Iranian Research Organization for Science and Technology (IROST) Innovative Food Technologies 2783-350X 4 4 2017 06 22 Evaluating the performance of artificial neural networks (ANNs) for predicting the effect of polymer concentration and operating voltage on the physical properties of electrosprayed particles Evaluating the performance of artificial neural networks (ANNs) for predicting the effect of polymer concentration and operating voltage on the physical properties of electrosprayed particles 31 43 409 10.22104/jift.2016.409 FA Ali Alehosseini PhD. Student of Food Materials and Process Design Engineering, Department of Food Nanotechnology, Research Institute of Food Science & Technology, Mashhad, Iran Mahboobe Sarabi Jamab Assistant Professor, Department of Food Biotechnology, Research Institute of Food Science & Technology (RIFST), Mashhad, Iran Behrouz Ghorani Assitant Professor, Department of Food Nanotechnology, Research Institute of Food Science & Technology Rassoul Kadkhodaee Associate Professor, Department of Food Nanotechnology, Research Institute of Food Science & Technology Saowakon Wongsasulak Assitant Professor, Department of Food Engineering, King Mongkut’s University of Technology Thonburi (KMUTT), Bangkok, Thailand Journal Article 2016 11 01 Electrohydrodynamic atomization has recently been introduced as a safe method for encapsulation of sensitive bioactive compounds. In this regard, the parameters affecting the size and morphology of particles, owing to their key role in the protection and controlled release of bioactive compounds, are of paramount importance. In this study the influence of whey protein concentrate (WPC), gum Arabic and skim milk concentration as well as the operating voltage on the size and morphology of electrosprayed particles was investigated and the experimental data were then used for simulation of the process using two feed-forward artificial neural networks: multilayer perceptron and radial basis function networks. These models along with different threshold functions were employed to predict the viscosity, surface tension, electrical conductivity of polymers solutions, and the size of particles. The results showed that multilayer perceptron network comprising a set of 5 inputs and a single hidden layer containing 4 neurons with hyperbolic tangent activation function which was trained by Levenberg-Marquardt (LM) learning algorithm at 1000 epochs predicted the desired outputs more accurately than the radial basis function network. The determination coefficients (<em>R²</em>) for particles size, viscosity, surface tension, and electrical conductivity of polymers solutions were found to be 0.958, 0.9991, 0.996 and 0.967, respectively. Examination of the sensitivity analysis diagram of the predicted values by multilayer perceptron neural network versus experimental data revealed random scattering of points in the close proximity of the regression line with a high determination coefficient which clearly verifies the high accuracy of the model in predicting the outputs. It should be noted that size of particles as well as viscosity and electrical conductivity of solutions significantly increased with the rise of polymers concentration (<em>p</em> < 0.05). The findings of this study may be of useful to predict the impact of the above-mentioned parameters on the physical properties of polymer solutions and the fabricated particles in order to select the most appropriate wall materiel based on the purpose of encapsulation of bioactive compounds. Electrohydrodynamic atomization has recently been introduced as a safe method for encapsulation of sensitive bioactive compounds. In this regard, the parameters affecting the size and morphology of particles, owing to their key role in the protection and controlled release of bioactive compounds, are of paramount importance. In this study the influence of whey protein concentrate (WPC), gum Arabic and skim milk concentration as well as the operating voltage on the size and morphology of electrosprayed particles was investigated and the experimental data were then used for simulation of the process using two feed-forward artificial neural networks: multilayer perceptron and radial basis function networks. These models along with different threshold functions were employed to predict the viscosity, surface tension, electrical conductivity of polymers solutions, and the size of particles. The results showed that multilayer perceptron network comprising a set of 5 inputs and a single hidden layer containing 4 neurons with hyperbolic tangent activation function which was trained by Levenberg-Marquardt (LM) learning algorithm at 1000 epochs predicted the desired outputs more accurately than the radial basis function network. The determination coefficients (<em>R²</em>) for particles size, viscosity, surface tension, and electrical conductivity of polymers solutions were found to be 0.958, 0.9991, 0.996 and 0.967, respectively. Examination of the sensitivity analysis diagram of the predicted values by multilayer perceptron neural network versus experimental data revealed random scattering of points in the close proximity of the regression line with a high determination coefficient which clearly verifies the high accuracy of the model in predicting the outputs. It should be noted that size of particles as well as viscosity and electrical conductivity of solutions significantly increased with the rise of polymers concentration (<em>p</em> < 0.05). The findings of this study may be of useful to predict the impact of the above-mentioned parameters on the physical properties of polymer solutions and the fabricated particles in order to select the most appropriate wall materiel based on the purpose of encapsulation of bioactive compounds. https://jift.irost.ir/article_409_bae8e11f3f86b84afa8cd4782b0e74c0.pdf