ABSTRACT
The excavation-induced tunnel displacement is an increasing problem with the intensive utilization of urban underground space. To rapidly and accurately predict the deformations in different scenarios, this study proposed a practical framework to build a digital twin of the tunnel–soil–foundation pit by a hybrid finite element (FE) and surrogate model approach. First, a parametric FE model is constructed, with 13 parameters as inputs and tunnel deformations at 5 positions as outputs. Then, a uniform design is adopted to generate the parameter sampling space of the FE examples, which reduces the number of samples for machine learning (ML) and computational cost. Finally, a surrogate model based on back propagation neural network (BPNN) is established, and the FE computation results and measured data from a real project are employed for verification. The results show that the accuracy of the validation set is more than 91.38%, the accuracy of the test set is more than 80%, and the error with the measured data is less than 1.5 mm. The BPNN–based surrogate model is high-efficient with a computational time of milliseconds. The established surrogate method could replace the expensive physics–based FE model for faster computation. The contribution is valuable in constructing digital twins of the tunnel during operation, especially suitable for safety assessment of existing tunnel in the scenario of adjacent engineering activities.
