Ting-Ju Wei

Ting-Ju Wei, Tung-Huan Su, Chuin-Shan Chen

Machine learning surrogate models have emerged as a promising approach for accelerating multiscale materials simulations while preserving predictive fidelity. Among them, the Orientation-aware Interaction-based Deep Material Network (ODMN) provides a hierarchical homogenization framework in which material nodes encode crystallographic texture and interaction nodes enforce stress equilibrium under the Hill--Mandel condition. Trained solely on linear-elastic stiffness data, ODMN captures intrinsic microstructure--mechanics relationships, enabling accurate prediction of nonlinear mechanical responses and texture evolution. However, its applicability remains fundamentally limited by the absence of a parametric mapping from arbitrary microstructures to the ODMN parameter space. This limitation necessitates retraining for each new microstructure. To address this challenge, we reformulate ODMN generalization as a microstructure-to-parameter inference problem and propose the TACS--GNN--ODMN framework. The proposed framework combines a Texture-Adaptive Clustering and Sampling (TACS) scheme for texture representation with a Graph Neural Network (GNN) for inferring micromechanical equilibrium parameters. This strategy enables the construction of fully parameterized ODMNs for previously unseen microstructures without retraining. Numerical results demonstrate that the proposed framework accurately predicts nonlinear mechanical responses and texture evolution across diverse texture distributions. The predicted responses show close agreement with direct numerical simulations (DNS), highlighting the framework as a generalizable and physically interpretable surrogate model for microstructure-informed multiscale materials simulations.

Ting-Ju Wei, Wen-Ning Wan, Chuin-Shan Chen

The Deep Material Network (DMN) has emerged as a powerful framework for multiscale materials modeling, enabling efficient and accurate prediction of material behavior across different length scales. Unlike conventional data-driven approaches, the trainable parameters in DMN possess clear physical interpretations-they encode the geometric characteristics of representative volume elements (RVEs) rather than serving as purely statistical fitting parameters . By employing a hierarchical tree structure, DMN learns the homogenization behavior associated with microstructural geometry. Consequently, it can be trained exclusively on linear elastic datasets while effectively extrapolating to nonlinear responses during online prediction, making it a highly efficient and scalable approach for multiscale simulations. From a broader perspective, DMN can be viewed as a physics-informed reduced-order model that captures the essential micromechanical features governing macroscopic behavior. Its hierarchical formulation provides a compact yet interpretable representation of the RVE response, significantly reducing computational costs compared to direct numerical simulations. This review elaborates on the theoretical foundation, training methodology, and recent extensions of DMN, emphasizing its role as a unifying framework that connects data-driven learning with physically interpretable multiscale modeling.

Ting-Ju Wei, Tung-Huan Su, Chuin-Shan Chen

Multiscale simulations are indispensable for connecting microstructural features to the macroscopic behavior of polycrystalline materials, but their high computational demands limit their practicality. Deep material networks (DMNs) have been proposed as efficient surrogate models, yet they fall short of capturing texture evolution. To address this limitation, we propose the orientation-aware interaction-based deep material network (ODMN), which incorporates an orientation-aware mechanism and an interaction mechanism grounded in the Hill-Mandel principle. The orientation-aware mechanism learns the crystallographic textures, while the interaction mechanism captures stress-equilibrium directions among representative volume element (RVE) subregions, offering insight into internal microstructural mechanics. Notably, ODMN requires only linear elastic data for training yet generalizes effectively to complex nonlinear and anisotropic responses. Our results show that ODMN accurately predicts both mechanical responses and texture evolution under complex plastic deformation, thus expanding the applicability of DMNs to polycrystalline materials. By balancing computational efficiency with predictive fidelity, ODMN provides a robust framework for multiscale simulations of polycrystalline materials.

Ting-Ju Wei, Chuin-Shan Chen

We present the Material Masked Autoencoder (MMAE), a self-supervised Vision Transformer pretrained on a large corpus of short-fiber composite images via masked image reconstruction. The pretrained MMAE learns latent representations that capture essential microstructural features and are broadly transferable across tasks. We demonstrate two key applications: (i) predicting homogenized stiffness components through fine-tuning on limited data, and (ii) inferring physically interpretable parameters by coupling MMAE with an interaction-based material network (IMN), thereby enabling extrapolation of nonlinear stress-strain responses. These results highlight the promise of microstructure foundation models and lay the groundwork for future extensions to more complex systems, such as 3D composites and experimental datasets.