Just another WordPress site - Ruhr-Universität Bochum
A microstructure-sensitive machine learning model for plasticity and strain hardening in polycrystalline metals
- Date: 25.09.2024
- Time:
- Place: MSE 2024 Materials Science and Engineering Congress and Exhibition, TU Darmstadt, Germany
Abstract
Constitutive models play a pivotal role in describing and predicting the mechanical behavior of materials under various loading conditions. Machine Learning (ML) techniques have significantly advanced the development of these models by leveraging mechanical data. A key advancement in this domain is the data-oriented formulation of the yield function—a foundational component of constitutive models—via ML. While initial yielding has been the primary focus of many ML-based approaches, the effect of strain hardening has not been widely considered. However, taking strain hardening into account is crucial for describing the deformation behavior of polycrystalline metals accurately. This study introduces an ML-based yield function formulated as a Support Vector Classification (SVC) model. This yield function incorporates strain hardening effects and is trained on a comprehensive 12-dimensional feature vector. This feature vector encapsulates stress and plastic strain components derived from crystal-plasticity finite element method (CPFEM) simulations. The simulations were performed on a three-dimensional representative volume element (RVE) comprising 343 grains. This setup aimed to replicate multi-axial mechanical testing of the polycrystal under proportional loading across 300 distinct directions. These directions were carefully chosen to guarantee comprehensive coverage of the entire stress space. This study demonstrates that the ML yield function, trained on this dataset, accurately captures not just the initial yielding but also the flow stresses within the plastic regime, exhibiting remarkable accuracy and robustness. The methodology developed in this study, which utilizes synthetic mechanical data from realistic CPFEM simulations and an ML-trained yield function incorporating strain hardening, marks a significant advance in microstructure-sensitive materials modeling and thus pave the way for obtaining digital material twins.