TUTORIAL-3
TITLE: Predictive Digital Twins: From Design to Deployment
Contributors: Michael Kapteyn, Anirban Chaudhuri, Sebastian Henao-Garcia, Prof. Karen Willcox
ABSTRACT:
Predictive digital twins have the potential to underpin intelligent automation across engineering, science, and society by enabling asset-specific data-driven analysis, prediction, and control or decision making. This is an embodiment of the the dynamic data-driven application systems (DDDAS) paradigm: asset-specific data are dynamically acquired and integrated into predictive models, which in-turn are dynamically executed to support the decision-support or automated control task at hand.
While the promise of predictive digital twins is now widely acknowledged, their implementation is generally limited to bespoke proofs-of-concept that require significant expertise and resources to design, deploy, and operate. One reason for this is the lack of established approaches and software tools for designing and deploying predictive digital twins in a way that is both generalizable and scalable. In this tutorial, we will present mathematical foundations, conceptual frameworks, and software tools that have been recently developed in the Willcox Research Group for the principled design and deployment of predictive digital twins.
We will start by focusing on early-stage design of predictive digital twins. Advancements in computational technologies, including reduced-order modeling, uncertainty quantification, and Scientific Machine Learning, offer significant potential for improving the accuracy, efficiency, and robustness of digital twins. However, the existence of these technologies presents the designer with tradeoffs across multiple axes, such as investment in technology development, computational costs, implementation complexity, accuracy, reliability, and robustness [1]. Understanding and effectively exploring these tradespaces is critical for ensuring that the resulting digital twin architecture is fit-for-purpose [2]. To this end, we will discuss theoretical frameworks [3] and software tools for defining, evaluating and optimizing a predictive digital twin architecture. Next, we will present a software platform and python SDK for implementing and deploying the resulting digital twin architecture in a way that is modular, asynchronous, flexible, and scalable.
Case studies and demonstrations will draw on various ongoing projects within the Willcox Research Group, with applications including unmanned aerial vehicles, space systems, civil engineering structures, and cancer patients.
References:
[1] Ferrari, A. and Willcox, K., 2024. Digital twins in mechanical and aerospace engineering. Nature Computational Science, 4(3), pp.178-183.
[2] National Academies of Sciences, Engineering, and Medicine, 2023. Foundational Research Gaps and Future Directions for Digital Twins. National Academies Press, Washington, D.C.
[3] Kapteyn, M.G., Pretorius, J.V. and Willcox, K.E., 2021. A probabilistic graphical model foundation for enabling predictive digital twins at scale. Nature Computational Science, 1(5), pp.337-347.
INSTRUCTOR: Dr. Michael Kapteyn
Dr. Michael Kapteyn is a research associate at the Oden Institute for Computational Engineering and Sciences within the University of Texas at Austin. He completed his PhD in Computational Science and Engineering at MIT under the supervision of Prof. Karen Willcox. Dr. Kapteyn’s research focuses on developing the mathematical and computational foundations necessary to enable predictive digital twins at scale. He is also interested in translating this research into industry application, with application areas of interest ranging from engineering systems to cancer patients. His digital twin research has been featured as a Nature Computational Science editors pick, recognized by AIAA with best paper awards, and has been covered by numerous media outlets. Michael holds a Bachelor of Engineering from the University of Auckland, and an SM in Aeronautics and Astronautics from MIT.