ABSTRACT
With increasing focus on life cycle cost (LCC), the operation, inspection and maintenance of tunnels must be time and cost efficient. Automated systems, high-tech sensors and deep learning strategies based on AI, are state of the art.
The focus of the topic is the development of new and more automated methods to inspect and assess the conditions of tunnels. In particular, data collection using mobile mapping systems, and the use of non-contact systems to measure and detect damages are of interest. Also, contributions on how to assess risks associated with detected damages and proposed countermeasures are welcome.
Efficient management of transportation tunnels (TT) is crucial for ensuring infrastructure safety and longevity. Predictive maintenance approaches, which enable proactive interventions and extend asset lifespan, offer significant advantages. However, traditional tunnel diagnostic methods can be time-consuming and expensive. Here it’s presented a retrospective analysis of deep learning-based methods for defect segmentation in TT images, specifically focusing on ethical, sustainable, and secure implementations within the MIRET (Management and Identification of the Risk for Existing Tunnels) framework.
Among the categories identified as high-risk in the AI-Act, AI systems used for the management and operation of critical infrastructure stand out. MIRET-Tunnel AI is the software associated with the MIRET methodology. It relies on Artificial Intelligence algorithms to detect defects in lining structures. These algorithms assist expert users in pre-assessing the inspection Big Data of the tunnel lining to elaborate vulnerability models employed for the qualitative risk calculation. MIRET primarily analyses transport networks, which can be inspected using the ARCHITA acquisition system, and they are generally considered critical infrastructure.
MIRET-Tunnel AI is not commercially sold software; rather, it forms part of a risk analysis service involving numerous technical roles and responsibilities. In such a case, compliance with the AI Act requires a self-declaration and an evidence archive to demonstrate adherent strategies, including diverse and well-balanced training datasets, fair metrics during model development, quality management, data privacy, cybersecurity and human supervision throughout the decision-making process. The compliance of MIRET-Tunnel AI is analyzed for each category.
Deep learning-based automated segmentation, coupled with multi-dimensional mobile data acquisition systems, reduces the need for physical inspections. This translates to improved efficiency, reduced traffic disruptions, and a more sustainable approach to tunnel management. Compared to the standard baseline, the introduced approach can demonstrate a significant decrease in equivalent carbon emissions.
