Physics-informed neural network modelling of uplift behaviour of segmental linings during shield tunnelling

Uplift of segmental linings in shield tunnels presents considerable challenges, potentially compromising the structural integrity of tunnels. The uplift movement can be physically modelled using a Timoshenko beam on a Winkler foundation. This study introduces an innovative method employing a physics-informed neural network (PINN) to solve the governing differential equations of shield tunnel linings under specified boundary conditions, known loads, and foundation parameters. Importantly, the PINN does not rely on empirical data for training; instead, it incorporates physics-based constraints to accurately capture spatial variations in load and foundation stiffness during grouting and construction phases. The PINN model was validated with field data from a shield tunnel in the Pazhou branch of the Guangzhou–Dongguan–Shenzhen intercity railway line. The results demonstrate the effectiveness of the model in predicting segment uplift. Furthermore, compared to traditional analytical solutions, the PINN model provides a more realistic representation of field conditions by integrating spatial variations in loading and foundation support. • PINN based framework for segment uplift prediction during shield tunnelling • Model predicts segment uplift, bending moments, shear forces, and deformations. • The influence of varying construction conditions on uplift behavior is investigated. • Variation in longitudinal foundation stiffness improves prediction accuracy.