What is (and what isn’t) a Digital Twin
And Why You Need One
Introduction
Digital Twins are an exciting new technology which will enable us to better manage complex assets like ships, cities, and factories.
Data Scientists and Data Engineers need an understanding of Digital Twins as they will require advanced models to be created and large volumes of real-time data will need to be managed.
This is the first article in a series. This article will introduce Digital Twins. In later articles we will look at specific implementation technologies and application domains.
This article introduces the concept of Digital Twin and discusses the potential benefits and applications.
When I use a word,’ Humpty Dumpty said, in rather a scornful tone, ‘it means just what I choose it to mean — neither more nor less.
Lewis Carroll, Through the Looking Glass [1]
While the term Digital Twin is relatively new (the term first appeared in 2002) it can be expected that there would be some ambiguity regarding the definition. As discussed by (Wright & Davidson, 2020), the risk of the misuse of the term is that the concept may be rejected as nothing more than hype. For this reason, we spend some time on developing a definition of a Digital Twin and also listing several implementations which are often incorrectly labelled as Digital Twins.
As will be discussed later in this article, Digital Twins find application in a wide variety of areas[2]. It should be noted that my background is the automation of continuous industrial processes[3] and as a result, the following will be written relating to Digital Twins to infrastructure and manufacturing.
Motivation
The primary motivation for developing Digital Twins is to enable the better management of complex assets and systems through their entire lifecycle.
Creating and maintaining anything but the simplest items requires that we develop a model of some kind. To construct a house, the designer will visualise the completed building and then translates those thoughts into a set of plans. These plans will then be used to fabricate the individual components which can then be assembled into the final structure.
It should be noted that the use of the model described above is basically unidirectional in that the data flows from concept to design, and finally to construction. Another point to consider is that the model is often discarded when the physical object has been created.
The Digital Twin concept extends this idea of creating a model of physical asset and comprises the following components,
- Digital replica of a physical thing
- Data driven model
- Simulation
Achieving the above requires,
- A model (including simulation)
- Sensors on a physical entity
- Visualisations (data lens)
History
NASA was the first to implement the concept of a Twin during the Apollo program (Ferguson 2020) which most notably contributed to bringing the stricken Apollo 13 spacecraft safely back to earth.
The term Digital Twin is attributed to Michael Grieves of the University of Michigan where the concept was incorporated into Product Lifecycle Management courses around 2003 (Jones et al., 2020; Semeraro et al., 2021)
Definition — What is a Digital Twin
The term Digital Twin is relatively young and there is much work being done to fully define what it actually means, in 2020 a comprehensive review of 93 paper was completed (Jones et al., 2020) and again in 2021 an even larger review of 150 paper was conducted (Semeraro et al., 2021).
Consensus requirements for a digital twin include,
- A physical object or system[4]
- A model of the object or system
- An evolving set of data relating to the object
- A means to dynamically update the model based on the data
Commonly, but not universally accepted requirements,
- A means to dynamically update the physical object or system based on the model
The IBM definition of a Digital Twin is,
A digital twin is a virtual representation of an object or system that spans its lifecycle, is updated from real-time data, and uses simulation, machine learning and reasoning to help decision-making.
(IBM, 2022)
In a 2021 paper (Semeraro et al., 2021), after examining 150 recent papers relating to Digital Twins[5], the authors presented that following definition,
A set of adaptive models that emulate the behaviour of a physical system in a virtual system getting real time data to update itself along its life cycle. The digital twin replicates the physical system to predict failures and opportunities for changing, to prescribe real time actions for optimizing and/or mitigating unexpected events observing and evaluating the operating profile system.
Below is a diagrammatic representation of the relationship between the digital and physical twin.
Foundations
Digital Twins are built with existing proven technologies.
Using the control of industrial processes as an example,
- Measurement of the process variables with instrumentation and visualisation with SCADA[6] systems is commonplace
- Process models and simulations have long been used during design, training and for process optimisation
Digital Twin Vs Model
The key difference between a Digital Twin and a Model is that a validated model describes a physical object at a point in time, while the digital twin has continually updated parameters which ensures that it describes the physical object at all times.
