Intelligent Design for Mechanized Drilling and Blasting Tunnels Based on Massive Numerical Simulations and Machine Learning Algorithms

Mechanized drilling and blasting method has the characteristics of the long excavation length,short step,and large excavation face.Tunnel design methods,primarily reliant on the existing engineering experience,are difficult to meet the requirements of mechanized construction.Based on the tunnels located on Yukun high-speed railway,this paper randomly generated calculation cases that are consist with the actual conditions utilizing MATLAB.Massive automated numerical simulations were then conducted by the finite difference software FLAC^3D.The results of which were stored in the database.The machine learning algorithms were introduced to establish a nonlinear relationship between the data in input and output layers of the trained RBF neural network.Where,the in-situ stress field,physical and mechanical parameters of the rock mass,excavation and supporting parameters were seen as the input data.Meanwhile the deformation of surrounding stratum and mechanical response of supporting systems were seen as the output data.The GA-SA algorithm was adopted to optimize the excavation and supporting parameters of a specific tunnel section.The results indicate that the RBF neural network has a high prediction accuracy with a determination coefficient (R²) of 0.836 for the axial force at the vault of the secondary lining.By repeatedly invoking the trained neural network during the iterative optimization process,the mechanical response related to the input data can be directly obtained.This methodology significantly reduces the computational time required for the numerical simulations,enabling the results to expeditiously converge to the optimal level.