Inverse identification of the consolidation coefficient in cyclic poroelastic consolidation using physics-informed neural networks

Accurate identification of the consolidation coefficient is essential for predicting pore pressure and settlement in cyclically loaded saturated soils, yet remains challenging due to sparse and noisy field data. While Physics-Informed Neural Networks (PINNs) excel in solving inverse problems, their application in coupled cyclic poroelastic consolidation remains limited. This study develops a PINN framework for both forward simulation of consolidation and inverse estimation of the consolidation coefficient under haversine loading. Results demonstrate that the forward PINN accurately captures the coupled consolidation response by reproducing both pore pressure dissipation and settlement evolution. In the inverse problem, the model successfully identifies the consolidation coefficient while simultaneously reconstructing pore pressure and displacement fields. Pore pressure observations are found to be more informative than displacement data; for the one-dimensional (1D) homogeneous case and sensor configurations considered, errors below 5% are obtained when at least two pore pressure sensors are used. The model also performs reliably using only early-cycle observations and remains robust under moderate uncertainty for the measurement configurations investigated. Parameter identification is robust to pore-pressure noise but sensitive to displacement noise. These results demonstrate the potential of PINNs as a data-efficient framework for cyclic consolidation problems within the scope of the adopted 1D model.