Alireza Doostan, University of Colorado
Ryan King, NREL
Recent advances in data acquisition systems along with modern data science and machine learning techniques have fostered the development of accurate data-driven approaches, such as inverse modeling for model calibration and system identification, in physical or biological sciences. In particular, system identification, i.e., deducting accurate mathematical models from measured observations, is key to improved understanding of complex phenomena, dominant feature analysis, design of experiments, and system monitoring and control. Furthermore, the emergence of multi-fidelity approaches provides further roles for data-driven models to contribute to outer loop studies.
This mini-symposium focuses on recent developments in deriving discrepancy terms, reduced order, or full models of non-linear dynamics from simulation or experimental data using techniques such as sparse regression, deep or physics-informed neural networks, and reinforcement learning. Of particular interest are methods addressing challenges regarding measurement noise, sampling strategies, identifiability, and scalability of data-driven techniques for model extraction in complex non-linear dynamical systems. Additionally, application of the extracted models to visualization, data compression/reconstruction, real-time controls, or uncertainty quantification are encouraged.