Experimental Investigation and Artificial Neural Network-Based Prediction of Thermal Conductivity of Metal Oxide-Enhanced Organic Phase-Change Materials

Abstract

This research article presents a comprehensive study on the prediction of thermal conductivity (TC) as a primary outcome for an artificial neural network (ANN) model in the context of nanoenhanced phase change materials (NEPCMs). To improve predictive accuracy and to reduce variation within the NEPCM dataset, a targeted dataset was employed, consisting exclusively of NEPCMs synthesized using paraffin wax (PW) and metal oxide nanoparticles. Unlike existing empirical models that predict TC of NEPCM without simultaneously considering multiple factors influencing it, this study integrates multiple factors, providing a more accurate prediction of NEPCM thermal conductivity. Additionally, the study explores the impact of synthesis parameters on the thermal performance of NEPCMs, focusing on the examination of factors such as the melting temperature of pure phase change material (PCM), nanoparticle size, and NEPCM composition. The thermal characterizations demonstrate outstanding thermophysical properties in NEPCMs, particularly in terms of thermal conductivity, phase change enthalpy, and thermal stability compared to their respective base PCM. An ANN TC prediction model demonstrates exceptional correlation (>99%) with reported NEPCMs, providing a reliable tool for TC forecasting in similar NEPCM categories. The backpropagation ANN model predicts NEPCM TC with a mean squared error (MSE) of 0.031124 within eight epochs. The dataset used exhibits high fit values, with R-values of 0.99825, 0.99208, and 0.9824 for training, validation, and testing, respectively. These values closely match experimentally determined TC, with less than 4% error.