Abstract:

In order to predict the filtration performance of melt blown nonwoven, a prediction method based on attribute reduction and support vector machine was introduced. Six reducts, each including three parameters, were extracted from the complete parameter set of fiber web structure of melt blown nonwoven fabric in ROSETTA environment using rough set theory. Twenty eight models, each based on either a support vector machine (SVM) or a back-propagation artificial neural network (BP-ANN), were established to predict the filtration performance by taking the parameters of each reduct and the complete parameter as inputs. A k-fold cross validation technique was applied to access the optimized structural parameters of the models. The results show that the prediction accuracy of the SVM-based model taking thickness, fiber diameter and pore as input parameters is higher than that of any other model. The values of its prediction precision for both filtration efficiency and pressure drop are higher than 98% and their variation coefficients are both lower than 2%. This indicates that these three parameters can be considered as key factors injluencing the filtration performances of melt blown nonwoven fabric. Generally, the prediction performance of SVM-based models are better than that of BP-ANN-based models.

Key words: melt blown nonwoven fabric, fiber web structure, attribute reduction, support vector machine, cross validation, filtration performance