Bayesian Updating of Bridge Condition Deterioration Models Using Complete and Incomplete Inspection Data

Abstract

Estimation of the time that a bridge or bridge component stays in a specific condition can guide decision-making on bridge maintenance and preservation. Statistical models of the time-in-condition rating (TICR) for bridges or bridge components are good candidates for this purpose. Typically, these models are calibrated using existing inspection data. Current practice tends to trim a large portion of the data that are deemed incomplete. However, there is actually a lot of useful information in these data (e.g., lower bounds for TICR), which should also be incorporated to establish better estimation of TICR. To address this, within the Bayesian framework, this paper proposes the adoption of a modified likelihood function to explicitly incorporate both complete and incomplete inspection data for model calibration. In addition, Bayesian model class selection is used to select the most appropriate models out of several candidate statistical models. The proposed approach is applied to establish TICR models for different types of bridges and bridge components in Colorado using National Bridge Inventory (NBI) data. The results and comparisons show the importance and necessity of explicitly incorporating incomplete inspection data in the model calibration and class selection for bridge condition deterioration models.