14:15 – 14:45
During transportation of Liquefied Natural Gas (LNG) in tanker ships, violent liquid sloshing can appear. We propose a digital twin to predict in real-time the operational envelope of a tanker. A critical part of the digital twin is the fluid dynamics model, for which we propose a novel approach: a multi-fidelity neural network. This approach uses a combination of a low- and high-fidelity model to significantly decrease the computational cost. The key element is the presence of the simplified physics in the low-fidelity solver, which results in a neural network that is capable of enhancing low-fidelity solutions also outside the training set.
Benjamin Sanderse is a tenure-track researcher in the Scientific Computing group at Centrum Wiskunde & Informatica (CWI) Amsterdam. He has a MSc in aerospace engineering and a PhD in numerical mathematics, in which he developed new fluid dynamic solvers for wind turbine applications. His research expertise is in numerical analysis, uncertainty quantification and computational fluid dynamics. An ongoing theme in his work is the development of reduced order fluid dynamic models that preserve certain properties (structure) of the full order models. The resulting reduced models are a crucial ingredient for real-time simulation in digital twin applications.