Explainable Artificial Intelligence for Understanding the Ageing Classes of Reinforced Concrete Bridge Components

Abstract

This article proposes an approach to the identification and interpretation of homogeneous ageing classes for reinforced concrete bridge components. The approach is articulated into three phases: in the first phase, homogeneous ageing classes are identified by considering the results of the visual inspections and the time sequence of condition states of the bridge components, applying a cluster analysis based on the k-means algorithm; in the second phase, the ageing class is predicted by means of a random forest algorithm, considering features of the bridge and of the components; in the third stage, the prediction is explained by applying a SHAP analysis. The results reveal that the prediction of the ageing class is influenced by the year of construction of the bridge and therefore of the component. This result opens up to a multiplicity of interpretations, which are considered in the article. The dependence of the ageing class on other variables is also discussed.