Constitutive Model Discovery from Physics-Enforced Neural Networks

Research Project, 2026 – 2030

When adopting new high-performance materials to build stronger, lighter, and more sustainable parts, we need constitutive models that accurately describe how these materials behave to utilize their full potential. However, their complex behavior under demanding operating conditions presents a significant modeling challenge. This project addresses that challenge by combining machine learning with constitutive modeling, while ensuring adherence to physical laws. Although the primary focus is on large plastic deformations, the developed methods will be broadly applicable for modeling both mechanical responses and other physical processes.Together with a PhD student, the PI will address three key challenges in combining constitutive modeling and machine learning: how to ensure adherence to physical constraints when enhancing complex constitutive models with neural networks; how to extract interpretable equations from these models; and how to train them when experimental data is scarce.Solving these challenges will accelerate research and adoption of new high-performance materials. Additionally, the modeling approaches will benefit research fields such as biomechanics where the behavior is complex and physical testing is limited. In summary, this research program lays the foundation for several future research challenges and projects.