Bridge Clustering for Systematic Recognition of Damage Patterns on Bridge Elements

In this paper, bridge clustering for the systematic recognition of damage patterns on bridge elements is described based on massive data sets of bridges. To achieve this primary object with the data from the Korean Bridge Management System (KOBMS), the research used the following procedures: (1) bridge clustering according to general, structural, and environmental characteristics that cause similar types of element damages by using a clustering algorithm; and (2) statistical investigation to extract element damage patterns on a cluster-by-cluster basis for evaluating clustering results. The case study for the purpose of validation was performed using the data sets of 1,944 prestressed concrete I-shaped (PSCI) bridges with 64,815 inspection records of the superstructure elements. Based on the clustering results, there were significant differences in the frequency of damage to bridge elements and the influence of the bridge characteristics on each damage. The predictive accuracy of damage occurrence was also improved from an area under the receiver operating characteristic (ROC) curve (AUC) value of 0.655 to an average AUC of 0.745 after clustering. The outcome of this research shows that it has potential to be applied to identify damage patterns on bridge elements in complex bridge data sets and estimate future changes in the conditions of bridges for preventive maintenance.