What is not a Digital Twin
Simulation Model
While a simulation model forms part of a Digital Twin. The key differentiator is that the simulation model is updated with measurements from the physical asset over time.
3D Model
Many organisations use the term Digital Twin synonymously with 3D model. At best a 3D model might be a component of a digital twin (Singh et al., 2021).
Common Data Environment
In the infrastructure and built environment space, many vendors and organisations are using Digital Twin to refer to one or more of,
- Building Information Management (BIM) as defined in ISO 19650
- Project Information Management (PIM)
- Common Data Environments (CDE)
While tools like BIM and CDE might be useful to organise and access data and even to provide a mechanism to facilitate access to models, in themselves they are not a Digital Twin.
Telemetry and Visualisation
Measuring parameters of a physical object or system and then visualising that data, does not alone constitute a Digital Twin (as the virtual twin, i.e., the model is missing).
Do I Already Have a Digital Twin?
In the Continuous Process Industries (CPI) it has been common practise (Rhinehart, 2021) to have process models which are periodically recalibrated (a.k.a twinning) to match the process. If this describes your operation, then you may well have a Digital Twin in place (without knowing it — refer to the previous comments regarding Digital Twins being largely based on existing well-established technologies).
Do I need a Digital TWIN?
Digital Twins are of most useful when the object or system is complex and changing over time (if not changing then can simply use a model, if not complex then can probably be understood using other methods).
Typical Applications
Digital twins are finding application in,
- Industry and Infrastructure (where it is also referred to as Industry 4.0)
- Utilities (SmartGrid)
- Product manufacturing (where it is referred to as PLM — Product Lifecycle Management)
- Cities (Smart Cities)
- Transportation (ITS — Intelligent Transportation Systems, Smart Roads)
- Gas turbines (Jet engines, Power Stations)
- Diesel engines (marine, power, locomotives, mine haul trucks)
- Healthcare
- Aerospace
Benefits
- Analysis of equipment performance in real time supporting Asset Performance Management (APM)
- Supports predictive maintenance
- Operational planning (what-if analysis through simulation)
- Operator training
- System optimisation
Implementing a Digital Twin
Three key aspects for a Digital Twin model identified by Wright and Davidson (Wright & Davidson, 2020),
- Sufficiently physics based (Phenomenological) to allow parameters to be updated based on new measurements from the physical twin
- Sufficient accuracy
- Sufficiently quick to execute
Finite Element Analysis (FEA) and computational fluid dynamics can be used. They are computationally intensive and techniques have been developed to produce lightweight models when is discussed by Grieves (Grieves, 2014).
Hybrid models can be created with utilize local high-fidelity models of key parts which are then combined with empirical (data-driven) models for the balance of the system.
Pure data-driven models are possible but not recommended (as results only valid over the range of training data) (Wright & Davidson, 2020). Note also that Empirical models (aka Data driven models) requires the physical object to exists before the model can be created.
Future Work
A number of areas have been identified in the literature as requiring future work before Digital Twins can be adopted more widely.
Validation
The application of Digital Twins to high value and safety critical systems requires that the systems can be validated (Wright & Davidson, 2020) as there is a need to be able to rely on the results produced.
MetaData
The requirements for consistent sensor data ontology has been identified (Wright & Davidson, 2020).
Lack of Standards
There is a lack of standards in model development and run-time environments (Singh et al., 2021).
Human in the Loop
Many implementations of Digital Twins lack a link from the virtual twin back to the physical twin. By inference there is a human in this loop and little work has been done to examine how this is best implemented (Jones et al., 2020).
Observations
Below are some general observations relating to the application on Digital Twins,
Benefits Realisation
It has been observed (Jones et al., 2020) that while many potential and perceived benefits have been reported in the literature there are few examples where these benefits have been quantified and validated. This is not uncommon in new and/or overhyped technologies.
Costs
The cost of developing high fidelity simulations of complex systems is exceptionally high and has typically been reserved for only a few well-resourced industries that have been able to justify the expense (petrochemical, aerospace etc.).
The high cost of implementation of high fidelity models required by Digital Twins was highlighted (Singh et al., 2021) and realistic cost benefit analysis needs to be conducted. (Also refer to the previous comments relating to the possible lack of tangible benefits described above).
