NUMERICAL INVESTIGATION OF END REINFORCEMENT TECHNOLOGY FOR SUPER-LARGE-DIAMETER SHIELD TUNNELS AND ONLINE PREDICTION USING MACHINE LEARNING

Shield tunneling has become a popular method in underground construction nowadays. In the process of shield launch or reception, end reinforcement of the connection between the shaft and the shield tunnel can avoid accidents such as excavation collapse or critical surface subsidence. However, the appropriate reinforcement method for super-large-diameter tunnels in soft soil remains unclear. Based on a real tunneling project, the Zhuhai tunnel with a 15 m diameter, this paper investigates the effect of different end reinforcement methods on tunneling-induced ground movements via the finite element method (FEM). The numerical model is validated based on field monitoring data and the optimal method (line type using full reinforcement) is subsequently suggested. After that, a detailed parametric study is carried out to distinguish the key model factors (elastic moduli of the reinforcement zones) influencing the ground movements during construction. Furthermore, to achieve real-time prediction of excavation-induced ground movements during construction, a modified multi-objective optimization procedure combined with a critical neuron network surrogate modeling is developed. The results show the proposed framework can efficiently predict breaking operation-induced model responses and the accuracy is more than 90%, which can significantly benefit engineering practice.