Machine learning guided automatic recognition of crystal boundaries in bainitic/martensitic alloy and relationship between boundary types and ductile-to-brittle transition behavior. (10th September 2021)

• Record Type: Journal Article
• Authors: Li, X.C.; Zhao, J.X.; Cong, J.H.; Misra, R.D.K.; Wang, X.M.; Wang, X.L; Shang, C.J.
• Is Part Of: Journal of materials science & technology. Volume 84(2021)
• Page Start: 49
• Page End: 58
• Publication Date: 2021-09-10
• Subjects: Machine learning -- Feature engineering -- Automatic recognition -- Lath structure -- Crystallography; Metals -- Periodicals; Materials science -- Periodicals; Materials science; Metals; Periodicals; 620.1105
• DOI: 10.1016/j.jmst.2020.12.024
• Languages: English
• ISSNs: 1005-0302

Abstract

Graphical abstract: Highlights: Machine learning was employed to microstructure recognition of bainitic steels. The model has good accuracy and very high efficiency. Deep understanding of coherent transformation mechanism is not essential. This approach can be extended to other metallic materials. Abstract: Gradient boosting decision tree (GBDT) machine learning (ML) method was adopted for the first time to automatically recognize and conduct quantitative statistical analysis of boundaries in bainitic microstructure using electron back-scatter diffraction (EBSD) data. In spite of lack of large sets of EBSD data, we were successful in achieving the desired accuracy and accomplishing the objective of recognizing the boundaries. Compared with a low model accuracy of <50 % as using Euler angles or axis-angle pair as characteristic features, the accuracy of the model was significantly enhanced to about 88 % when the Euler angle was converted to overall misorientation angle (OMA) and specific misorientation angle (SMA) and considered as important features. In this model, the recall score of prior austenite grain (PAG) boundary was ∼93 %, high angle packet boundary (OMA>40°) was ∼97 %, and block boundary was ∼96 %. The derived outcomes of ML were used to obtain insights into the ductile-to-brittle transition (DBTT) behavior. Interestingly, ML modeling approach suggested that DBTT was not determined by the density of high angle grain boundaries, but significantly influenced by the density of PAG and packet boundaries. The study underscores that ML has a great potential in detailed recognition of complex multi-hierarchical microstructure such as bainite and martensite and relates to material performance.