Rather than attempting to build inductive models using machine learning, this research attempts to leverage the engineering multi-models constructed during the development process of the system to build a digital twin.

The aim is to increase our understanding of

• how to reuse engineering multi-models to build a digital twin.
• how can DSE with alternative faulty scenarios be used to provide a why-that capability for assisting with determining likely faults based on data from the original twin.
• which co-simulation approaches are suited for calibration and support real-time what-if analyses.
• which kinds of models can be used for that purpose.

We also want to be able to characterise when and how to employ a digital twin, and which kind of faults that can be diagnosed from the engineering multi-models most commonly used in practice.

The DiT4CPS project has seven overall tasks:

• Getting data: This task is dedicated to getting data about the physical CPSs from the physical sensors and actuators corresponding to the different constituent models from the multi-model of the CPS.
• Co-simulation: This task targets enhancing the existing Maestro co-simulation capability takes the time-series of data as input to the FMUs representing the multi-models or the digital version of the CPS.
• Monitoring: This task is responsible for the monitoring of the difference between how the physical CPS performs compared to what the co-simulation of the corresponding multi-model predicts will be added as a new service.
• Semantic foundation: This task is responsible for the semantic foundation of the INTO-CPS co-simulation expanded to cope with uncertainties.
• Calibration/adaptation: This task will produce a service enabling calibration/adaptation of a CPS, in particular in relation to its environment, in particular for applications such as the mobile robots.
• What-if analysis: This task is responsible for providing human operators with a capability to experiment with potential interactions in the control of the real CPS such that what-if analysis can be achieved.
• Industrial cases: In order to test the limits of the research results we have a number of industrial sized cases that we can fully control and work on from an academic perspective. These will be used as case studies for different scientific publications.