Search

Due to the variability and randomness of the geotechnical media, it is difficult to determine the constitutive parameters of the geotechnical materials in a numerical model. In this paper, a novel probabilistic back-analysis framework to update geotechnical parameters and better predict displacement due to shield excavations has been proposed. The surrogate model for the three-dimensional finite element model of shield tunneling is proposed to improve the computational efficiency based on proper orthogonal decomposition and artificial neural network (POD-ANN). In view of the cumbersome process of training the surrogate model, a training framework based on ABAQUS-PSO is presented. Considering the uncertainty factors such as the prior information of the geotechnical parameters and observation errors, probabilistic back-analysis based on Bayesian theory is implemented. The POD-ANN surrogate model and Bayesian theory are integrated to achieve an efficient back-analysis, and Delayed Rejection Adaptive Metropolis is used to solve the joint posterior distribution of the geotechnical parameters. To corroborate the effectiveness of the proposed framework, it was applied to the displacement prediction of shield tunneling under a railway project. The results show that the proposed framework can significantly improve computational efficiency and prediction accuracy. [ABSTRACT FROM AUTHOR]