A digital twin is a virtual model of a physical object or system used for simulations. Any object can be simulated this way – like wind farms, jet engines and even entire cities – to get a deeper understanding of its real-world counterpart. Digital twin technologies can replicate processes to collect data vital for future predictions. It’s like a bridge between the digital and physical world, representing the physical object throughout its entire lifecycle.
Hurdles
Although providing much more information than conventional models, digital twins are challenging to implement and use. “The substantial computational requirements for real-time calibration, prediction and control present a significant hurdle”, Mücke says. “When dealing with real-world systems we only have access to very limited information through sensors”, he explains. “To ‘fill in the gaps’ in information, we use simulations based on physical laws. To make sure that the simulations reflect the observed reality, they must be calibrated. This requires numerous simulations, each of which takes a long time to run. And even though the simulations are based on physical laws, they are not perfect. Therefore, it is also important to identify where the models are lacking, by comparing with the sensor observations without incorporating the observed noise.
Deep learning
In his thesis Mücke addresses these challenges by leveraging deep learning for efficient real-time data assimilation. His comprehensive and novel contributions to the field of digital twins enhance the practicality and performance of digital twins.
Information
‘Deep learning for real-time inverse problems and data assimilation with uncertainty quantification for digital twins’
Author: Nikolaj Mücke
Promotors: Prof.dr.ir. Kees Oosterlee (UU) and Prof.dr. Sander Bohté (CWI / UvA)
Defense date: 27 June
Photo: Ole.CNX/Shutterstock.com
