Bayesian optimization for effective thermal conductivity measurement of thermal energy storage: An experimental and numerical approach

The increasing demand for cooling and refrigeration poses an urgent need in designing efficient and low-cost thermal energy storage systems for future energy systems. While multiple effects may affect the heat transfer behaviors during thermal energy storage, these effects can be lumped into one parameter, the effective thermal conductivity. Effective thermal conductivity provides a simple and reliable solution for accurate numerical simulations in designing a thermal energy storage system. In this study, a novel experimental, numerical and Bayesian optimization-based method is developed and validated that allows for fast and accurate measurement of the effective thermal conductivities over a wide temperature range. The method can also be applied to other bulky and heterogeneous structures that cannot be considered as continuous media. An experimental setup and a 3D numerical model were developed for the plate-type thermal energy storage. After a thorough algorithm comparison, Bayesian optimization using Gaussian process was selected to search for the effective thermal conductivities with high accuracy (root mean square error < 2 K and R-squared between 0.975 and 0.992). The effective thermal conductivities measured using deionized water as the phase change material were validated by a COMSOL simulation. With the accurate effective thermal conductivity results, we revealed that neglecting the effective thermal conductivity for the solid phase while still using conduction models will lead to significant errors in the simulation. A duo arch-shaped graphite sheet-based macrofiller is designed and inserted into the plate-type thermal energy storage, which increased the effective thermal conductivities by around 20% and suppressed the subcooling effect.