Data-Driven Inverse Design of Double-Layer Air-Filter Microstructures for Targeted Filtration Performance

The design of multilayer fibrous filters remains challenging owing to the complex interactions among microstructural parameters, airflow resistance, and particle capture mechanisms. In this study, a data-driven inverse design framework is developed to derive dual-layer filter microstructures that satisfy predefined filtration performance targets. A data set of 338 geometries is generated using the computational fluid dynamics and discrete phase model simulations to compute the pressure drop and collection efficiency. Two neural network (NN) models are trained as surrogate predictors and coupled with a genetic algorithm (GA) to efficiently explore the high-dimensional design space. The GA-derived filter configurations are validated using high-fidelity simulations. The results show that the outer layer plays a dominant role in regulating hydraulic resistance, whereas the inner layer governs particle capture, particularly near the most penetrating particle size range. This framework provides a predictive pathway for tailoring microstructures to meet application-specific filtration requirements.