Applicability of neural network to constitutive model under multi-axial stress field

2020 Fiscal Year Final Research Report

• Project/Area Number: 19K15136
• Research Category: Grant-in-Aid for Early-Career Scientists
• Allocation Type: Multi-year Fund
• Review Section: Basic Section 23010:Building structures and materials-related
• Research Institution: Niigata University
• Principal Investigator: Teranishi Masaki 新潟大学, 研究推進機構, 助教 (80798322)
• Project Period (FY): 2019-04-01 – 2021-03-31
• Keywords: 材料構成則 / リターンマッピングアルゴリズム / ニューラルネットワーク / 金属材料 / 多軸応力場 / 非線形硬化則

Outline of Final Research Achievements

In this study, we developed a novel constitutive model with low calculation cost and high reproducibility of hardening phenomena under multi-axial stress field. In the process, the neural network was employed because of its high calculation efficiency and high performance of regression. Firstly, the proposed method was employed the stress-strain relation under uni-axial stress field. The results obtained by the trained neural network under random loading agreed well with test data, along with a 70 % reduction in calculation time as compared with a conventional method. Moreover, the proposed method was extended to the multi axial stress field, then a 90 % reduction in calculation time was confirmed.

Free Research Field

建築構造

Academic Significance and Societal Importance of the Research Achievements

本研究では，ニューラルネットワークを用いて，多軸応力場において，硬化挙動を再現でき，かつ，計算負荷が小さい材料構成則を開発した．本手法を実装したFEMコードを用いて，これまで計算困難であった大規模解析を実用的な時間内で遂行可能となり，その結果により構造物の変形・破壊過程が明らかになり，構造物の適切な耐震性評価に繋がると期待される．