PSI - Issue 84

Federico Foria et al. / Procedia Structural Integrity 84 (2026) 304–312

307

Fig. 3. ARCHITA.

In contrast to standard inspections nowadays, ARCHITA moves the inspection phase from on-field to back-office guaranteeing more reliable and cost-time efficient data (compared to e.g. visual inspection, trolley). The inspections were concluded with just a few nights of interruption of service, obtaining: • The geometric model of a tunnel and the HD photo, thermal camera image and point cloud for defect detection. • The HD photo and the thermal image allow mapping of the defects to assess the lining condition for more efficient design-planning. A critical aspect of the management and operation of tunnels is the inspection phase and, afterward, the vulnerability elaboration and risk assessment. Deep learning, particularly convolutional neural networks (CNNs), has emerged as a promising solution for automated defect segmentation in various image analysis tasks. The successful implementation of deep learning in the MIRET methodology led us to the MIRET-Tunnel AI, a software based on artificial intelligence algorithms, for detecting defects in lining structures, assisting expert users in the pre-assessment of the qualitative risk associated with tunnel structures. MIRET primarily analyses transport infrastructure, which can be considered critical infrastructure, they are inspected using the ARCHITA and then the MIRET-Tunnel AI is used to process the acquired data (Foria et al. 2025). 2.2 Defect Detection: MIRET-Tunnel AI

Fig. 4. The image shows the two workflows for defect analysis: the top one describes the training of artificial intelligence; the bottom one describes its use.