Comparative analysis of artificial neural networks and adaptive neuro-fuzzy inference system for biocomposite material synthesis and property prediction

Biocomposite materials (BMs) are becoming increasingly prevalent in modern applications. Estimating their production values involves various techniques, depending on the proportions of materials used. Among these techniques, artificial neural networks (ANN), fuzzy logic, statistical methods, and the adaptive neural fuzzy inference system are prominent. In this study, polyester biocomposites have been synthesized experimentally by adjusting the quantities of methyl ethyl ketone peroxide (MEKP), cobalt octoate (Co Oc) metal catalyst, marble factory waste, modified castor oil (MCO), and polyester raw material (UP) in specific ratios. The testing and analysis of these materials are conducted to determine parameters such as bulk density (BD), thermal conductivity coefficient (TCC), and activation energy (Ea). Subsequently, input and output values of the BMs are obtained, and ANN and adaptive neuro-fuzzy inference system (ANFIS) methods are employed for assessment. Both networks are trained and modeled using experimental data to construct their respective architectures. Validation of the models has been performed using data separate from the training set. A comparison between the actual values and those predicted by the network architectures revealed that the ANN method yielded outcomes with an average error of 0.3849 %, outperforming ANFIS. The findings showed that while ANFIS produced superior predictions for the Ea output value, the ANN structure fared better in predicting output values the BD and TCC.