The impact of clustering in the performance prediction of transportation infrastructures

Abstract

In the context of transportation infrastructures management, bridges are a critical asset due to their potential of becoming network’s bottlenecks. Unfortunately, this aspect has been emphasized due to several bridge failures, occurred in the last years worldwide, resulting from climate change-related hazards. Given this, it is important to establish accurate tools for predicting the structural condition and behavior of bridges during their lifetime. The present paper addresses this topic taking into account one of the statistical models most used and generally accepted in existing bridge management systems—Markov’s stochastic approach, which is further described. These statistical models are highly susceptible to the data that feeds them. Quite often, the step related with data cleaning and clustering is not properly conducted, being the most commonly available data sets adopted in bridge’s performance prediction. This paper presents a comparative analysis between different performance predictions. The only different between consecutive scenarios corresponds to the subset of bridges database used in each analysis. It was found that the development of good data clusters is of utmost importance. Contrarily, the use of poor clusters can lead to deceiving results which hinder the actual deterioration tendency, thus leading to wrong maintenance decisions.