Artificial neural network modeling for predicting pore size and its distribution for melt blown nonwoven

Abstract

To improve the feasibility of developing melt blown nonwoven filtering material with given pore size specifications, the predictive power of a back-propagation (BP) artificial neural network (ANN) that takes the processing parameters as its inputs for pore size and its distribution, characterized by the variation coefficient of pore size, was investigated. Twenty-seven samples of melt blown nonwoven were produced and their images were collected using the scanning electron microscopy (SEM) method. The pore sizes were measured using digital image processing technology in which maximum entropy thresholding image segmentation based on a genetic algorithm was adopted. Seven BP ANN models were constructed by varying the number of neurons in the hidden layer. Metering pump frequency, mesh belt frequency, and the distance from die to collector (DCD) were chosen as the inputs of BP ANN. The results show that BP ANN can effectively reflect the nonlinear relationship between the processing parameters, and the pore size and its distribution. The mean absolute percentage errors (MAPE) between the predicted values and the measured values of the 7 models are all below 5%. Among these 7 models, the one that contains 7 neurons in its hidden layer has the minimum predictive error. The ANN model has stronger predictive power than the multiple linear regression model.