Process optimization of polyvinylidene fluoride/polypropylene gradient composite filter media based on BP neural network

Mask was an important epidemic prevention barrier to prevent virus from entering human body through respiratory system and mucous membrane. The disposable mask had some problems, such as rapid decline of filtration efficiency with electrostatic attenuation, large respiratory resistance, short service life and so on. Electrospun nanofiber membrane was compounded with melt blown cloth to reduce the dependence of particle filtration on static electricity and realize long-term filtration. Polyvinylidene fluoride (PVDF) nanofiber membrane was prepared by electrospinning with N, N-dimethylformamide (DMF) as solvent. Then it was coated with polypropylene (PP) melt blown base cloth to prepare PVDF/PP nano/micron structure composite fiber membrane. The effect of electrospinning process parameters on the aerosol filtration performance of composite fiber membrane was experimentally studied. The ternary quadratic polynomial model was established to optimize the spinning process and predict the fiber membrane resistance. At the same time, the back propagation (BP) neural network model was constructed to predict the fiber membrane resistance. The results show that the effects of voltage, receiving distance, injection speed, spinning solution concentration and fiber membrane surface density on the filtration efficiency and filtration resistance are consistent. When the concentration of spinning solution is 15wt% and the area density is 3 g/m2, the optimized spinning process parameters are voltage of 30 kV, receiving distance of 16.8 cm and injection speed of 1.6 mL/h. The filtration resistance predicted by polynomial model is 76.79 Pa, the relative error is 9.23%, and the error coefficient of variation (CV) value is 59%. The filtration resistance predicted by BP neural network is 81.25 Pa, the relative error is 1.99%, and the error CV value is 48%. The experiments show that the ternary quadratic model and BP neural network have high prediction accuracy. © 2022 Beijing University of Aeronautics and Astronautics (BUAA). All rights reserved.