Background
Wind loads acting on a WPU are inherently difficult to measure due to temporal and spatial variations in the inflow caused by atmospheric turbulence, vessel motion, and wing-vessel interaction effect. Measuring the deformed shape of a structure on the other hand is much simpler. Hence, the question is whether the deformed shape of the WPU can be used as a reliable and accurate proxy to monitor wind loads acting on a WPU.
Purpose and Objective
The underlying objective of this thesis is to investigate if a physics-informed neural network can be developed to accurately and reliably estimate the (unknown) loads acting on a structure of (known) deformed shape, with the case of the Oceanbird Wing 560 as reference example.
The thesis will focus on:
-Establish a suitable deep neural network structure, including any physical constraints in the loss function.
-Generate training data from full-scale 3D FEA models.
-Validate and improve the neural network to be able to predict loads acting on the wing sail.
The thesis work shall be documented in a written report. Depending on interest, the scope of thesis could be extended to include model-scale lab-testing to verify the methodology – to be discussed.
Skills
-Strong background in structural mechanics.
-Familiarity with FEA tools (e.g., ANSYS, Abaqus) and CAD software.
-Strong interest (or even experience) in machine learning, deep neural networks and python scripting.
-Analytical and structured with a curious mindset.
Contact
Andreas Fieber, Team Manager – Simulations and Analytics, andreas.fieber@alfalaval.com