Physics-Informed AI

Our research in physics-informed AI focuses on inverse problems in which important physical properties cannot be measured directly but must be inferred from indirect, noisy, and limited observations. We establish physics-informed machine-learning methods that embed governing equations, such as partial differential equations from mechanics, into deep-learning models to estimate latent heterogeneous physical fields. A major application is the estimation of spatially varying bone stiffness and elastic moduli from deformation data using linear elasticity and physics-informed neural networks. This line of work aims to make AI more reliable for scientific and engineering inference by combining data, domain physics, and uncertainty-aware computational modeling.

Physics-Informed Machine Learning to estimate heterogeneous stiffness (elastic moduli) of bones based on the bone deformation data
- Goal: To infer the unobserved underlying elastic properties based on observed noisy deformation data using the law of physics (partial differential equation; PDE).
- Methodologies: Physics-Informed Neural Network, deep neural network
- PDE Equation: Linear Elasticity with heterogeneous elastic moduli

This work is done with a Ph.D. students, T. Srikitrungruang, S Aghaee, an undergraduate student, M. Lemon, and Collaborator, Dr. Yuxiao Zhou.
[J9] T. Srikitrungruang, M. Lemon, S. Aghaee Dabanghan Fard, J. Lee, Y. Zhou, "Robust Physics-Informed Neural Network Approach for Estimating Heterogeneous Elastic Properties from Noisy Displacement Data", Advanced Science, Accepted, doi.org/2025, 10.1002/advs.202508445.
- Advanced Science is a premier interdisciplinary journal with high impact (IF > 14).

Google Sites

Report abuse