Tunnel alignment design integrating physics-informed deep learning and enhanced optimization algorithm

The design of tunnel alignment based on mechanical stability principles faces significant challenges, including the arduous and time-consuming nature of conventional reliable data generation, limitations in efficient deformation prediction, and reliance on empirical approaches or engineering analogy for optimization. These challenges hinder efficient stability evaluation and optimal design. To this end, a hybrid framework is raised to implement automatic parametric modelling and data generation, efficient stability prediction, and intelligent optimization of tunnel design. Within this work, secondary development on existing software enables artificial data generation. A novel machine learning algorithm with the consideration of physics-informed and data-driven loss functions for higher precision is developed. The coordinates of four control points on the tunnel alignment are sought by an enhanced intelligent optimization algorithm. Validation using a numerical engineering case indicates that: (1) the secondary development for parametric modelling and automatic simulation is effective, reducing the time cost in data generation, (2) the proposed Transformer-Convolutional Neural Network (Trans-CNN) algorithm achieves satisfying performance in vault deformation prediction, with an R 2 value of 0. 9832 and a mean error around −0. 11%, outperforming the conventional CNN algorithm, (3) introducing of loss function revealing the vault deformation features ensures the precision of optimization process. Application of this framework on the design and optimization suggestions of underground engineering traversing complicated geological conditions holds immense potential. The mathematical framework underlying this study offers a transferable paradigm for other engineering domains requiring simulation-based optimization under physical constraints.

Authors

• Hui Li
• Weizhong Chen
• Xianjun Tan
• Hongming Tian
• Yaoyao Meng

Keywords

• Tunnel alignment
• Numerical analysis
• Machine learning
• Intelligent optimization