Thermal conductivity prediction with Fully Connected Neural Network

Conference: AIIPCC 2022 - The Third International Conference on Artificial Intelligence, Information Processing and Cloud Computing
06/21/2022 - 06/22/2022 at Online

Proceedings: AIIPCC 2022

Pages: 5Language: englishTyp: PDF

Authors:
Zhang, Xingxing; Li, Shaobo (State Key Laboratory of Public Big Data, Guizhou University, Guiyang, China)
Yu, Liya; Song, Qisong (College of Mechanical Engineering, Guizhou University, Guiyang, China)
Dong, Rongzhi (Department of Computer Science and Engineering, University of South Carolina, USA)
Wang, Zhongyu (Key Laboratory of Advanced Manufacturing Technology (Ministry of Education), Guizhou University, Guiyang, China)

Abstract:
In this paper a Fully Connected Neural Network (FCNN) model is developed for the prediction of the material thermal conductivity. The traditional theoretical experimental research and computational simulation could not meet the needs of scientists' exploration of new materials. In recent years, data-driven machine learning algorithm has been used in material screening and performance prediction. Thermoelectric materials have become a hot topic in the field of materials research because of their ability to realize the interactive conversion of electric energy and thermal energy, small size, light weight, long service life and environmental friendliness. The hypothesis verification method based on a large number of experimental observations and the method of system calculation using the first principle are difficult to meet the requirements for rapid discovery of new materials in the current era due to the long research period. Therefore, the use of machine learning algorithms to assist in the development of new thermoelectric materials is of great significance. In this paper, a material thermal conductivity prediction model based on Fully Connected Neural Network (FCNN) is established, which is applied to the thermal conductivity dataset in the Materials Project database. The model is evaluated by ten-fold crossverification, and good prediction results are obtained. The mapping model between descriptive and thermal conductivity of materials is established by using machine learning algorithm, which can be used for large-scale material screening and guiding experimental research.