic uncalibrated functions of Cox models can appear complex and inapplicable for maintenance management applications. Therefore, it is considered useful to incorporate both semi-parametric and parametric models in an integrated strategy for modeling survival data. The development of an integrated methodology to predict the distribution of durations between changes in condition states by incorporating both Cox regression analysis and parametric models for concrete bridge elements, is presented in this paper.