|This Frontiers of Materials event will provide a forum on “physics-informed machine learning (ML)” with applications to the (multi-scale) modeling and design of material systems and manufacturing processes. While data-driven ML methods have found a wide range of applications in materials modeling, design, and manufacturing, the predictions of purely data-driven models may still be physically inconsistent due to observational biases. Recent research in materials informatics has addressed this challenge by enforcing underlying physical constitutive rules, constraints, and initial/boundary conditions in data-driven model development, thereby creating physics-informed ML models. These “domain-aware” physics-informed ML models are shown to achieve better prediction accuracy and interpretability even when trained with small datasets, in addition to faster training and improved generalization performance compared to purely data-driven approaches.
This event will welcome presentations on modeling, multi-scale modeling, and the design of materials and manufacturing processes across different length scales (ranging from the atomistic scale to the macro-scale) using physics-informed ML techniques. Studies on the integration of physics-informed ML models into experimental datasets of materials and manufacturing processes will also be of interest.
Additional topics are listed as follows:
- Application of physics-informed ML to small datasets of materials and manufacturing processes
- Application of physics-informed ML to big datasets of materials and manufacturing processes
- Transfer learning approaches combined with physics-informed ML to study physical material behavior across different length scales
- Uncertainty quantification for material systems and manufacturing processes with the aid of physics-informed ML
- Application of physics-informed ML in metal additive manufacturing
- Application of physics-informed ML in constitutive model development for materials behavior