Abstract: In order to effectively predict the pore size and distribution of spunbond nonwovens, thirty polypropylene spunbond nonwovens were produced by varying metering pump frequency and mesh belt frequency. The pore size of these samples was measured by digital picture processing technique. By changing the number of neurons in hidden layer, seven back-propagation(BP) artificial neural network(ANN) models were established by inputting metering pump frequency and mesh belt frequency. These models were used to predict the pore size and its variation coefficient. The results show that the mean absolute percentage errors among the predictions of these 7 models are all less than 5%. And when the hidden layer contains 5 neurons,the prediction accuracy of this model is the highest, which further proves the high prediction accuracy of BP ANN. Moreover, the prediction of the BP ANN model is more effective than that of the multiple linear regression model.

Key words: artificial neural network; spunbond nonwovens; pore size and its distribution; digital image processing technology