A strategy combining machine learning and multiscale calculation to predict tensile strength for pearlitic steel wires with industrial data. (September 2020)

• Record Type: Journal Article
• Authors: Jiang, Xue; Jia, Baorui; Zhang, Guofei; Zhang, Cong; Wang, Xin; Zhang, Ruijie; Yin, Haiqing; Qu, Xuanhui; Song, Yong; Su, Lan; Mi, Zhenli; Hu, Lei; Ma, Han
• Is Part Of: Scripta materialia. Number 186(2020)
• Page Start: 272
• Page End: 277
• Publication Date: 2020-09
• Subjects: Materials -- Periodicals; Metallurgy -- Periodicals; Metalen; Legeringen; Materiaalkunde; Metals, metalworking and machinery industries; Metals; Electronic journals; 620.11
• DOI: 10.1016/j.scriptamat.2020.03.064
• Languages: English
• ISSNs: 1359-6462

Abstract

Graphical abstract: Image, graphical abstract Abstract: Manufacturing process of pearlitic steel wires involves multiple steps, generating numerous influencing parameters for tensile strength. Therefore, it is difficult to build globally-optimized tensile strength model by costly, time-consuming experimental trials or physical theoretical calculations. Here, a new strategy combining machine learning with multiscale calculation was promoted to construct tensile strength model based on high-dimension, small-size industrial datasets. Process space was transformed to microscopic structure space by thermodynamic, kinetic and finite element calculations, which was then fed to machine learning algorithms. Gradient Tree Boosting and Gaussian Process models show excellent prediction accuracy with maximum relative error less than 2.0 %.