A Neural Network Approach to Reducing the Costs of Parameter-Setting in the Production of Polyethylene Oxide Nanofibers

Abstract

Nanofibers, which are formed by the electrospinning process, are used in a variety of applications. For this purpose, a specific diameter suited for each application is required, which is achieved by varying a set of parameters. This parameter adjustment process is empirical and works by trial and error, causing high input costs and wasting time and financial resources. In this work, an artificial neural network model is presented to predict the diameter of polyethylene nanofibers, based on the adjustment of 15 parameters. The model was trained from 105 records from data obtained from the literature and was then validated with nine nanofibers that were obtained and measured in the laboratory. The average error between the actual results was 2.29%. This result differs from those taken in an evaluation of the dataset. Therefore, the importance of increasing the dataset and the validation using independent data is highlighted.

Keywords: PEO nanofibers; artificial neural networks; electrospinning.

Figure 1

Schematic presentation of the ANN model development.

Figure 2

Schema of the hyperparameter tuning process.

Figure 3

The basic configuration of a laboratory-made electrospinning device. (A,B,D) The Solidwork^® designs. (C) Macroscopic view of the electrospinning device.

Figure 4

Correlation between the real and predicted dataset diameter values, using the testing dataset.

Figure 5

Correlation between the laboratory and the predicted dataset’s diameters, using the testing dataset.

Figure 6

Prototype for the prediction model, using the MATLAB framework.

Figure 7

Solvent codes, as shown on the prototype model’s screen.