Time-dependent convergence prediction, uncertainty quantification of creep rock behavior and design optimization of deep tunnel support using ai methods
Please login to view abstract download link
The creep behaviour of rock formation, the phenomenon induces a significant evolution in time of tunnel convergence, has been largely observed in many deep underground construction projects. The possibility to predict accurately this time-dependent convergence and uncertainty quantification of creep rock behaviour are essential to avoid setbacks related to the jamming phenomenon during the excavation by TBM as well as the damage of tunnel support system at long term. This study aims at demonstrating the efficiency of the AI methods in predicting the increase in time of tunnel’ convergence and in quantifying the uncertainty of the time-dependent behaviour of rock mass through which the design optimization of the support system can be conducted. Following that, in the first stage, a recent innovative deep learning technique is chosen and adapted to predict the evolution in time of tunnel convergence. Then the probabilistic inversion by Bayesian inference is carried out to quantify the mechanical properties that characterize the short and long-term behaviour of rock mass. Finally, in the step, the combination of the Kriging surrogate with the quantile-based optimization approach is performed to optimize the support system of tunnel. The validation of the proposed method is undertaken using both the synthetic data and in-situ measurements.
