A novel Artificial Neural Network-based model for predicting dielectric properties of banana fiber filled with polypropylene composites

Abstract

The objective is focusing on the prediction of dielectric properties of the polypropylene composites reinforced with banana fiber using an Artificial Neural Network (ANN). To prepare the composites in accordance with ASTM requirements, randomly oriented banana fibers were combined with polypropylene at volume fractions of 20%, 30%, 40%, and 50%. For these composites, the impedance analyzer was used to determine dielectric characteristics such as the dielectric constant, tan δ, and ac conductivity. To estimate the dielectric properties, an artificial neural network is used with a supervised training strategy. The data set was assembled using ReLU, sigmoid, and tanh, three activation functions. Forecasting the outcome variables used temperature, frequency, filler content, and polymer content as input factors. Comparing the model utilizing ReLU to the other two activation functions, the MSE value was 0.32, and the R2 value was 0.98. Dielectric parameter values from both experiments and ANN modeling show a similar pattern. The dielectric properties of fiber-reinforced polyester matrix composites can be accurately predicted using ANN, reducing the need for manual intervention.