Development of low melting point alloy with high latent heat based on predictive models and standardized experiments

This study addresses discrepancies in existing literature by implementing standardized sample preparation procedures and characterization methods, thereby establishing the most comprehensive phase change point database for eutectic low-melting-point alloy (LMPA) to date. A machine learning-based (random forest) phase change point prediction model was developed, effectively overcoming the practical limitations of traditional activity coefficient-based approaches. Furthermore, a latent heat prediction model incorporating melting entropy, mixing entropy, and the solid–liquid heat capacity difference was proposed, demonstrating superior accuracy for quaternary and quinary systems. Guided by these models, two high-performance LMPA (HLHD65 and HLHD76) were designed: HLHD65 achieves 10.31 % and 10.54 % higher mass and volumetric latent heat density, respectively, than Indalloy 140, while HLHD76 demonstrates improvements of 20.13 % and 20.07 % over Wood's metal. This work provides a robust foundation and establishes a paradigm for developing advanced thermal energy storage materials.