Modeling and optimization of electrospun PAN nanofiber diameter using response surface methodology and artificial neural networks

Response surface methodology (RSM) based on a three-level, three-variable Box-Benkhen design (BBD), and artificial neural network (ANN) techniques were compared for modeling the average diameter of electrospun polyacrylonitrile (PAN) nanofibers. The multilayer perceptron (MLP) neural networks were trained by the sets of input–output patterns using a scaled conjugate gradient backpropagation algorithm. The three important electrospinning factors were studied including polymer concentration (w/v%), applied voltage (kV) and the nozzle-collector distance (cm). The predicted fiber diameters were in agreement with the experimental results in both ANN and RSM techniques. High-regression coefficient between the variables and the response (R2 = 0.998) indicates excellent evaluation of experimental data by second-order polynomial regression model. The R2 value was 0.990, which indicates that the ANN model was shows good fitting with experimental data. Moreover, the RSM model shows much lower absolute percentage error than the ANN model. Therefore, the obtained results indicate that the performance of RSM was better than ANN. The RSM model predicted the 118 nm value of the finest nanofiber diameter at conditions of 10 w/v% polymer concentration, 12 cm of nozzle-collector distance, and 12 kV of the applied voltage. The predicted value (118 nm) showed only 2.5%, difference with experimental results in which 121 nm at the same setting were observed. © 2012 Wiley Periodicals, Inc. J Appl Polym Sci, 2012