Physics Informed Machine Learning through Symbolic Regression

Explore a novel framework using symbolic regression to identify ground truth models from scarce and noisy data in this hour-long lecture by Dr. George M. Bollas. Discover how this approach successfully identifies partial differential equation (PDE) models from time-variant data, outperforming similar methods when data is limited. Learn about the framework's robustness to noise and scarcity, successfully recovering models with up to 50% noise. Compare this approach to Physics-Informed Neural Networks (PINN) using NVIDIA's Modulus software package, and examine the benefits and drawbacks of symbolic regression versus neural networks. Explore applications in fault detection through a genetic programming algorithm that augments dynamic system models. Gain insights into Dr. Bollas' interdisciplinary research merging energy technology, process systems engineering, and model-based systems engineering, with applications in various industries including energy, chemical, manufacturing, naval, and aerospace.

DDPS | ‘Physics Informed Machine Learning through Symbolic Regression’