Property Predictions for Dual‐Phase Steels Using Persistent Homology and Machine Learning. Issue 3 (20th January 2020)

• Record Type: Journal Article
• Authors: Wang, Zhi‐Lei; Ogawa, Toshio; Adachi, Yoshitaka
• Is Part Of: Advanced theory and simulations. Volume 3:Issue 3(2020)
• Page Start: n/a
• Page End: n/a
• Publication Date: 2020-01-20
• Subjects: machine learning -- microstructure–property linkage -- persistent homology -- persistent images; Science -- Simulation methods -- Periodicals; Science -- Methodology -- Periodicals; Engineering -- Simulation methods -- Periodicals; Engineering -- Methodology -- Periodicals; 507.21
• DOI: 10.1002/adts.201900227
• Languages: English
• ISSNs: 2513-0390

Abstract

Abstract: Materials informatics seeks to establish microstructure–property linkage hidden in materials. A topological analysis of persistent homology and machine learning are combined to model microstructure–property linkage for dual‐phase steels, where a descriptor of persistent images is employed to characterize the microstructure and stress–strain curves are predicted using an artificial neural network. The correlations between stress and microstructure descriptor of persistent images are estimated using sensitivity analysis, Bayesian information criterion, and the least absolute shrinkage and selection operator (LASSO), respectively. The three methods identify consistent correlations, indicating that persistent images are capable of interpreting properties. Furthermore, the established artificial neural network model exhibits good accuracy and satisfactory property prediction performance. The proposed approach is expected to provide a new avenue for materials informatics and thus promote materials research. Abstract : Metallurgical microstructure features often ignore the complexities of the microstructure in geometry and thus easily underestimate the materials' properties. Persistent homology is applied to characterize the topological microstructure features, and a descriptor of persistent image is employed to quantify the microstructure. Furthermore, microstructure–property linkage is further modeled using a machine‐learning approach.