Hybrid modeling for optimizing electrospun polyurethane nanofibrous membranes in air filtration applications

Abstract

Nanofibers have gained recognition as promising materials for air filtration due to their high surface area-to-volume ratio, adjustable porosity, and exceptional mechanical properties. However, optimizing their structural characteristics to maximize filtration efficiency while minimizing pressure drop remains challenging due to the complexity of the electrospinning process. This study presents an artificial intelligence-based methodology to establish relationships between electrospinning parameters, nanofiber morphology, and filtration performance. An advanced statistical approach is used to systematically collect and analyze data, followed by modeling these relationships using artificial neural networks (ANN) and analytical formulas to enhance predictive accuracy. A genetic algorithm (GA) is subsequently utilized to refine electrospinning parameters, facilitating the production of nanofibers with enhanced filtration efficiency and optimized airflow resistance. The optimized nanofiber membranes are validated experimentally to assess their real-world performance. The findings demonstrate the potential of AI-driven design in fine-tuning nanofiber structures for advanced air filtration applications. The optimized sample achieved a filtration efficiency of 96%, a pressure drop of 110.23 Pa, and a quality factor of 0.0297. This study underscores the effectiveness of combining AI with electrospinning to develop high-performance air filtration materials.

Keywords: Air filtration; Artificial intelligence; Nanofibrous membrane; Optimization; Polyurethane.

Fig. 1

a Electrospinning setup, b filtration performance analysis setup.

Fig. 2

a Pearson’s heatmap correlation for control and response parameters, b data distribution of basis weights of the layer, c data distribution of layer thickness, d data distribution of nanofiber diameter for different samples.

Fig. 3

a Architecture of the developed ANN models, and b relative importance of input neurons.

Fig. 4

Workflow of the study to fine-tune the electrospinning parameters of PU nanofibrous air filters for the maximization of filtration efficiency and the minimization of pressure drop.

Fig. 5

Performance of the RSM model during the training step for a basis weight, b layer thickness, and c nanofiber diameter, and during the testing step for d basis weight, e layer thickness, and f nanofiber diameter. Performance of the ANN model during the training step for g basis weight, h layer thickness, and i nanofiber diameter, and during the testing step for j basis weight, k layer thickness, and l nanofiber diameter.

Fig. 6

Performance of the ANN-GA optimization process during a maximization of filtration efficiency, b maximization of pressure drop, c minimization of pressure drop, and d simultaneous maximization of filtration efficiency and minimization of pressure drop based on the cost function.

Fig. 7

a Filtration efficiency for different particle sizes before and after optimization, b pressure drop at various airflow rates before and after optimization, c quality factor of the optimized sample at different airflow rates, d filtration efficiency for PM₃₀₀ of the optimized sample at different relative humidities, e SEM image of the optimized sample, f nanofiber diameter distribution of the optimized sample.