Aarhus University Seal

This center is an umbrella for a collection of research projects that research/develop/make use of digital twins. A digital twin is a digital replica of physical assets, processes, people, places, systems or devices, created and maintained in order to answer questions about its physical counterpart, its physical twin in a near real-time setting. Coupled with new sensor technology, such replica can provide a new layer of engineering insight, which will be valuable in improving a product performance, and providing a seed for the next generation of the product. The conceptual idea of using a “twin” in an engineering setting dates back to NASAs Apollo program in the 1970s and it was taken up in a manufacturing setting in 2002.

This center was established because of a generous donation from the Poul Due Jensen Foundation with 12MDKK for the basic research project called DiT4CPS. The target for this center is to support the full cycle including basic research, applied research, innovation in particular together with SMEs and teaching for our students about digital twins. The research and investment in modernizing digital twins is predicted to have a significant importance in the future for example by the Gartner group.  In an increasingly digitalized industry and society, this can have an enormous impact. As a consequence, a number of companies have claimed to provide digital twin-based solutions, e.g., ANSYS and Siemens. With this center we hope to be able to provide an independent analysis of the underlying limitations and research challenges that typically is ignored in commercial marketing material.

Since digital twins are rather new there is still a lot of uncertainty about how it can be used both in different application fields/domains as well as under different circumstances for different kinds of systems. We primarily focus on how digital twins can be created from models developed during the engineering of a Cyber-Physical System (CPS), and can be used during its deployment. That way, we maximize the value of such models (typically discrete event models based on discrete mathematics and continuous-time models based on partial differential equations). We typically denote such models as first principle models since they are based on the necessary mathematics of the appropriate field.

The digital twin will often need to be calibrated and run with data coming from the sensors of the CPS. In the engineering of CPSs, the complexity is managed using heterogenous models (also called multi-models) of the different constituent systems as well as the envisaged environment that the CPS is intended to operate in. In case the engineering models include Computational Fluid Dynamics (CFD) or Finite Element Analysis (FEA) it will be necessary to automatically transform these into their Reduced Order Models (ROMs) approximations in order to be able to simulate these in a reasonable amount of time. For some CPSs there is a significant difference of potential environments that they can operate in and these have substantial influence on the actual behaviour of the physical twin. Another way to accomplish such approximations is to use machine learning on massive amount of data about a constituent system and then create what we call an inductive model of that.

Here are some of the questions that our research intends to tackle:

  • How can engineering models be reused in the production of Digital Twins?
  • How to make surrogate models (e.g., Neural Networks or Reduced Order Models) that are fast enough for real time monitoring?
  • How can the digital twin ensure that the physical twin is operating within an expected environment?