Design of Efficient Radiative Coolers with Nanoparticles via Deep Neural Networks

Highly demanded energy from conventional cooling systems overloads power grids and significantly contributes to greenhouse gas emissions. To efficiently address global warming, radiative cooling is an innovative technology that can reduce temperatures below the surrounding environment without the need for energy consumption. In a variety of applications, radiative coolers are utilized in various forms. For instance, they can be attached to vehicles or building exteriors to replace conventional cooling methods that result in environmental pollution or combined with solar cells to enhance their efficiency. Radiative coolers require strong emission within the atmospheric transparent window, which ranges from 8 μm to 13 μm. Depending on their application, radiative coolers can be engineered to be either solar-transparent, with high transmittance in the solar spectral region (0.3-2.5 μm), or solar-opaque with high reflectance in the solar spectrum. Due to the need for intricate spectral properties, the designs of diverse proposed radiative coolers are complex for both types, leading to high fabrication costs and complicated assembly processes.<br/><br/>Here, we propose an innovative inverse design methodology using a deep neural network, a type of deep learning algorithm, for creating thin-film radiative coolers. The designs for solar-transparent and solar-opaque radiative coolers consist of a polyethylene (PE) film with a thickness of 10 μm, embedded with nanoparticles based on Maxwell-Garnett effective medium theory. The nanoparticles are composed of SiO₂, TiO₂, Si₃N₄, Al₂O₃ and HfO₂, which are commonly used in radiative cooling applications. Using Maxwell-Garnett effective medium theory can suggest the appropriate material ratios to achieve the desired intricate optical constants, making it possible to find suitable materials for the design of radiative coolers. In our study, the calculation process of effective medium theory is integrated with a deep neural network to enhance design quality.<br/><br/>Deep neural networks are highly effective for designing photonic structures due to their ability to rapidly and accurately learn complex patterns. Our deep learning models, each tailored for specific purposes, identify the optimal optical constants, volume ratios of materials, and particle size distributions for the thin films. These three models are utilized in sequence to inversely design photonic structures, ensuring the desired optical properties for radiative cooling. The first model proposes optimal refractive indices for high emission within the atmospheric transparent window. Subsequently, the second model suggests the material volume ratios corresponding to proposed refractive indices. Finally, the third model, which considers optical properties within the solar spectral region, outputs the particle size distribution tailored for each type of radiative cooler. Consequently, our deep learning models can propose design variables when the desired optical properties are provided.<br/><br/>Finally, we assessed the optical and thermal performance of our inversely designed radiative coolers through optical simulations and net power calculations. The solar-transparent radiative coolers achieve high performance with 93.2% emissivity in the atmospheric transparent window and 93.3% transmittance in the solar region. In contrast, the solar-opaque radiative coolers show 94.9% emissivity in the 8 μm to 13 μm region and 90.4% reflectance in the solar region with a PE film thickness of 100 μm. Additionally, a solar cell integrated with a solar-transparent radiative cooler improves efficiency by reducing its temperature by 16.4 K during operation, with a heat transfer coefficient of 10 Wm^-2K^-1. Furthermore, the solar-opaque radiative coolers provide a cooling effect, capable of reducing the temperature by approximately 2.06 K under the same heat transfer coefficient. These two types of radiative coolers offer customized solutions for various applications.