Experimental study and ANN-based prediction of melting heat transfer in a uniform heat flux PCM enclosure

Abstract

Abstract In this paper, heat transfer during the melting process of n-octadecane as a phase change material (PCM) is experimentally studied. This study is followed by an artificial neural network (ANN) to predict the melting characteristics of PCM. Experiments are performed in a rectangular enclosure subjected to a uniform heat flux in one vertical side. Melting heat transfer is characterized by observing the solid-liquid interface and recording the temperature distribution in the enclosure. Experimental results indicate that heat transfer during the melting process is dominated by natural convection. A multilayered perceptron feed-forward neural network trained by the Levenberg-Marquardt algorithm is used to predict the Nusselt number and the melted volume fraction. Rayleigh, Fourier and Stefan numbers are set as input parameters of the network. The optimal structure of the ANN to predict the Nusselt number show high accuracy in estimating the heat transfer characteristics during melting by achieving the mean square error and the correlation coefficient of 4.42 × 10−6 and 0.999, respectively. Based on the proposed ANN, the majority of the data falls within ±6.23% and ±6.54% of the Nusselt number and the melted volume fraction, respectively.