image: a, Schematic of three typical applications in emissivity engineering, including thermal camouflage, radiative cooling, and gas sensing. The basic elements of Deep Q-learning network (DQN)-based framework, including materials and structures for inputs and action decisions for outputs. b, Schematic of the DQN framework. The state consists of two materials and five layers of thickness of the multilayer, then the state parameters are fed into the DQN to generate an Action. Then take the action to update the state. Transfer matrix method (TMM) is adopted to simulate the new state, and reward is obtained to feed back to neural network. The new state is fed into the DQN for next iteration. Each pair of state, action and reward is recorded as dataset to train the neural network, so that it can take the action that increases accumulated reward and finally get the corresponding state with the maximum reward. c, Emissivity spectrum of the WS-TE optimized for thermal camouflage. The reward is defined as the difference between the average emissivity outside and inside the atmospheric window (8-13 μm). d, Optimization process of WS-TE for thermal camouflage. Maximum of reward as a function of the percentage of calculated structures. Only 4.428% of the structures are calculated to find the optimal structure for thermal camouflage.
Credit: by Shilv Yu, Peng Zhou, Wang Xi, Zihe Chen, Yuheng Deng, Xiaobing Luo, Wangnan Li, Junichiro Shiomi, and Run Hu
Wavelength-selective thermal emitters (WS-TEs) have been frequently designed to achieve desired target emissivity spectra, as a typical emissivity engineering, for broad applications such as thermal camouflage, radiative cooling, and gas sensing, etc. However, previous designs require prior knowledge of materials or structures for different applications and the designed WS-TEs usually vary from applications to applications in terms of materials and structures, thus lacking of a general design framework for emissivity engineering across different applications. Moreover, previous designs fail to tackle the simultaneous design of both materials and structures, as they either fix materials to design structures or fix structures to select suitable materials.
In a new paper published in Light Science & Application, a team of scientists, led by Professor Run Hu form School of Energy and Power Engineering, Huazhong University of Science and Technology, China, and co-works have proposed a general deep learning framework based on deep Q-learning network algorithm (DQN) for efficient optimal design of WS-TEs across different applications. Employing this framework, they designed three multilayer WS-TEs for thermal camouflage, radiative cooling and gas sensing, respectively. The materials of the WS-TEs are autonomously selected by DQN algorithm from the same common material library according to the target emissivity spectra of different applications and the structural parameters are optimized simultaneously. The three designed WS-TEs all presents excellent performance, which are experimentally fabricated, measured and the actual emissivity spectra match well with the target one. As such, the proposed framework is demonstrated to be feasible and efficient in achieving reverse design of WS-TEs within a vast optimization design space. More importantly, it offers a general framework for emissivity engineering across different applications and paving the way for efficient design of nonlinear optimization problems beyond thermal metamaterials.
The proposed framework is a general design approach for emissivity engineering that is highly scalable across the design parameters of the WS-TMs, including material, structure, dimension, and target function. The core of the framework is that the DQN algorithm can receive various design parameters and output a decision to update the parameters. In the continuous iterative update, DQN gradually learns how to make appropriate decisions to finally achieve the optimal design. These scientists summarize the advantage of their proposed framework:
“The merits of the deep Q-learning algorithm include that it can 1) offer a general design framework for WS-TEs beyond one-dimensional multilayer structures; 2) autonomously select suitable materials from a self-built material library and 3) autonomously optimize structural parameters for the target emissivity spectra.”
“Considering the 8 available materials, this structural configuration leads to 8×7×505= 1.75×1010 potential candidate structures. The demand of simultaneous material selection and structure optimization, together with the sheer volume of optimization space, renders manual design impractical and presents significant challenges to conventional machine learning methods.” they added.
“Additionally, the input parameters of the DQN framework are highly flexible in materials, structures, dimensions, and the target functions, paving the general solution to other nonlinear optimization problems beyond emissivity engineering.” the scientists forecast.
Journal
Light Science & Applications