A PINN Framework for Perturbed Poromechanical Models

Solving partial differential equations with errors or performing regression with a small number of data alone may not yield accurate solutions. A Physics Informed Neural Network (PINN) model can offer a robust approach to address this issue. By incorporating both the governing physics and sensor data, PINNs can effectively address the discrepancies introduced by an inaccurate knowledge of material parameters, thus leading to a perturbed model, and guide the model towards the true solution. In this work, we develop a PINN-based approach for a perturbed model in poromechanics. The results on a synthetic test case show that PINNs can represent a promising tool to integrate sensor measurements with an underlying physical model, even if the knowledge of the latter is inaccurate.