Physics-informed digital twins for industrial heating processes (twin4heat)

Industrial heating processes, which are currently widespread in foundation industries, are often described as transitioning to Industry 4.0. With rapid advances in the field of machine learning, data-driven approaches have become increasingly popular. Whilst the machine learning accurately models the physical process within the vicinity of the experimental data used for training, it fails to generalise away from these data due to its lack of understanding of the underlying physical process. Amongst the technology making this possible is the ability of “digital twins” to fully simulate the heat transfer processes which represents the physics of the actual processes, i.e., physics-informed digital twins.

In this project, by incorporating existing physical principles of industrial heating processes into machine learning algorithms we will create more powerful models that learn from available data and build upon our existing scientific knowledge. It will be able to predict the state far away from the experimental data points and thus demonstrate superior performance with respect to the naiver approaches. If the project is successful, it will develop new methods to realise the benefit of the digital twin in next generation and intelligent industrial heating processes by providing engineering designers with leading indicators, helping them to trace and handle interdependencies and uncertainties.

Dr. Yukun Hu

University College London

Yukun.hu@ucl.ac.uk