Integrating data mining and machine learning to discover high-strength ductile titanium alloys

Chengxiong Zou,Jinshan Li,William Yi Wang,Ying Zhang,Deye Lin,Ruihao Yuan,Xiaodan Wang,Bin Tang,Jun Wang,Xingyu Gao,Hongchao Kou,Xidong Hui,Xiaoqin Zeng,Ma Qian,Haifeng Song,Zi-Kui Liu,Dongsheng Xu

DOI: https://doi.org/10.1016/j.actamat.2020.10.056

IF: 9.4

2021-01-01

Acta Materialia

Abstract:<p>Based on the growing power of computational capabilities and algorithmic developments, with the help of data-driven and high-throughput calculations, a new paradigm accelerating materials discovery, design and optimization is emerging. Titanium (Ti) alloys have been chosen herein to highlight an integrated computational materials engineering case study with the aim of improving their strength and ductility. The electronic properties of elemental building blocks were derived from high-throughput first-principles calculations and presented in the form of the Mendeleev periodic table, including their electron work function (Ф), Fermi energy (EF), bonding charge density (Δρ), and lattice distortion energy. The atomic and electronic insights of the composition–structure–property relationships were revealed by a data mining approach, addressing the key features/principles for the design strategies of advanced alloys. Guided by defect engineering, the deformation fault energy and dislocation width were treated as the dominating criteria in improving the ductility. The proposed yield strength model was utilized quantitatively to present the contributions of solid-solution strengthening and grain refinement hardening. Machine learning was used collaboratively with fundamental knowledge and feed back into a new training model, shown to be superior to the empirical molybdenum equivalence method. The results draw a conclusion that the integration of data mining and machine learning will not only generate plausible explanations and address new hypotheses, but also enable the design of strong and ductile Ti alloys in a more efficient and cost-effective way.</p>

materials science, multidisciplinary,metallurgy & metallurgical engineering

What problem does this paper attempt to address?

This paper aims to discover high-strength ductile titanium alloys by integrating data mining and machine learning methods. With the enhancement of computational power and the development of algorithms, the discovery, design, and optimization of materials are entering a new paradigm. The study takes titanium alloys as an example, utilizing high-throughput first-principles calculations to obtain the electronic properties of element building blocks, and presenting them in the form of the Mendeleev periodic table, including the electronic work function (Ф), Fermi level (EF), bond charge density (Δρ), and lattice distortion energy, etc. The relationships between composition, structure, and performance are revealed by data mining methods, with a focus on the key features and principles for designing advanced alloys.

The paper proposes deformation energy and dislocation width as the main criteria for improving ductility by utilizing defect engineering. The yield strength model proposed quantitatively demonstrates the contributions of solid solution strengthening and grain refinement hardening. Machine learning is combined with fundamental knowledge to form new training models, which outperform traditional methods like the Mo equivalent method. The results indicate that the combination of data mining and machine learning can not only generate reasonable explanations and new hypotheses but also design high-strength and ductile titanium alloys in a more efficient and cost-effective manner.

Keywords: high-throughput calculations, machine learning, electronic work function, similar atomic environments, bond charge density.