FRIDAY SEMINAR SERIES - Physics-guided Machine Learning for Structural Dynamics

The availability of various data types and increasing amounts of data, coupled with emerging sensing technologies, artificial intelligence, and machine learning models, have strengthened the inverse modeling of structural dynamical systems. However, the use of "black-box" or purely data-driven models has limitations due to the strong dependence on sensor data. The quality, representativeness, and choice of sensor data for training highly influence the effectiveness of learned models. Moreover, the learned models lack physical interpretation of underlying systems or processes. In this talk, I will discuss leveraging domain knowledge to enhance machine learning models' effectiveness, thereby converting sensor data into interpretable information in the form of physically interpretable and generalizable models. The talk covers three main topics: (1) sparse structural system identification, a parsimonious modeling technique based on sparse regularization constrained by underlying physics; (2) physics-informed Neural ODEs for integrating physics-based models with deep learning models; and (3) full-field structural monitoring and vibration analysis using event-based cameras and physics-guided machine learning. The proposed frameworks are demonstrated through laboratory experiments and real-world case studies.