Bridges are essential for maintaining functional transportation systems, and their failure can significantly jeopardize both safety and economic stability. The primary factors contributing to the deterioration of bridge structural components include exposure to environmental conditions and the continuous increase in traffic volume. As a result, timely and consistent maintenance interventions are essential to maintaining adequate performance levels. Traditionally, this has been achieved through visual inspections to monitor the progression of deterioration.
In recent years, there has been a shift towards utilizing Structural Health Monitoring (SHM) systems to support decision-making. SHM systems offer a more reliable and continuous assessment of bridge conditions. However, up to now, only a limited number of bridges have these monitoring systems in place. In this context, this paper proposes a novel framework that incorporates the concept of transfer learning within Bayesian networks (BNs) to enhance the evaluation of SHM-based management strategies for bridges. A BN represents the joint distribution between a set of variables, through a directed acyclic graph and a set of conditional probability tables (CPTs) and it is an invaluable tool for bridge management, as it enables probabilistic updates on the bridge's condition throughout its lifespan. Transfer learning in BNs allows to leverage knowledge gained from monitoring a domain of bridges and to apply it to a related domain, thus addressing the inherent uncertainties in loads, mechanical strength, damage models, and numerical modelling.
The integration of transfer learning within Bayesian networks also facilitates the optimization of SHM system deployment and management. The proposed framework involves several key phases: (i) identifying the source and the target domain; (ii) identify common damage scenarios to be transferred and the transferable components; (iii) adapt the BN to ensure that the source domain fits the target domain; (iv) incorporate the transferred knowledge into the Bayesian network for the target domain for decision-making or predictive tasks; (v) assess the performance of the BN (validation). By adopting this advanced methodology, the framework aims to increase the efficiency and reliability of bridge infrastructure management, ensuring safer and more economically sustainable transportation networks.