Probabilistic Digital Twins for Predicting Chaotic Bifurcations in High-Speed Aeroelasticity
PI: Ashwin Renganathan (Aerospace)
This project focuses on digital twin technology development by contributing to its mathematical and statistical foundations. The overarching goals of this project include the development of mathematical, statistical, and computational approaches for probabilistic digital twins for complex engineered systems. Digital twins are a digital representation of a physical asset that can interface with the physical asset to assimilate new information to update itself. In this project, we consider chaotic bifurcations typical in high-speed flows, particularly in the hypersonic regime, displaying strong fluid-thermal-structural interactions (FTSI). Specifically, we consider aeroelastic flutter and the onset of limit cycle oscillations (LCO) in hypersonic vehicles, which can have a catastrophic effect if they occur during flight. A probabilistic digital twin method can accurately predict the onset of post-flutter LCO, the associated path to divergent oscillations, and identify optimal operating conditions that can potentially avoid flutter.
Our intellectual merits are anchored on two fronts. First, we propose a novel nonlinear model calibration approach with active learning that can efficiently calibrate a biased physics model with measurement data. This method formulates a multilayered generalization of the well-established Kennedy-O’Hagan formalism for model calibration. Full high-fidelity models (ROMs) for high-speed FTSI don’t exist, and reduced-order models are inadequate. Our proposed approach calibrates the ROMs with limited scope experimental measurement, thereby bridging the gap between ROMs and high-fidelity models. Second, we propose fast optimization algorithms that show promise of faster convergence and reduced computation compared to existing methods in the state of the art. This builds on calibrated models developed with our calibration approach and, critically, enables real-time decision-making for the digital twin. The fast optimization method will be used to provide feedback coupling to the physical twin thereby closing the loop on this project.
List of expected skills. Python programming, high-performance computing (usage of SLURM, MPI4Py), PyTorch, GPyTorch, and BoTorch.
List of specific objectives for work supported by this call. The graduate student will spend 50% of their time working on this project. This include (i) conducting research, (ii) service to ICDS, and (iii) involvement in collaboration with other ICDS faculty. We will use this effort to generate preliminary data that will then be used to submit proposals to DOD agencies (AFOSR and ONR) as well as manuscripts for publication.
Connection to ICDS mission. We will develop probabilistic AI/ML methods to reduce, interpret, and learn data. This project will include both large-scale data generation by running f inite-volume based multiphysics codes on Roar Collab, as well as developing AI/ML methods on that data with GPU acceleration