Integrating Bayesian Networks into Enhanced Bridge Management: A Data Fusion Approach

Bridges hold paramount importance in modern infrastructure, serving as crucial conduits for transportation and connectivity between regions. However, their prolonged exposure to diverse environmental and operational factors pose formidable challenges to their structural health and performance. In response, Structural Health Monitoring (SHM) has emerged as an invaluable tool, providing real-time data on bridge behavior and potential damage. Complementing SHM, visual inspections remain essential for identifying visible defects. Integrating the information from both SHM data and visual inspections through efficient data fusion techniques becomes imperative to maximize their combined potential. Among the diverse data fusion methods, Bayesian networks stand out as a promising solution for handling uncertain and incomplete information since they offer a principled approach to combine disparate data sources by capturing probabilistic relationships between variables. Despite their potential, however, their use in SHM is still in its infancy. In order to bridge this research gap, this paper proposes an original framework comprising the following key stages: (i) select the possible damage scenarios to be monitored by SHM; (ii) reproduce the selected damage scenarios in a numerical model of the bridge; (iii) perform the Bayesian Model Class Selection among the selected scenarios; (iv) update information on the Bayesian network (evidence and/or conditional probabilities), previously assembled with all the possible input variables (SHM, visual inspections and so on) for evaluating the risk of bridge failure. The proposed framework’s effectiveness is demonstrated by means of a case study bridge, showcasing the promising avenue for improving bridge condition assessment and maintenance practices.