3rd World Congress on Artificial Intelligence in Materials & Manufacturing (AIM 2025): Development of Novel ML Methodologies IV
Program Organizers: Remi Dingreville, Sandia National Laboratories; Ali Riza Durmaz, Fraunhofer Institute Iwm
Tuesday 3:40 PM
June 17, 2025
Room: Platinum Ballroom 7&8
Location: Anaheim Marriott
Session Chair: Paul Conway, Loughborough University
3:40 PM Invited
Strongly Physics-Constrained Neural Networks for Mechanical Superresolution: Vivek Oommen1; Andreas Robertson2; George Karniadakis1; Remi Dingreville2; 1Brown University; 2Sandia National Laboratories
Neural operators facilitate the efficient estimation of the solution fields to partial differential equations. However, these methods are restricted by the availability of sufficient data for training. This problem is especially pronounced in crystal mechanics applications because at high resolutions the computational cost of running simulations becomes overwhelming. We propose a framework for multi-resolution training of neural operators in crystal mechanics, where the training is only supervised with low-resolution datasets generated from acceptably cheap simulations. To fill in the missing information, we introduce a new variant of neural operators: Structure- Preserving UNets. These networks are strongly physics-constrained: the deformation compatibility and stress equilibrium PDEs are directly incorporated into the architecture. In this talk, we present the novel architecture that makes physically consistent and computationally efficient predictions at high resolution. To understand the proposed framework’s strengths and weaknesses, we compare it against traditional softly constrained physics-informed neural networks.
4:10 PM
Beyond Bespoke Models: Foundational Vision Transformers for Microstructure Representation and Machine Learning of Microstructure-Property Relationships in Alloys: Sheila Whitman1; Aditya Jain1; Marat Latypov1; 1University of Arizona
Machine learning is rapidly emerging as a powerful approach to establishing microstructure–property relationships in structural materials. Most existing machine learning efforts focus on the development of task-specific models for each individual class of materials and individual properties. To go beyond bespoke models, we propose utilizing foundational vision models for the extraction of task-agnostic microstructure features and subsequent lightweight machine learning. We demonstrate our approach on two case studies: stiffness of synthetic two-phase microstructures learned from simulation data and Vickers hardness of superalloys learned from experimental data. Our results show the potential of foundational vision models for robust microstructure representation and efficient machine learning of microstructure–property relationships without the need for expensive task-specific training or fine-tuning. We further explore the extension of this approach to include additional alloy information (composition, processing) besides the microstructure for multimodal representation and learning of alloy properties.
4:30 PM
Generative Priors for Regularizing Ill-Posed Problems:
Applications to 3D Polycrystalline RVE's: Michael Buzzy1; Surya Kalidindi1; Andreas Robertson2; 1Georgia Institute of Technology; 2Sandia National Laboratory
Many problems in Materials Science are ill-posed, meaning that, given a task there may be one, many, or no possible solutions. Classic examples of ill-posed problems include materials design and microstructure reconstruction, where a user may be interested in obtaining a microstructure which corresponds to a desired property or structural descriptor. When solving inverse problems, the existence (or lack thereof) of multiple solutions presents a persistent problem in high dimensional spaces, where the set of possible solutions becomes incredibly vast and difficult to enumerate. New algorithms utilizing generative priors provide a promising avenue for regularizing these high dimensional ill-posed problems. This talk will discuss the theory and benefits of generative priors, as well as demonstrate their practicality by solving materials design and microstructure reconstruction problems relating to 3D Polycrystalline RVE's.