Conference Agenda

Routine visual inspections of bridges are safety-critical activities that are still manual, time-consuming, and subjective to the inspector’s interpretation. This paper presents a scalable automated visual inspection pipeline for semantic segmentation of bridge defects, targeting the 19-class dacl10k benchmark dataset. We perform the first systematic comparison of pretraining paradigms on dacl10k, including CNN -supervised, ViT-supervised, masked image modelling, and self-supervised learning across 10 configurations under a controlled training protocol. The pretraining paradigm consistently dominates architectural choice, with DINOv2-L, pretrained on 142 million unlabelled images, achieving a mean Intersection-over-Union (mIoU) of 49.16%. Applying our native multi-label training approach to EVA-02-L, the dacl10k challenge-winning backbone, achieves 48.97% mIoU versus their 47.80% single-model result, demonstrating that training design is an independent performance factor. A three-model ensemble achieves 51.08% mIoU, exceeding the challenge winner’s score by using half the number of models. A prototype inspection system (InSpectralytiX) is deployed in a HuggingFace Gradio Space, demonstrating end-to-end feasibility from raw image to a per-class defect map. The future work targets automated condition scoring for bridge asset management integration, supporting structural health monitoring at the local level when performed repeatedly.

Adaptive service life prediction of reinforced concrete structures requires continuous, real-time assessment of the chloride-induced depassivation phase. Within this study, an established probabilistic service life model can be combined with embedded corrosion wire sensor data to enable adaptive, data-driven lifetime prediction under real exposure conditions. An integrated corrosion monitoring system with embedded filament wire sensors was employed in the instrumented prestressed concrete research bridge Concerto. These wire sensors enable direct, real-time detection of the depassivation front by measuring the abrupt increase in electrical resistance upon wire fracture. The resulting depassivation times are interpreted and incorporated as input parameters in an established probabilistic lifetime model. By updating model parameters based on wire sensor data, uncertainties are reduced and predictions are adapted to the actual structural condition. Evaluation of the long-term measurements from the Concerto research bridge demonstrates its applicability under real exposure conditions and illustrates how sensor-derived depassivation data can be integrated into established probabilistic lifetime models as a basis for data-driven maintenance planning and reliable estimation of remaining service life. Adaptive service life prediction of reinforced concrete structures requires continuous, real-time assessment of the chloride-induced depassivation phase. Within this study, an established probabilistic service life model can be combined with embedded corrosion wire sensor data to enable adaptive, data-driven lifetime prediction under real exposure conditions. An integrated corrosion monitoring system with embedded filament wire sensors was employed in the instrumented prestressed concrete research bridge Concerto. These wire sensors enable direct, real-time detection of the depassivation front by measuring the abrupt increase in electrical resistance upon wire fracture. The resulting depassivation times are interpreted and incorporated as input parameters in an established probabilistic lifetime model. By updating model parameters based on wire sensor data, uncertainties are reduced and predictions are adapted to the actual structural condition. Evaluation of the long-term measurements from the Concerto research bridge demonstrates its applicability under real exposure conditions and illustrates how sensor-derived depassivation data can be integrated into established probabilistic lifetime models as a basis for data-driven maintenance planning and reliable estimation of remaining service life.

Corrosion is the main deterioration process affecting reinforced concrete, and it can lead to a loss of serviceability and a reduction in structural reliability over time. The probabilistic model for corrosion initiation is now well established, and for practical applications, chloride ingress into concrete is usually described by Fick’s second law of diffusion. In this model, the chloride content at a given depth and exposure time depends on the surface chloride concentration and the apparent diffusion coefficient. The model parameters and the time of corrosion initiation are uncertain and vary with environmental exposure and concrete properties. Reducing the uncertainty in these parameters is therefore important to improve the prediction of the time to corrosion initiation. In this study, a Bayesian Network framework is proposed to refine the uncertainty in the model parameters by combining information from three different types of tests: (i) destructive tests on core samples providing a chloride content profile at three depths, (ii) non-destructive tests, using half-cell potential measurements, which provide an indication about the state of corrosion and (iii) continuous monitoring data, which provide information about the initiation time by detecting the change in electrochemical response when chlorides reach the reinforcement level. The evidence from these three monitoring strategies is used to update the probability distributions of the model parameters and the time of corrosion initiation.

The management of aging steel bridges poses significant challenges to bridge owners, who must maintain structural integrity under fatigue-induced deterioration while operating within limited maintenance budgets. Although post-weld treatment methods have demonstrated potential to improve fatigue performance, a probabilistic quantification of their comparative reliability benefits remain limited in literature. This study presents a probabilistic S-N based reliability approach to evaluate and compare as-welded condition and three post-weld treatment methods, including burr grinding/TIG, needle/hammer peening, and HFMI. The approach is applied to a welded steel detail under variable amplitude loading, incorporating uncertainties in the fatigue resistance, damage accumulation and stress modelling. The results show that post-weld treatments significantly extend service life. Furthermore, an analysis shows the potential of SHM also contributing to service life extension; however, subjected to a specific outcome probabilities