Research Progress of Artificial Neural Networks in Material Science
KANG Jing1, MI Xiaoxi1, WANG Hailian1, WU Lu2, SUN Kai2, TANG Aitao1,
1 College of Materials Science and Engineering, Chongqing University, Chongqing 400045, China 2 Nuclear Power Institute of China, Chengdu 610005, China
Abstract: The traditional “trial and error” material research methods have the disadvantages of long cycle, high cost and great contingency, which can't meet the needs of modern material research and development. It has become a research hotspot all over the world to improve the pertinence of research and development, shorten the cycle of material research and development, and reduce the cost of material research and development. With the continuous accumulation of data and the continuous development of computer technology, data intensive material scientific discovery has gradually become the fourth paradigm of scientific research. It is the current research trend of materials to find the “gene” that can reflect the material's intrinsic characteristics from a large number of data. The artificial neural network method is widely used in the field of materials science because of its advantages of self-learning, associative storage and high-speed search for optimal solution. Researchers use machine learning models such as artificial neural network to mine the experimental or theoretical calculation data of materials under the guidance of expert experience and theory, it can be transformed into reliable knowledge and can assist intelligent decision-making, so as to establish a one-to-one mapping relationship between microstructure and macro performance of materials. In the early days, artificial neural network was mainly used to find the relationship between the macro parameters of materials and the macro performance of materials, such as the composition design of materials, the optimization of process parameters, and the search of environmental para-meters that affect the performance of materials. With the development of computer technology and the rise of computer simulation, artificial neural network was used to learn the calculation results of the first principle. It is used to describe the interaction between the systems at the atomic scale, so as to achieve the balance between the calculation speed and accuracy. The convolution neural network methods, such as a deep neural network method, has unique advantages in image processing, which makes it more widely used in the field of material characterization, such as microstructure identification and reconstruction in SEM and TEM. With the help of artificial neural network and other methods, it is a possible way to realize the ultimate goal of material design to realize the cross scale relationship between micro, meso and macro properties of materials. This paper reviews the development history of artificial neural network, explains the principle of BP neural network and convolution neural network which are widely used in the field of materials at present, summarizes the application of artificial neural network in the field of material macro performance, calculation simulation, material characterization, etc., probes into the shortcomings of the application of artificial neural network in the field of materials, and finally discusses the development trend of artificial neural network in materials research.
康靓, 米晓希, 王海莲, 吴璐, 孙凯, 汤爱涛. 人工神经网络在材料科学中的研究进展[J]. 材料导报, 2020, 34(21): 21172-21179.
KANG Jing, MI Xiaoxi, WANG Hailian, WU Lu, SUN Kai, TANG Aitao1,. Research Progress of Artificial Neural Networks in Material Science. Materials Reports, 2020, 34(21): 21172-21179.
Huang N C. Application of artificial neural network in rubber damper design. Master's Thesis, Qingdao University of Science and Technology, China, 2018(in Chinese).
Yazdanmehr M, Anijdan S H M, Bahrami A.Computational Materials Science, 2009, 44(4), 1218.
[25]
Zhu H W. Research of the nonlinear identification algorithm for the thermal process system. Master's Thesis, North China Electric Power University, China, 2018(in Chinese).
朱洪伟. 热工系统的非线性辨识算法研究. 硕士学位论文, 华北电力大学, 2018.
[26]
Armaghani D J, Shoib R S N S B, Faizi K, et al.Neural Computing & Applications, 2017, 28(2), 391.
[27]
Hinton G E, Salakhutdinov R R.Science, 2006, 313(5786), 504.
[28]
Zhou Z H. Machine learning, Tsinghua University Press, China, 2016(in Chinese).
周志华. 机器学习, 清华大学出版社, 2016.
[29]
Liu Y, Zhou X L, Hu Z K, et al.China Forestry Science and Technology, 2019, 4(1), 115(in Chinese).