Physics-informed neural networks for resin flow prediction in fibre textiles

Machine learning is a fast-growing area being increasingly applied to engineering problems. The balance between data driven approach and physics-based modelling makes the concept of physics-informed neural networks (PINNs) in general particularly attractive. In this work, we will adopt the PINNs to solve the differential equation system that governs the Newtonian flow in porous media, where Darcy’s law is incorporated with two-phase flow configuration. In order to enable the PINN model to consider physically meaningful parameters, the governing equations are non-dimensionalised. Numerical examples are presented for 2D problems, considering heterogeneous permeability fields that are representative of real-world scenarios. The predictions are compared against a well-established traditional numerical solver.