Machine-Learning-Assisted Design of a Robust Biomimetic Radiative Cooling Metamaterial

Abstract

Recently, biomimetic photonic structural materials have significantly improved their radiative cooling performance. However, most research has focused on understanding cooling mechanisms, with limited exploration of sensitive parameter variations. Traditional numerical methods are costly and time-consuming and often struggle to identify optimal solutions, limiting the scope of high-performance microstructure design. To address these challenges, we integrated machine learning into the design of Batocera LineolataHope bionic photonic structures, using SiO 2 as the substrate. Deep learning models provided insights into the complex relationship between bionic metamaterials and their spectral response, enabling us to identify the optimal performance parameter range for truncated cone arrays (height-to-diameter ratio (H/D bottom ) from 0.8 to 2.4), achieving a high average emissivity of 0.985. Experimentally, the noon temperature of fabricated samples decreased by about 8.3 °C. This data-driven approach accelerates the design and optimization of robust biomimetic radiative cooling metamaterials, promising significant advancements in standardized passive radiative cooling applications.