Data-driven digital twin for global load and motion prediction of floating structures
This thesis aims at the implementation and validation of a data-driven digital twin — a faithful representation of a dynamical system — for global load and motion prediction based on surface elevation at the hull with machine learning (ML). Training data is generated using an existing strip theory code that calculates the vertical bending moment based on nonlinear wave simulations. A machine learning model is then trained to map the surface elevation to the vertical bending moment, eventually replacing the strip theory code. Once trained, a machine learning model becomes a static input-output mapping. Consequently, one major benefit of using ML over numerical methods is the almost instant execution time and thus real-time capability.
The scope of this work covers the following tasks:
- Data generation using existing strip theory code and surface elevation data,
- Literature review of state of the art, choice of suitable ML method,
- Application and evaluation of chosen method.
- Demonstrated programming experience in Python and TensorFlow,
- Knowledge of neural networks,
- Curiosity, excellent skills in independent work and communication.
Please find the original tender at https://cgi.tu-harburg.de/~dynwww/cgi-bin/teaching/project?tx_news_pi1%5Baction%5D=detail&tx_news_pi1%5Bcontroller%5D=News&tx_news_pi1%5Bnews%5D=34&cHash=f5f0c1b235d082ddb3635c0553621bed