Digital Twins: Bringing artificial intelligence to Engineering

digital twins bringing artificial intelligence to engineering

Digital Twins are increasing in usage but are often used in multiple contexts and in a simplified manner. Most references to the Digital Twin actually refer to a Digital shadow i.e. maintaining a digital copy of a physical object that is updated periodically. In a more complete sense, the Digital Twin concept relates to simulation and interaction of complex, multiple physical objects in a digital environment (typically for Engineering and Construction)

I am interested in the idea of Digital Twin because my teaching at the #universityofoxford applies more to AI in engineering (as opposed to say financial services).

Also, Digital Twins relate to the idea of Physics based modelling in Engineering. A wind tunnel is an example of Physics based model. Hence, one could think of a corresponding digital entity to the physical model which simulates the behavior of the model in a digital sense.

For this reason, digital twins are one of the best conceptual mechanisms for incorporating artificial intelligence into large-scale, dynamic engineering problems – especially considering existing ideas of physics-based modelling in engineering.

Digital twin technology is already used in various industrial sectors such as aerospace, infrastructure and automotive.

A paper I recently read talks about how Digital twins can be implemented through surrogate modeling.

The paper uses a discrete damped dynamic system to explore the concept of a digital twin.

An image of this idea is as below

digital twins bringing artificial intelligence to engineering 1

Image source

The paper uses Gaussian process (GP) emulator within the digital twin technology is explored. GP has the inherent capability of addressing noisy and sparse data.

GP is a probabilistic machine learning technique that attempts to infer a distribution and then use that distribution to predict unknown points.

GP has two distinct advantages over other surrogate models:

  • GP is a probabilistic surrogate model, it is resistant to overfitting.
  • GP can measure the uncertainty which can then be used in the decision-making process

Additional notes from the paper

  • GP not only model and also the example (spring) is a relatively simple one for explanation
  • As IoT proliferates, digital twins would get more complex based on increasing data being reflected in the virtual world from the physical world
  • Digital twins / surrogate modelling approach suits dynamically evolving systems
  • Typically, the digital twin starts from an ‘initial model’ which is often a physics-based model.
  • Over time, as more and more components can be modelled virtually, digital twins of larger (composite) objects would become the norm ex aircraft, automobiles etc

Paper link:

The role of surrogate models in the development of digital twins of…

Source Prolead brokers usa