Artificial Intelligence No 58: Digital twins v.s. traditional IoT applications
A system of systems - image source IBM

Artificial Intelligence No 58: Digital twins v.s. traditional IoT applications

In the recently concluded Digital Twins course at the University of Oxford – the question of what exactly is a digital twin is still high on the agenda.

There are a few reasons for this

  • Companies are defining digital twins in their own image (strengths)
  • Companies are defining Digital twins in their own technical limitations (ex: currently not all twins perform complex simulations)
  • Finally, some functions of digital twins can also be done using traditional IoT applications. For example, remaining useful life as an application can be a twin but also a conventional application

So, its important to define twins in terms of functionality that only digital twins can implement

The term ‘Digital Twin’ has several definitions, but for the proposes of this post, we shall consider a Digital Twin is defined as a dynamic representation of a physical system using interconnected data (1). This definition makes the Digital twin a complex system in itself. A complex system is defined as “a large network of components, many-to-many communication channels, and sophisticated information processing that makes prediction of system states difficult” (2).

So, how do you distinguish Digital twins v.s. traditional IoT applications?

If we consider Digital twins from a systems perspective, then things become much more clearer. Firstly, twins have specific characteristics that can be only implemented when viewed as a system (or indeed a system of systems)

  • Systemic view – scenario Fast forwarding
  • Systemic view - What if – fast forwarding
  • Systemic view – Failure – fast forwarding
  • Supply chain twin
  • Cost optimization twin
  • Smart cities – urban planning, pollution, power
  • Commissioning new equipment
  • Decommissioning equipment
  • Virtual commissioning
  • Twins of things that can be modelled
  • Twins of things / processes that cannot be modelled physically
  • Additive manufacturing
  • Twins to AR/VR
  • Causal twins
  • Graph based twins
  • Low code based twins
  • Human in the loop modelling for twins

?All the above in most cases take a systems perspective

Which algorithms can be used to implement digital twins? That’s one more differentiator

  • System dynamics?
  • Discrete event modelling?
  • Agent based modelling?

?On the other hand, most IoT applications can be considered as a simple twin ie from a non systemic perspective

  • Improved customer?service
  • Maintenance?of machines?
  • Automotive Insurance:
  • Driver experience improvement
  • Predictive analytics?????
  • Industrial robotics?????
  • Inventory management??
  • Predictive maintenance:??Remaining useful life
  • Process Automation
  • Robotics including autonomous robotics
  • Fleet management
  • Adas advanced driver management systems
  • Industrial robotics:?
  • Onboard diagnostics (automotive)

That's the best way I can think of to differentiate digital twins from traditional IoT applications.

By this perspective, it implies that

a) ?Digital twins need a systems thinking perspective

b) IoT applications can be seen as a simplified digital twin (not as a system) - so the twin can be seen as a superset of traditional IoT Edge based applications.

c) Digital twins need a capacity for simulation (updated)

comments welcome if you can explain this better!

1.?????https://hal.archives-ouvertes.fr/hal-03262607

2.?????https://link.springer.com/chapter/10.1007/978-3-319-38756-7_4

3.?????https://link.springer.com/chapter/10.1007/978-3-642-35758-9_20

?

Image source IBM https://www.solvingforpattern.org/2012/06/30/ibm-system/

Geoffrey Payne

Data Scientist with masters degree in AI, leveraging 10 years as software developer | Python | Pytorch | NLP - Natural Language Processing | Computer Vision | SQL | AWS | C#

2 年

Digital Twins is still a new concept for me, but I recall Paul Clarke pointing out that the difference between a Digital Shadow and a Digital Twin is that a Digital Shadow creates a simulator model built from data input, whilst a Digital Twin has a 2 way communication from the data source. For example a monitor recording road usage can general a Digital Shadow simulator of that road usage. However if based on the incoming data it can calculate whether a traffic light should be changed and send an instruction to that traffic light to do so, then the Digital Shadow becomes a Digital Twin. It seemed to me that this was a key defining point.

Joseph A di Paolantonio

SensAE are better than IoT projects; mature with connection, communication, contextualization, collaboration, causation, conceptualization and cognition into Sensor Analytics Ecosystems

2 年

I agree that digital twins need a system perspective. I don’t see IoT applications as simplified digital twins, as an IoT application need not perform any simulation, but I can not imagine a digital twin that does not have multiple simulations, even if each simulation is simplistic.

Dr. PG Madhavan

Digital Twin maker: Causality & Data Science --> TwinARC - the "INSIGHT Digital Twin"!

2 年

Ajit Jaokar You say, "IoT applications can be seen as a simplified digital twin". This can be taken to mean Digital Twin is a SUPERset. Which the definition that I use...

要查看或添加评论,请登录

Ajit Jaokar的更多文章

社区洞察

其他会员也浏览了