Sequential data assimilation for digital twin modeling of shield tunnel structure-soil interaction systems

Developing digital twin models for long-serviced shield tunnels is important for the safe operation and maintenance decision-making of tunnels. However, shield tunnel structure-soil interaction systems involve complex multi-physics coupling behaviors and may suffer from localized defects. Establishing a digital twin model capable of tracking and accurately predicting the system states and behaviors remains challenging. To address this issue, by extending the ensemble Kalman filter with subset simulation (EnKF-SuS), a rigorous and efficient Bayesian updating method, to the recursive Bayesian framework, a sequential data assimilation (DA) scheme for digital twin modeling was proposed, which aims to update the model and estimate possible system states by assimilating new observations. To validate the performance of the proposed method, finite element models incorporating tunnel construction and thermo-hydro-mechanical (THM) coupling were developed. The reliability of forward modeling was validated against the analytical solutions of segmental lining mechanical responses and field data from an energy tunnel. Then, based on a real-world energy tunnel and a scenario involving the hydraulic performance degradation of the lining joint respectively, the uncertainty quantification results and computational time of the developed algorithm for updating model inputs (including time-invariant/variant parameters) and predictions were examined. Results indicate that the proposed method can timely track and accurately estimate the time-varying system states and behaviors by integrating the model with sparse observational data.