Probabilistic Physics-Based Machine Learning for Digital Twinning Enabled by Bridging Supercomputing and Real-Time Computing
Presenter
DescriptionThis lecture will focus on the construction of a digital twin instance (DTI) – that is, a DT of an individual instance of a product after it has been manufactured and equipped with sensors that provide vital information during its deployment. It can serve many purposes, including: enabling a predictive rather than scheduled maintenance; performing reliable structural health monitoring; enabling model predictive control; and enabling operation at performance limits. Specifically, the lecture will present a feasible probabilistic framework for enriching a computational model with sensor data to reduce its model-form uncertainty; and continuously update it to enhance its predictive ability. The framework is grounded in a physics-based, machine learning approach for extracting from sensor data information that is not captured by a deterministic computational model and infusing it into a lower dimensional stochastic counterpart constructed using a randomized, projection-based model order reduction method enabled by supercomputing. As a benefit, the stochastic, projection-based reduced-order model delivers near-real-time numerical predictions in the form of confidence intervals that contain the true values of the quantities of interest, within a specified confidence level. The lecture will demonstrate the framework with applications pertaining to car crash analysis and model predictive control of autonomous aircraft landing.
TimeMonday, June 2716:30 - 17:00 CEST
LocationBoston 3 Room
SessionMS2H - HPC in Reduced Order Modelling for Advanced Mechanics Simulations and Digital Twinning
Session Chair
Event Type
Minisymposium
Chemistry and Materials
Computer Science and Applied Mathematics
Engineering
