Physics‐informed neural network for predicting extrusion force in basket granulation

Abstract To enhance understanding of basket granulation, this study presents a custom‐built, instrumented radial extruder designed to evaluate how system parameters, including hole diameter, hole count, die thickness, and roller radius, affect extrusion force and extrusion rate. By adapting the Hagen–Poiseuille and orifice flow equations, semi‐empirical models were developed to predict extrusion rate and extrusion force, respectively. A novel composite solidity‐aspect ratio (CSAR) index was introduced to quantify extrudate quality by integrating both aspect ratio and solidity into a single metric. High CSAR values corresponded to elongated, compact extrudates, while low values indicated fractured, irregular morphologies. A physics‐informed neural network (PINN) was developed to predict extrusion force by integrating the developed extrusion force model into the machine learning framework. A proof‐of‐concept study experimentally validated that the PINN achieved better predictive accuracy for extrusion forces compared to conventional artificial neural networks (ANNs).