Starch foam material performance prediction based on a radial basis function artificial neural network trained by bare‐bones particle swarm optimization with an adaptive disturbance factor

Abstract

A novel model based on a radial basis artificial neural network and bare‐bones particle swarm optimization tuned with adaptive disturbance factor for predicting the performances of starch‐based foam materials was established. The ethylene–vinyl acetate/starch mass ratio, glycerin content, and NaHCO3 content were used as the input variables, whereas the tensile strength and rebound rate were taken as the output variables of the model. The prediction results show that model predictions were in great accordance with the experimental values. The root mean square error of prediction and the correlation coefficients were 0.0256 and 0.9873; this indicated the good performance of the model. The model predicted that the tensile strength of the starch‐based foam materials would decrease slowly with increasing glycerin content and showed a V‐shaped variation with increasing NaHCO3 content. The rebound rate increased with increasing glycerin content and presented an inverted V‐shaped variation with increasing NaHCO3 content. The predicted results were consistent with the experimental results. © 2016 Wiley Periodicals, Inc. J. Appl. Polym. Sci. 2016, 133, 44252.

Predicting mechanical performance of starch-based foam materials

“…In Table 2, we compare performance of the MLR for resilience and GPR for tensile strength with that of more complicated models, including the back-propagation (BP) ANN, radial basis function (RBF) ANN, particle swarm optimization (PSO) RBF ANN, support vector machine (SVM), and bare-bones particle swarm optimization with adaptive disturbance factors (BBPSO-AD) RBF ANN reported in a previous study. 29 From the ARD and RMSE, it can be seen that the MLR and GPR lead to improved accuracy while maintaining simplicity. This is an important issue to consider when developing data-driven approaches for materials performance predictions.…”

“…Here, our goal is to develop simple and accurate predictions for resilience and tensile strength based on mixture proportions of components for starch-based/EVA foam materials though datadriven methods, which have been applied in a wide variety of fields for similar purposes. [26][27][28][29] The former is achieved through a multivariate linear regression (MLR) and the latter through a nonlinear Gaussian process regression (GPR). We compare performance measurements based on the MLR and GPR with those based on more complicated models, such as the support vector machine (SVM) and different versions of the artificial neural network (ANN), and find that simpler models here could lead to higher accuracy.…”