To leverage advancements in machine learning for metallic materials design and property prediction, it is crucial to develop a data-reduced representation of metal microstructures that surpasses the limitations of current physics-based discrete microstructure descriptors. This need is particularly relevant for metallic materials processed through additive manufacturing, which exhibit complex hierarchical microstructures that cannot be adequately described using the conventional metrics typically applied to wrought materials. Furthermore, capturing the spatial heterogeneity of microstructures at the different scales is necessary within such framework to accurately predict their properties. To address these challenges, we propose the physical spatial mapping of metal diffraction latent space features. This approach integrates (i) point diffraction data encoding via variational autoencoders or contrastive learning and (ii) the physical mapping of the encoded values. Together, these steps offer a method to comprehensively describe metal microstructures. We demonstrate this approach on a wrought and additively manufactured alloy, showing that it effectively encodes microstructural information and enables direct identification of microstructural heterogeneity not directly possible by physics-based models. This data-reduced microstructure representation opens the application of machine learning models in accelerating metallic material design and accurately predicting their properties.
In finite element models (FEMs), two- or three-dimensional Representative Volume Elements (RVEs) based on a statistical distribution of particles in a matrix can predict mechanical material properties. This article studies an alternative to 3D RVEs with a 2.5D RVE approach defined by a one-plane layer of 3D elements to model the material behavior. This 2.5D RVE relies on springs applied in the out-of-plane direction to constrain the two lateral deformations to be compatible, with the goal of achieving the isotropy of the studied material. The method is experimentally validated by the prediction of the tensile stress–strain curve of a bi-phasic microstructure of the AlSi10Mg alloy. Produced by additive manufacturing, the sample material becomes isotropic after friction stir processing post treatment. If a classical plane strain 2D RVE simulation is clearly too stiff compared to the experiment, the predictions of the stress–strain curves based on 2.5D RVE, 2D RVE with no transversal constraint (called 2D free RVE), and 3D RVE simulations are close to the experiments. The local stress fields within a 2.5D RVE present an interesting similarity with 3D RVE local fields, but differences with the 2D free RVE local results. Since a 2.5D RVE simplifies one spatial dimension, the simulations with this model are faster than the 3D RVE (factor 2580 in CPU or taking into account an optimal parallel computation, a factor 417 in real time). Such a discrepancy can affect the FEM2 multi-scale simulations or the time required to train a neural network, enhancing the interest in a 2.5D RVE model.
With the ongoing development of high-throughput characterization approaches for metallic materials, full-field measurements and their rapid statistical analysis have become increasingly essential for capturing material heterogeneities and providing a comprehensive understanding of metal processing and microstructure effects. In the context of mechanical properties, understanding the influence of microstructure on the localization of plasticity at small scales is crucial for describing a material’s behavior. While experimental techniques for measuring plasticity continue to improve, there remains a significant need for rapid and statistically robust analysis methods of resulting data. Here, we propose a framework based on computer vision and event merging for the automated extraction of deformation events from high-resolution digital image correlation measurements. Notably, the proposed approach demonstrates versatility, enabling its application with different material systems, microstructure, and deformation temperature conditions ranging from room to elevated temperatures. A case study is presented to statistically evaluate the impact of processing methods, (i.e., wrought vs additive manufacturing), on the distribution of deformation events during monotonic loading of 718 alloy. Significant differences in plastic localization behavior are observed between the wrought and additively manufactured 718 materials. These differences are discussed in light of their microstructural features.
Encoding metal plasticity captured from high-resolution digital image correlation (DIC) can be leveraged to predict a wide range of monotonic and cyclic macroscopic properties of metallic materials. To capture the spatial heterogeneity of plasticity that develops in metals, latent space features describing plasticity of a small region are spatially mapped across a large field of view while maintaining the same spatial relationships as the experimental measurements. Latent space feature maps capture the complexity and heterogeneity of metal plasticity as a low-dimensional representation. These feature maps are then used to train a convolutional neural network-based model to predict monotonic and cyclic macroscopic properties. The approach is demonstrated on a large set of face-centered cubic metals, enabling rapid and accurate property prediction. The effects of hyperparameters and training strategies are analyzed, and the extension of the proposed approach to a broader range of metallic materials and loading conditions is discussed.