Physics-Informed Machine Learning for Pressure Drop Prediction in Electrospun Fibrous Filters

Pressure drop across filtration media is a crucial parameter in the design and performance of air filtration applications, and a low pressure drop is always desirable. Conventional analytical models for the pressure drop for electrospun nanofibrous materials, derived from porous media theory, rely on simplified structural assumptions and often fail to generalize across different polymers, fiber diameters, and fabrication conditions. In this article, we present various machine learning models, including a Physics-informed Neural Network (PINN) model for the pressure drop, which are developed using a comprehensive set of input parameters encompassing the thickness of the membrane, diameter of nanofibers, porosity and input air velocity. Our study establishes that a physically motivated constraint on the permeability gives substantially improved accuracy in prediction for PINN compared to other models. The PINN model shows an excellent R 2 value of around 0.983 against the test data. Results demonstrate that physics-informed models can accurately capture pressure–velocity relationships across diverse electrospun membranes and offer a generalizable alternative to classical semi-empirical correlations.