Physics-informed neural networks for granular flows modelling

Physics-Informed Neural Networks (PINNs) have recently emerged as a powerful framework for solving forward and inverse problems involving partial differential equations, by embedding physical laws directly into the training process of neural networks. A key advantage of PINNs lies in their ability to infer hidden or hard-to-measure physical quantities from limited data. In this work, we explore the capability of PINNs to investigate dry granular flows, complex systems that are challenging to study due to their non linear rheology and the high computational cost of traditional simulations. To describe the flow dynamics, we adopt the μ ( I ) rheology, a well-established constitutive model for granular materials. We propose an approach based on a simple and accessible experiment: the collapse of a granular column. Using synthetic data generated from this setup, we demonstrate that PINNs can accurately reconstruct the pressure field and reliably identify key rheological parameters such as μ s and μ 2 . This study highlights the potential of PINNs as a low-cost, flexible tool for rheological characterization of granular materials, potentially replacing more complex and expensive experimental protocols, like rheometers.