LifeCycle Mismatch
There is a significant mismatch (Singh et al., 2021) in the life cycle of a typical engineered asset (building, ship, factory etc.) and the software systems that would be used to manage the virtual components of the Digital Twin.
Virtual to Physical Connection
Much of the literature does not include any description of a link from the virtual to physical model (Jones et al., 2020). The view of Jones is that without this there really seems to be little point to implementing a Digital Twin which is a view to which I strongly concur.
It should also be noted that (Grieves, 2014) in his original definition of a Digital Twin included the virtual to physical link.
Data Management and Model Creation
There are many existing facilities with long asset lives that have limited existing data and no models available.
Resiliance
The use of Digital Twins can lead to more highly integrated and optimised system.
The impacts on resilience need to be considered as greater connectedness and optimisation can lead to less resilient systems (Martin, 2020). Martin explores the idea that optimising a Complex Adaptive Systems will lead to a loss of resilience.
In some respects, concepts like Smart Cities might not actually work.
Conclusion
Digital Twins promise to improve the way complex assets and systems a built and operated. Data Scientists will be required to develop models for virtual twins while Data Engineers will be involved in the processing the real-time dataflows from physical to virtual and virtual to physical worlds.
To Explore Further
First look at new AWS IoT TwinMaker service.
Take a look at iTwins.js, an open-source data platform to federate and visualise infrastructure which can be used as platform on which a Digital Twin can be constructed.
Read Further
Thanks for reading, hope you enjoyed this article.
To explore further,
- AWS IoT TwinMaker
- Subscribe to email notifications
- Click on the ‘follow’ button at the top of the article
- For all things Industry 4.0, check out my Industrial Digital Transformation & Industry 4.0 publication
- Feel free to join my network on LinkedIn (remember to mention that you have read the article)
To support medium authors, consider a subscription.
Remember to click on the subscribe and follow button,
References
Ferguson , S. (2020). Apollo 13: The First Digital Twin. Siemens AG. Retrieved 2022–01–30 from https://blogs.sw.siemens.com/simcenter/apollo-13-the-first-digital-twin/
Grieves, M. (2014). Digital twin: manufacturing excellence through virtual factory replication (White paper, Issue.
IBM. (2022). What is a digital twin? IBM. Retrieved 2022–01–30 from https://www.ibm.com/topics/what-is-a-digital-twin#:~:text=A%20digital%20twin%20is%20a,Twin%20Exchange%20(01%3A41)
Jones, D., Snider, C., Nassehi, A., Yon, J., & Hicks, B. (2020). Characterising the Digital Twin: A systematic literature review. CIRP Journal of Manufacturing Science and Technology, 29, 36–52.
Martin, R. L. (2020). When More is Not Better. Harvard Business Review Press.
Rhinehart, R. (2021). Understanding the digital twin, part 1. Control Global. Retrieved 2022–01–30 from https://www.controlglobal.com/articles/2021/understanding-the-digital-twin-part-1/
Semeraro, C., Lezoche, M., Panetto, H., & Dassisti, M. (2021). Digital twin paradigm: A systematic literature review. Computers in Industry, 130, 103469.
Singh, M., Fuenmayor, E., Hinchy, E. P., Qiao, Y., Murray, N., & Devine, D. (2021). Digital twin: origin to future. Applied System Innovation, 4(2), 36.
Wright, L., & Davidson, S. (2020). How to tell the difference between a model and a digital twin. Advanced Modeling and Simulation in Engineering Sciences, 7(1), 1–13.
Footnotes
[1] Quotation sourced from, How to tell the difference between a model and a digital twin (Wright & Davidson, 2020)
[2] Product development, the built environment (buildings, roads, bridges etc.), aircraft, ships and so on
[3] Pulp and Paper manufacturing, Waster and Wastewater treatment, mineral processing, power generation etc.
[4] In many respects a Digital Twin during the design stage is non-sensical
[5] Worth a read, as the authors utilised machine learning and data analytics techniques to assist with the analyses of the large number of papers
[6] SCADA systems are described in my post Processing SCADA Alarm Data Offline with ELK
