Prediction of Dielectric Properties of Cement-Graphite Mixture using Neural Network Models

Abstract

Prediction of dielectric properties of the cementgraphite mixture has been conducted based on two neural network models namely Radial Basic Function and Multilayer Perceptron. The prediction based on neural network is easier as it does not involve complicated empirical expressions and also it does not require special details such as the volume fraction, and dielectric constant of each element in the mixture which is necessary for dielectric mixing models. Dielectric measurement based on APC 7 connectors is carried out to obtain the dielectric properties of the cement-graphite mixture with different percentage of graphite for training purpose. The comparison of predicted results shows that Radial Basic Function introduces less error in prediction compare to Multilayer Perceptron. The discrepancy presented by the Radial Basic Function is less than 0.025 and 0.16 for dielectric constant and loss factor respectively in the frequency range between 100 MHz to 2000 MHz.