Uncertainty Quantification Strategies for Multi-Physics Systems and Digital Twins
February 24 — 26, 2025
This workshop will explore innovations and challenges in understanding computational methods for solving multi-physics models. Digital twins that mimic real-world dynamics can provide support in decision-making. To this end, fast surrogate models must be combined with large, time-consuming simulations, to achieve a desired outcome. Particular interest is in twins that link more than one process-based model and in investigating how to build linked emulators of such systems. This workshop brings together experts with applications in a number of different areas, working on emulation, linked emulators, and using these surrogates to assist in the construction and use of digital twins.
For any early career researchers or those looking for additional background in this area, there will be a tutorial in the afternoon on Monday, February 24, 2025 of the workshop.
If you are planning to attend only the workshop, note that the workshop dates are Tuesday, February 25 through Wednesday, February 26. (updated 12/2/24)
Lightning Talks
This workshop will include lightning talks for early career researchers (including graduate students). In order to propose a lightning session talk, you must first register for the workshop, and then submit a proposal using the form that will become available on this page after you register. The registration form should not be used to propose a lightning session talk.
The deadline for proposing has been extended toJanuary 8, 2025. If your proposal is accepted, you should plan to attend the event in-person.
Karen Veroy-Grepl
Eindhoven University of Technology
D
W
Danny Williamson
University of Exeter
Schedule
Monday, February 24, 2025
12:30-13:00 CST
Check-in and Welcome
13:00-14:30 CST
Tutorial Part One
14:30-15:00 CST
Coffee Break
15:00-16:30 CST
Tutorial Part Two
Tuesday, February 25, 2025
8:30-9:00 CST
Check in and Breakfast
9:00-9:45 CST
Digital Twins for Wind Energy and Leading Edge Erosion Detection
Speaker: Sue Minkoff (Brookhaven National Lab)
One of the main sources of renewable energy is wind, which generates tremendous power while also reducing the need for greenhouse gas-emitting traditional power sources such as hydrocarbons and coal. However, many wind turbines installed in the early 2000's are nearing the end of their lifespan, and the problem remains of how to maintain, reduce, or decommission these aging turbines in a cost efficient way. In this talk we describe a digital twin for damage detection and maintenance scheduling of wind turbines which can track the condition of a wind turbine under different operating conditions. A key concern for wind energy that contributes to power production losses and high maintenance costs is deterioration of the turbine blades over time from environmental stressors such as lighting strikes, icing and accumulation of airborne particles which can result in leading edge erosion of the blades. We will discuss surrogate modeling of the turbines and classification of levels of leading edge erosion via machine learning.
9:45-10:00 CST
Q&A
10:00-10:05 CST
Tech Break
10:05-10:50 CST
Surrogates of multi-physics models, in climate and convection, with sequential design for simulators with sharp transitions using Deep Gaussian Processes
Speaker: Serge Guillas (University College London (UCL))
We first demonstrate the embedding of a Gaussian Process emulator of high resolution convection processes within a coarse climate numerical model. It leads to a reduction of some of the well know biases in climate modelling. We then present a new type of emulator of any feed forward multi-physics system, by linking GP emulators of individual simulators, with large gains over the composite emulator of the whole system. The Deep Gaussian Process (DGP) is then presented as a surrogate that shares the structure of the linked emulator but enables the emulation of highly non-linear simulators without the knowledge of individual sub-processes. We then examine sharp changes in the outputs a computer simulator. These often indicate bifurcations or critical transitions within the investigated system, e.g. laminar v. turbulent behavior in fluid dynamics. An efficient approach that localizes these changes using DGPs with a minimal number of evaluations is introduced. We demonstrate the efficacy and efficiency of the proposed framework on the Rayleigh–Bénard convection.
