核沸騰伝熱性能の回帰学習に基づく伝熱面の最適化手法の構築

抄録

Boiling-based two-phase heat transfer, which leverages latent heat transport associated with phase change, has been widely applied in cooling technologies for power plants, air conditioning systems, and electronic devices due to its high heat transfer coefficient. However, with the increasing performance and miniaturization of electronic devices, power density has surged, leading to escalating cooling costs. To address this issue, research has focused on improving heat transfer performance by promoting boiling nucleation through microscale surface modifications. Nevertheless, the numerous design parameters involved make trial and error a time-consuming process. In this study, a heat transfer coefficient prediction model is built using machine learning trained on diverse existing experimental data, and a sensitivity analysis is conducted to evaluate each input’s contribution to the output. This approach suggests the potential for rapidly and accurately determining design guidelines for surface modifications.