Prediction of diffusion coefficients in fcc, bcc and hcp phases remained stable or metastable by the machine-learning methods. (15th January 2021)

• Record Type: Journal Article
• Authors: Wei, Zhenbang; Yu, Jinxin; Lu, Yong; Han, Jiajia; Wang, Cuiping; Liu, Xingjun
• Is Part Of: Materials & design. Volume 198(2021)
• Publication Date: 2021-01-15
• Subjects: Diffusion coefficient -- Melting temperature -- Elastic stiffness constant -- Machine-learning methods -- Features selection methods; Materials -- Periodicals; Engineering design -- Periodicals; Matériaux -- Périodiques; Conception technique -- Périodiques; Electronic journals; 620.11
• DOI: 10.1016/j.matdes.2020.109287
• Languages: English
• ISSNs: 0264-1275

Abstract

Abstract: Diffusion coefficient play a crucial role in material designing, and physical phenomenon explaining during the material preparation and post-treatment. However, it is unavailable in some metallic systems. In this paper, based on basic physical properties (including atom properties, lattice parameters, melting temperature, elastic stiffness constant and etc.), the diffusion activate energy model were developed by machine-learning methods. First, the melting temperature ( T m ) and elastic stiffness constant ( C ij ) models were built by machine-learning methods to fill the absent values in properties. Second, the diffusion activate energy ( Q ) model was built, and a hybrid features selection method was used to decrease features from 73 to 11 in the model. The T m, C ij and Q models showed a good predictive ability and goodness of fit. Finally, features in the models were analyzed and compared with the parameters in various prior models. This work provides further understanding on the mechanism of the melting process, elastic deformation and diffusion process. Moreover, the models could be able to provide an easy and reliable method to obtain the diffusion coefficients in bcc, fcc, and hcp alloys when they are needed but unavailable. Graphical abstract: Unlabelled Image Highlights: Machine-learning models of melting temperature, elastic constants and diffusion coefficients were constructed. A hybrid features selection method were developed to decrease the number of features in the models. Diffusion coefficients in bcc, fcc and hcp alloys remained the stable and metastable were well predicted. The effects of features on melting temperature, elastic constants and diffusion coefficients were discussed.