Artificial neural network prediction of unconstrained close-contact melting of phase change materials within spherical capsules

Close-contact melting is a multiphase multiscale phenomenon that occurs during unconstrained melting of phase change materials (PCM) in a heated capsule and is characterized by the formation of a liquid film between the sinking/rising solid PCM and the heated wall. This research investigates the utilization of the artificial neural network (ANN) to predict the melting rate of PCM with and without nano-additives during the close-contact melting process within a spherical capsule. The main parameters controlling the close-contact melting process, namely the capsule size, the heated wall temperature, and the thermophysical properties of the PCM, are accounted for in the ANN model through the introduction of six dimensionless groups: Fourier (Fo), Stefan (Ste), Grashof (Gr), Galileo (Ga), Archimedes (Ar), and Prandtl (Pr). To train the neural network, a comprehensive dataset comprising over 1000 data points from 50 different experimental studies reported in the literature was employed. The ANN model proved successful in predicting the quantitative and qualitative influence of the control parameters. Most notably, it was found that the PCM melting rate accelerates by increasing Ste, Gr, Ga, and Ar numbers, and by decreasing Pr number, however, the extent of their respective impacts varies significantly. This work demonstrates the usefulness of ANN for the analysis of problems that are challenging to simulate using traditional computational methods due to the presence of fluid-solid interactions and multiphase multiscale features such as melting and thin liquid film formation.