Paper Number
ML7                         My Program 

Session
AI and ML in Rheology

Title
Non-local physics-informed neural networks for forward and inverse solutions of granular flows

Presentation Date and Time
October 22, 2025 (Wednesday) 1:30

Track / Room
Track 6 / Sweeney Ballroom C

Authors (Click on name to view author profile)

1. Zolfaghari, Saghar (Northeastern University, Mechanical and Industrial Engineering)
2. Jamali, Safa (Northeastern University, Mechanical and Industrial Engineering)

Author and Affiliation Lines (in printed abstract book)
Saghar Zolfaghari and Safa Jamali
Mechanical and Industrial Engineering, Northeastern University, Boston, MA 02115

Speaker / Presenter 
Zolfaghari, Saghar

Keywords
artificial intelligence; granular materials; machine learning; non-Newtonian fluids

Text of Abstract
Understanding/predicting dense granular flows remains a major challenge due to the inherently nonlocal nature of grain interactions, especially in quasi-static or slowly sheared regions. Traditional local rheological models fail to capture spatial cooperativity effects that are prominent in many granular systems. The nonlocal granular fluidity (NGF) model addresses this limitation by introducing a diffusive-like partial differential equation for a fluidity field, governed by a key material-dependent parameter: the nonlocal amplitude A. However, determining A from experiments or simulations is known to be difficult and typically requires extensive calibration across multiple geometries. In this work, we present a data-driven platform based on Physics-Informed Neural Networks (PINNs) embedded with the NGF model, capable of solving granular flows in a forward or inverse manner. By training the model on transient flow in a planar shear configuration under gravity, we infer the full velocity, pressure, and stress fields. We observe that small variations in the nonlocal amplitude A possibly lead to sharp, bifurcation-like transitions in flow behavior, transitioning from localized shear zones to bulk flow regimes. To further address the challenge of identifying A, we also implement an inverse PINN architecture that treats A as a trainable parameter and learns its value directly from velocity data. This approach demonstrates the feasibility of data-driven parameter inference in complex nonlocal models and opens up new possibilities for characterizing granular materials from sparse experimental observations.