Mining the Correlations Between Optical Micrographs and Mechanical Properties of Cold-Rolled HSLA Steels Using Machine L
- PDF / 3,367,210 Bytes
- 17 Pages / 595.276 x 790.866 pts Page_size
- 42 Downloads / 166 Views
TECHNICAL ARTICLE
Mining the Correlations Between Optical Micrographs and Mechanical Properties of Cold‑Rolled HSLA Steels Using Machine Learning Approaches Berkay Yucel1 · Sezen Yucel1 · Arunim Ray2 · Lode Duprez2 · Surya R. Kalidindi1 Received: 4 June 2020 / Accepted: 24 August 2020 © The Minerals, Metals & Materials Society 2020
Abstract This paper demonstrates the feasibility of extracting quantitative linkages between optical micrographs and mechanical properties of cold-rolled HSLA (high-strength low alloy) steels measured in standardized tension tests. These linkages were established by bringing together modern toolsets for (i) image segmentation, (ii) rigorous statistical quantification of segmented microstructures, (iii) low-dimensional representation of microstructure statistics, and (iv) building surrogate models using emergent machine learning approaches. A salient aspect of the overall approach presented in this paper is that the extracted linkages exhibited remarkable predictive accuracy while utilizing only three features identified objectively (i.e., unsupervised) in the proposed overall workflow. Keywords Segmentation · Microstructure quantification · Structure-property linkage · Machine learning
Introduction The existence of causal relationships between a material’s microstructure and its mechanical properties is a foundational tenet in the field of materials science and engineering [1–3]. Since the material’s microstructure is most commonly documented as images (obtained from optical and electron microscopes), it is only reasonable to ask exactly how much information about the mechanical properties of the material is actually encoded in such micrographs? Clearly, the answer to this question depends on the type of image(s) collected. Typical optical micrographs obtained from metallic samples usually provide raw information on the geometry of the microstructure, i.e., information on the spatial distributions of phases/grain structures as well as their sizes and morphologies [4, 5]. They clearly do not provide any direct information on the material’s chemical composition or dislocation densities. Furthermore, they mainly reflect the consequences of the last process steps; information from the * Surya R. Kalidindi [email protected] 1
Woodruff School of Mechanical Engineering, Georgia Tech, Atlanta, USA
OCAS NV/ArcelorMittal Global R&D Gent, Ghent, Belgium
2
processing steps prior to the last one is often significantly altered or lost. Given these limitations, it is reasonable to expect that the micrographs of metal samples might be correlated to their tensile properties (e.g., yield strength, ultimate tensile strength), if one restricts attention to alloys with a fixed chemical composition and a fixed initial state (i.e., fixed set of processing conditions prior to the last processing step), and a limited window in the process space (e.g., different heat treatments after a specified thermo-mechanical processing). Therefore, the central question that needs to be answered is
Data Loading...
