Data‐driven physics‐based digital twins via a library of component‐based reduced‐order models

Abstract

This work proposes an approach that combines a library of component-based reduced-order models with Bayesian state estimation in order to create data-driven physics-based digital twins. Reduced-order modeling produces physics-based computational models that are reliable enough for predictive digital twins, while still being fast to evaluate. In contrast with traditional monolithic techniques for model reduction, the component-based approach scales efficiently to large complex systems, and provides a flexible and expressive framework for rapid model adaptation-both critical features in the digital twin context. Data-driven model adaptation and uncertainty quantification is formulated as a Bayesian state estimation problem, in which sensor data is used to infer which models in the model library are the best candidates for the digital twin. This approach is demonstrated through the development of a digital twin for a 12ft wingspan unmanned aerial vehicle. Offline, we construct a library of pristine and damaged aircraft components. Online, we use structural sensor data to rapidly adapt a physics-based digital twin of the aircraft structure. The data-driven digital twin enables the aircraft to dynamically replan a safe mission in response to structural damage or degradation.

Hybrid analysis and modeling, eclecticism, and multifidelity computing toward digital twin revolution

“…Recently, we formulated an eclectic interface modeling approach as a key enabler for emerging DT technologies in many sectors [277]. However, just like any technology, it comes with its own needs and challenges [62,112,136,162,164,325]. Therefore, an accurate characterization and modeling of the interface is crucial to derive the consistent boundary conditions and upscaling laws in a large variety of scientific applications.…”

Multifidelity computing for coupling full and reduced order models

“…This is depicted in Fig 7, showing the effect of different approaches on the resulting predictions of temperature fields. The FOM trajectory corresponds to the solution of both the 2D Boussinesq equations in FOM space, then projecting the obtained fields on the basis functions of θ (see Eq (31)). For the rest, the streamfunction fields are obtained from ROM predictions and fed into FOM solver to compute the temperature fields.…”

“…This should serve and advance the applicability of the emerging digital twin technologies in many sectors [26]. However, just like any technology, it comes with its own needs and challenges [27][28][29][30][31][32]. In practice, two modeling paradigms are in order.…”

A nonintrusive nonlinear model reduction method for structural dynamical problems based on machine learning

“…It remains exciting, which algorithm suits best for which task field. In some modeling methods, ML is used to estimate model parameters [5] or applied to multiscale models [6], while others use data for model adaptation of digital twins [7].…”