PSI - Issue 78

Francesco Mariani et al. / Procedia Structural Integrity 78 (2026) 875–882

881

Fig. 6. Damage detection on pier scenario

Fig. 7. Probabilistic damage assessment on pier scenario

necessitating immediate intervention. In the scenario under consideration, all nine pier bases are subjected to varying levels of damage. Each predicted elastic modulus is represented by a distribution of 200 samples obtained from the BNN’s posterior estimates. By comparing these probabilistic distributions to the defined thresholds, the di ff erent levels of attention required for each pier become evident. Specifically, piers 1, 6, 7, and 9 exhibit distributions with a large proportion of samples between the warning and critical thresholds, suggesting the need for monitoring but a relatively low probability of critical failure. In contrast, piers 2, 3, 4, 5, and 8 show a significantly higher probability of being within the critical damage zone. These predictions suggest that these piers should be prioritized for inspection and potential maintenance, thereby demonstrating the utility of the BNN framework for supporting informed, risk-based decision-making. This study presents a BNN framework for the probabilistic detection and quantification of structural damage in a complex, real-world post-tensioned concrete box girder bridge. The proposed approach demonstrates the capability of BNNs to produce accurate and reliable predictions even when trained on relatively small datasets. A core strength of the framework lies in its probabilistic nature, which enables the model to output full distributions of possible damage states rather than single-point estimates. This feature allows for a more comprehensive understanding of structural health, accounting not only for the most likely condition of each element but also for the associated uncertainty. As a result, the framework provides valuable support for risk-based decision-making, empowering infrastructure managers to prioritize inspections and allocate maintenance e ff ortsmore e ffi ciently. Elements with a high probability of critical 8. Conclusions