"Predictive Digital Twins: From physics-based modeling to scientific machine learning" Prof. Willcox

CIS Digital Twin Days 2021 | 15 Nov. 2021 | Lausanne Switzerland
Prof. Karen E. Willcox, Director, Oden Institute for Computational Engineering and Sciences, University of Texas, Austin
Predictive Digital Twins: From physics-based modeling to scientific machine learning
Abstract
A digital twin is an evolving virtual model that mirrors an individual physical asset throughout its lifecycle. Key to the digital twin concept is the ability to sense, collect, analyze, and learn from the asset’s data. To make digital twins a reality, many elements of the interdisciplinary field of computational science, including physics-based modeling and simulation, inverse problems, uncertainty quantification, and scientific machine learning, have an important role to play.
In this work, we develop a probabilistic graphical model as a formal mathematical representation of a digital twin and its associated physical asset. We create an abstraction of the asset-twin system as a set of coupled dynamical systems, evolving over time through their respective state-spaces and interacting via observed data and control inputs. The abstraction is realized computationally as a dynamic decision network. Predictive capabilities are enabled by physics-based reduced-order models. We demonstrate how the approach is instantiated to create, update and deploy a structural digital twin of an unmanned aerial vehicle.
cis.epfl.ch

[@birukgirma4443]

thank you so much for this wonderful presentation

[@prashkd7684]

So the bottm line is that simple black box method of capturing system dynamics is not the solution. For a digital Twin, you need to apply first principle to model the system and THEN reinforce it with dynamic field data.

[@thefastandthedead1769]

Well said!

[@yaong49]

Thank you for your presentation!!!

[@foju9365]

Amazing talk

[@lucyswift1980]

amazing presentation!

[@Qatium]

Awesome video - thanks for sharing 👋

[@cbxxxbc]

Tour de force - great!

[@dlee4736]

Awesome talk

[@fslurrehman]

The term DT is gaining traction these days in research but I find it repetition of idea used in product/building life cycle. Similarly Reduced Order Model has been there in prototype-model and in phenomenological elements in finite element analysis.
I have seen that sometimes, researchers coin new terminology or buzz word that help them to publish their work as new tech.

[@kongeo7431]

Can someone explain, for prediction tasks, why should we do all of this modeling, instead of building a Deep learning model on the historic data of the physical asset? and retrain it every x amount of time to be up to date. I can understand the interpretability advantage of physics-driven, but is there any other advantage?

[@el.omondi]

yes, spatial interpratation

[@DiabolicMagicSquare]

Because prediction will need training the models? it is a dumb and brute force way to do things. Things would quickly go out of hand. You need lot of computing power.

[@67254215415413]

This is just an observer?

[@nailbalkan7991]

Vv

[@tomberger8628]

There was no machine learning in this talk.