A new mathematical model allows the virtual design representation known as a "digital twin" to be used in a range of engineering systems, from spacecraft to entire cities.
A digital twin is a computational model that evolves over time and continuously represents the structure, behavior, and context of a unique physical “asset," like a bridge, vehicle, or any object that needs monitoring really.
Through advanced mathematical modeling techniques, sensors, and supercomputers, experts from the Oden Institute for Computational Engineering and Sciences, Massachusetts Institute of Technology (MIT), and industry partner the Jessara Group improved the capabilities of the digital twin.
The researchers tested out their idea on an unpiloted aerial vehicle (UAV), creating a structural digital twin of a custom-built unmanned aerial vehicle instrumented with state-of-the-art sensing capabilities.
The team's study appeared in the March 2021 journal Nature Computational Science .
To test their model, the team outfitted a 12-foot-wingspan UAV with sensors from The Jessara Group. The "stickers" collected strain, acceleration, and other relevant data from the UAV, which then informed the virtual representation.
As the overall state of the UAV changes over time, the digital twin updates its own state so that it matches the physical aircraft.
According to the study, the digital twin was able to analyze sensor data to extract light-damage information, predict how the structural health of the UAV would change in the future, and recommend changes in its maneuvering to accommodate those changes.
With a combination of mathematical monitoring and sensors, the technology can potentially be applied beyond unmanned aircraft to any applications that experience wear-and-tear and require regular checks, including wind turbines, a bridge, or a nuclear reactor.
“The value of integrated sensing solutions has been recognized for some time, but combining them with the digital twin concept takes that to a new level,” said Jacob Pretorius, chief technology officer of The Jessara Group and co-author on the Nature paper . “We are on the cusp of an exciting future for intelligent engineering systems.”
The study was funded by the Air Force Office of Scientific Research, the SUTD-MIT International Design Centre, and the Department of Energy Advanced Scientific Computing Research program.
The team's approach provides a unifying mathematical representation of the relationship between a digital twin and its associated physical asset – one that is not specific to a particular application or use. The researchers’ model mathematically defines a pair of physical and digital dynamic systems, coupled together via two-way data streams as they evolve over time. In the case of the UAV, for example, the parameters of the digital twin are first calibrated with data collected from the physical UAV so that its twin is an accurate reflection from the start.
The research could help make the use of digital twins more widespread, says Karen Willcox, MIT visiting professor and director of the Oden Institute for Computational Engineering and Sciences at the University of Texas at Austin.
“We could imagine a digital twin of just about any system," Willcox tells Tech Briefs.
In a short Q&A with Tech Briefs below, Willcox explains more about why digital twins are so valuable across all industries – for both decision-making and design.
Tech Briefs: To set the stage a bit, how is a “digital twin” different from traditional simulation, and the way simulation has traditionally predicted the behavior of a part or system of parts?
Prof. Karen Willcox: I like the definition proposed in an AIAA/AIA 2020 Position Paper .
“A set of virtual information constructs that mimics the structure, context and behavior of an individual/unique physical asset, or a group of physical assets, is dynamically updated with data from its physical twin throughout its life cycle and informs decisions that realize value.”
This definition highlights three key aspects that distinguish a digital twin from a traditional simulation: First, a digital twin is asset-specific. It is not a generic computational model, like a simulation of a generic unmanned aerial vehicle (UAV), but rather it is an individualized, “patient-specific” model targeting simulation of one specific UAV. And so, if our fleet comprised many hundreds of UAVs, we would have hundreds of corresponding personalized digital twins.
Second, a digital twin is not a static computational model, but rather a living model that is dynamically updated to follow the physical twin through its lifecycle.
And third, key to the digital twin concept is the tight multi-way integration between data, models, and decisions.
While, of course, there are exceptions, these three aspects are not generally prevalent in the way that simulation has been used traditionally.
Tech Briefs: Why is a digital twin especially valuable in aerospace and in medical?
Prof. Karen Willcox: Despite the clear differences in the aerospace and medical fields, I am struck by the commonalities in their challenges. Applications in both fields typically have an underlying of complex nonlinear physical phenomena involving multiphysics and multiscale dynamics. These dynamics often lead to high sensitivities, making it difficult to make reliable predictions. In both fields, cyber-physical considerations are increasingly central to successful decision-making, as sensing, automation, and software/hardware interactions play larger roles. And whether it is an aerospace system or a human patient, there is a complex lifecycle involving multiple stages, multiple stakeholders, and an evolving asset state that undergoes degradation, damage, maintenance, and overhaul.
All these challenges point to the value of having an asset-specific/patient-specific model that combines predictive models and data to drive key decisions, especially when it enables a decision-maker to factor in multiple lifecycle considerations.
Tech Briefs: What other applications could especially benefit from a digital twin?
Prof. Karen Willcox: We could imagine a digital twin of just about any system. In my view, the most benefit is derived when variability or uncertainty play a large role in affecting outcomes. It could be variability from asset to asset (or patient to patient), or it could be variability of a single asset over its lifetime, or it could be uncertainty due to our inability to characterize or control elements of the system, including external factors. In all cases, there is clear value in having a dynamic data-driven personalized model to aid in decision-making, because then our decisions can be optimized to the situation at hand, rather than being made by assessing averages or probabilities across a population.
Tech Briefs: What are the limitations of digital twins?
Prof. Karen Willcox: It is true that we are in the age of big data, but when it comes to building a digital twin of something as complex as an aircraft or a person, the data are never enough. And I would venture to say that the data will never be enough. The advances in sensing technologies have been amazing, but our measurements are almost always sparse and indirect.
And it is not just a matter of more, or better, sensors. Sensors on board an engineering system add cost, weight, power consumption and thermal loads. Data from the natural world or from a human patient will always be temporally and spatially sparse – in addition to the expense, it is highly intrusive to collect data in these settings.
The limitation of our big data is one of the reasons that a digital twin must have an underpinning of predictive physics-based models. Yet putting it all together – complex physical models over multiple spatial and temporal scales, big data collected over a lifecycle spanning multiple stages and stakeholders, and high-consequence decisions fraught with uncertainty – remains a significant undertaking that goes beyond state-of-the-art in theory, scalable algorithms, and software implementations.
Tech Briefs: Do you envision these limitations being addressed in the future?
Prof. Karen Willcox: Yes, absolutely. But it will take continued investment in physics-based modeling and simulation capabilities, in addition to machine learning and data science.
What do you think? Share your questions and comments below.