Abstract:

This lecture presents a multi-source information fusion and intelligent algorithm approach for the whole-life-cycle safety risk assessment of tunnel engineering, covering geology, design, construction, operation, and maintenance. At its core, the method leverages sensing technologies, machine learning, and numerical simulation to enable data-driven dynamic assessment and control of safety risks. In geological modeling, a 3D transparent geological model is constructed based on the Markov principle, integrating real-time vision and borehole data. Deep learning is applied to dynamically identify geotechnical parameters, thereby digitizing the geological environment. During the design and construction stages, machine learning methods are used to achieve high-precision predictions of structural parameters, forming a data-driven and machine-learning-based design methodology. In construction, embodied intelligence is introduced: sensors monitor the tunnel boring machine (TBM) attitude and the excavation face in real time, forming a sensing–decision–execution closed loop to dynamically control ground settlement and achieve risk-controllable intelligent tunneling. In the operation and maintenance phase, mobile vision devices automatically identify surface defects (such as cracks, leakage, and spalling) and upload data in real time via wireless network, enabling automatic detection and localization of defects and improving operational efficiency. For safety assessment, multi-source information from the above stages is integrated, and numerical simulation techniques such as the phase-field method are used to simulate defect evolution considering ground–structure interaction, which can quantify safety risk for the tunnel.