Erosion prediction of grouted and lined subsea tunnels over full life cycle: Physics-informed neural networks with nonlinear seepage theory

Physics-Informed Neural Networks (PINNs) with the Hansbo nonlinear seepage theory are established to study the erosion prediction of grouted and lined subsea tunnels. The erosion process is divided into three stages: low-speed, developing, and accelerated erosion. The Partial Differential Equations of penetration pressures during the different erosion stages are derived according to the Hansbo nonlinear seepage theory. The proposed PINNs framework employs a meshfree computational approach to conduct forward analysis of the advection-dispersion process with time-dependent permeability. This method eliminates the constraints of complex grid generation and mesh distortion inherent in traditional numerical simulations. It allows for the integrated prediction of penetration pressure, ion concentration, and erosion depth. Furthermore, the meshfree architecture facilitates an efficient inverse analysis to estimate the unknown parameters like permeability coefficient K and bulk modulus K v from sparse data. The framework is fully verified through numerical simulation, experimental data, and analytical solutions. The effects of initial water pressure and configuration optimization are discussed. The Hansbo seepage law vs. Darcy seepage law and PINNs model vs. Data-Driven Machine Learning models are compared. These findings reveal that the proposed PINNs with nonlinear seepage theory can ensure high accuracy and exceptional noise robustness to study ion erosion problems.