Identification of influencing factors on bridge damages using Bayesian network

In Japan, bridge inspections are compulsorily performed in 5‐year cycles. With the institutionalization of the inspection cycle, essential data have been continuously accumulated. However, effective data utilization requires trend analysis and causal analysis for a group of bridges. In this study, a method for determining factors affecting deterioration is established. The analysis is performed for concrete and steel bridges with Bayesian networks by utilizing data on bridge inspection and repair, and open data such as traffic census and rainfall. For concrete and steel bridges, the target members are the deck slab and main structural members, whereas the damage type is “Delamination/rebar exposure” and “corrosion,” respectively. The validity of the selected explanatory variables is verified by crossvalidation using separately prepared test data; evidently, the maximum damage rating prediction accuracy is 86%. Furthermore, the influencing factors extracted in this study are reasonable for the two damages, thus indicating the possibility of probabilistically extracting influencing factors for specific damages by Bayesian networks.