Accelerated Development of Refractory Nanocomposite Solar Absorbers using Bayesian Optimization
- PDF / 1,737,508 Bytes
- 9 Pages / 432 x 648 pts Page_size
- 74 Downloads / 199 Views
MRS Advances © 2019 Materials Research Society DOI: 10.1557/adv.2019.468
Accelerated Development of Refractory Nanocomposite Solar Absorbers using Bayesian Optimization Qiangshun Guan1, Afra S. Alketbi1, Aikifa Raza1, TieJun Zhang1,* 1
Department of Mechanical Engineering, Masdar Institute, Khalifa University of Science and Technology, P.O. Box 54224, Abu Dhabi, United Arab Emirates. *Correspondence: [email protected]
ABSTRACT Machine learning-based approach is desired for accelerating materials design, development and discovery in combination with high-throughput experiments and simulation. In this work, we propose to apply a Bayesian optimization method to design ultrathin multilayer tungstensilicon carbide (W-SiC) nanocomposite absorber for high-temperature solar power generation. Based on a semi-analytical scattering matrix method, the design of spectrally selective absorber is optimized over a variety of layer thicknesses to maximize the overall solar absorptance. Our nanofabrication and experimental characterization results demonstrate the capability of the proposed approach for accelerated development of refractory light-absorbing materials. Comparison with other global optimization methods, such as random search, simulated annealing and particle swarm optimization, shows that the Bayesian optimization method can expedite the design of multilayer nanocomposite absorbers and significantly reduce the development cost. This work sheds light on the discovery of novel materials for solar energy and sustainability applications.
INTRODUCTION Machine learning promises great potential in materials development and discovery for clean energy production and sustainability [1]. In materials design, a thorny problem that materials scientists commonly confront is to decide the choices of combinatorial materials, stoichiometric ratios, and fabrication conditions to achieve desired materials properties [2]. The conventional way to solve this problem is to use a trial-and-error method, where (i) several candidate designs are proposed from primary intuition or prior experiences; (ii) subsequent fabrication, characterization, and testing of these designs are perform to update our knowledge of the designs; and (iii) new designs
Downloaded from https://www.cambridge.org/core. Gothenburg University Library, on 29 Jan 2020 at 18:56:41, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1557/adv.2019.468
are proposed based on the lessons acquired from preceding experiments. This process is usually iterative until a successful design is achieved or a final failure occurs. Recently, the Bayesian optimization method, developed on the basis of Bayesian statistics and decision theory for tuning hyperparameters of machine learning algorithms [3], is attracting more attentions from scientists to enable accelerated materials design and discovery. Applications of Bayesian optimization in materials science include but are not limited to finding single- and binary-compon
Data Loading...
