Electrospinning predictions using artificial neural networks

Abstract

Electrospinning is a relatively simple method of producing nanofibres. Currently there is no method to predict the characteristics of electrospun fibres produced from a wide range of polymer/solvent combinations and concentrations without first measuring a number of solution properties. This paper shows how artificial neural networks can be trained to make electrospinning predictions using only commonly available prior knowledge of the polymer and solvent. Firstly, a probabilistic neural network was trained to predict the classification of three possibilities: no fibres (electrospraying); beaded fibres; and smooth fibres with > 80% correct predictions. Secondly, a generalised neural network was trained to predict fibre diameter with an average absolute percentage error of 22.3% for the validation data. These predictive tools can be used to reduce the parameter space before scoping exercises.