The whole article can be accessed below the following excerpt.
What are some uses of AI in the design, operation and maintenance of tunnel systems and structures?
Nick Mirsepassi: AI, particularly machine learning and more specifically deep learning, is transforming the design and analysis of tunnel structures. Deep learning models excel in identifying complex patterns within geological and structural data, enabling the optimization of ground support designs tailored to specific conditions. These models can process and learn from available data, improving the prediction of tunnel stability and deformation and the selection of appropriate support systems.
AI-driven automation also streamlines documentation by generating detailed reports, 3D models and construction drawings directly from the design process. Furthermore, deep learning enhances sensitivity analysis as neural networks can rapidly assess the impact of varying ground conditions or design parameters on tunnel stability, uncovering potential risks or inefficiencies.
By integrating these advanced technologies, engineers can achieve higher precision, reduced design time and more reliable outcomes during both the planning and construction phases of tunnelling projects.
Gencer Koc: The day-to-day work in tunnel systems requires significant amount of analysis involving long modelling and computation times. One- and three-dimensional computational fluid dynamics [CFD] analyses—carried out as part of fire and life safety assessments as well as tunnel ventilation engineering—are among the most demanding tasks in terms of cost
and time. They involve creation of model geometries and the preparation of complex models using CFD software. The preparation of models and the run times of the analysis using CFD software can take from a couple of days to weeks, depending on the size of the domain under consideration.
Advanced machine learning algorithms, including artificial neural networks, can be used to replace or partially replace conventional CFD analysis tools. When trained on high-quality data, machine learning tools have the potential to provide accurate results involving much shorter analysis runtimes as compared to the use of currently available CFD tools.
Also, the improvement of processes, adopted during pre- and post-processing of analysis, have the potential to significantly save on time and cost, leaving more room for design iterations and optimization.
