Prediction on effective thermal conductivity of composite phase change materials from mesoscale structure

This study proposes a numerical approach combined Quartet Structure Generation Set (QSGS) method with machine learning to investigate the effective thermal conductivity of composite phase change materials (CPCM). The QSGS method is utilized to construct the mesoscale structure of the CPCM, enabling the determination of the effective thermal conductivity (k eff ) through finite element simulations . The generated structures reveal the formation of three-dimensional thermal conduction network, influenced by key structural parameters such as porosity, solid diameter, and core growth probability. Machine learning-based predictive models are developed to estimate k eff and are compared against finite element simulation results. The results showed that a 10 % increase in porosity decreased the k eff by about 8.45 %. The k eff of CPCM increases with increasing temperature. The k eff can be increased by increasing the solid diameter, and the k eff decreases with the decrease of C dd . Among the machine learning models tested, the extreme gradient boosting model demonstrated superior performance, achieving a prediction accuracy with a correlation coefficient of 96 % relative to finite element results . These findings provide valuable insights into the thermal conductivity behavior and heat transfer mechanisms in composite phase change materials . • Thermal conductivity network structure of CPCM was established using QSGS method. • The meso-structural parameters on effective thermal conductivity was studied. • The prediction model of effective thermal conductivity was proposed.