10:50-11:05 CST
Q&A
11:05-11:35 CST
Coffee Break
11:35-12:20 CST
Strategic Framework for Designing Trustworthy Deep Learning Surrogate Models
Speaker: Danial Faghihi (University at Buffalo)
Deep learning models, whether serving as surrogate representations of high-fidelity simulations or directly informed by physical data, have become indispensable for alleviating the computational bottlenecks inherent in the inference and optimization tasks associated with digital twins of complex physical systems. This reliance emphasizes the critical need for rigorous validation and uncertainty quantification methods to guide model construction and facilitate trustworthy deployment in high-consequence decision-making processes. In this work, we present the Occam Plausibility Algorithm for Surrogate models (OPAL-surrogate), a systematic framework for discovering “optimal” deep learning surrogate models across the expansive model space, encompassing diverse deep learning classes, architectures, and hyperparameters. The framework leverages hierarchical Bayesian inference for the principled determination of both network parameters, hyperparameters, and model plausibility, complemented by model validation under uncertainty to assess prediction accuracy and reliability. Adhering to these principles OPAL-surrogate provides an efficient strategy for adaptively balancing the trade-off among model complexity, accuracy, and prediction uncertainty. We demonstrate the effectiveness of OPAL-surrogate through two illustrative modeling problems: the deformation of porous materials for building insulation and the simulation of turbulent combustion flow governing shear-induced ablation of solid fuels in hybrid rocket motors. Further extensions to the framework in digital twin settings, including its application to the selection of “right” models for patient-specific brain tumor treatment and for controlling the nano-manufacturing processes of self-assembled materials, will also be discussed.
12:20-12:35 CST
Q&A
12:35-13:35 CST
Lunch Break
13:35-14:20 CST
TBA
Speaker: Derek Bingham (Simon Fraser University)
14:20-14:35 CST
Q&A
14:35-14:40 CST
Tech Break
14:40-15:25 CST
Lightning talks
Tao Li (New York University), "Risk-Aware Long-Short-Term Adaptive Twining in Urban Traffic Digital Twins"
Louise Kimpton (University of Exeter), "Bayesian Hierarchical Gaussian Processes for Multi-level Models"
Lekha Patel (Sandia National Laboratories), "Real-Time Linked Emulation in Digital Twins of Complex Cyber-Physical Systems"
15:25-15:40 CST
Q&A
15:40-16:30 CST
Social Hour
Wednesday, February 26, 2025
8:30-9:00 CST
Check in and Breakfast
9:00-9:45 CST
Model Order Reduction in Digital Twins
Speaker: Karen Veroy-Grepl (Eindhoven University of Technology)
9:45-10:00 CST
Q&A
10:00-10:05 CST
Tech Break
10:05-10:50 CST
TBA
Speaker: Reinhard Laubenbacher (University of Florida, Gainesville)
10:50-11:05 CST
Q&A
11:05-11:35 CST
Coffee Break
11:35-12:20 CST
Hybridizing Differential Equations and Neural Operators
Speaker: Emil Constantinescu (Argonne National Laboratory)
We present a methodology that combines neural ordinary differential equations (NODEs) with partial differential equations (PDEs) solved by the method of lines, establishing a hybrid framework for differential equation modeling. This approach addresses computational challenges in systems requiring fine temporal and spatial resolution through the integration of machine learning techniques with traditional numerical methods. Neural networks are employed to learn continuous mappings between different scales of resolution, enhancing both accuracy and computational efficiency. The strategy discussed here leverages NODEs and partial knowledge of the system to learn source dynamics at a continuous level, offering a robust framework for approximating coupling operators and improving the performance of low-order solvers. The effectiveness of this hybrid approach is demonstrated through numerical experiments using the two-scale Lorenz 96 equation, the convection-diffusion equation, and 2D/3D Navier-Stokes equations.
12:20-12:35 CST
Q&A
12:35-13:35 CST
Lunch Break
13:35-14:20 CST
Deep Gaussian processes for estimation of failure probabilities in complex systems
Speaker: Annie Sauer Booth (North Carolina State University)
IMSI is committed to making all of our programs and events inclusive and accessible. Contact [email protected] to request disability-related accommodations.
In order to register for this workshop, you must have an IMSI account and be logged in. Please use one of the buttons below to login or create an account.