Development of novel symbolic regression models for prediction of nano filters efficiency using CFD and DPM

This study presents a comprehensive CFD-DPM framework for analyzing the filtration performance of nanofiber media, with a specific focus on parameters relevant to face mask applications while maintaining general applicability. A detailed parametric study is conducted to assess the influence of inlet velocity, particles density and particle number density on filtration efficiency and pressure drop. The results demonstrate a complex, non-linear relationship between these parameters, where filtration efficiency is most sensitive to inlet velocity, exhibiting a decrease from approximately 96% to 42% as velocity increases. To understand these complex interactions, advanced Artificial Intelligence techniques are employed. An Artificial Neural Network model achieved exceptional predictive accuracy (R² > 0.999) and is subsequently used for a global sensitivity analysis via the Morris method, which quantitatively ranked inlet velocity as the most influential parameter. Furthermore, a novel, highly accurate explicit correlation for predicting filtration efficiency is derived using Symbolic Regression, achieving a coefficient of determination R² > 0.998 for both training and testing datasets. Since the efficiency is found out to be much less sensitive to the number of inlet particles, the correlation is harnessed to derive an averaged expression for the efficiency as a function of the inlet velocity and particles density. It acquires R² > 0.999, demonstrating excellent accuracy of the function, which can be used for practical purposes. However, a shorter correlation is also derived with R² > 0.98 for stronger interpretation.