Session: 12-01-02 Wave theories II
Paper Number: 125048
125048 - Physics-Informed Neural Networks for Modeling Linear Waves
Numerical simulation of waves is of essential importance for different applications, including designing ships and coastal and offshore structures. While current methods offer varying degrees of wave simulation accuracy, there's a significant potential for developing more accurate and cost-effective techniques. One promising new method is the training of Physics-Informed Neural Networks (PINNs) for these simulations. PINNs can take advantage of all available data about waves of interest, which makes the model more reliable. In addition, few to no data points are needed for training these networks, which lessens the burden of data gathering. The current study discusses the simulation of linear waves (Airy waves) using PINNs, focusing solely on the governing equations. This study investigates how to implement various components of the governing equations and boundary conditions in the model and addresses the treatment of different unknowns, including constants and variables. The way the governing equation and boundary conditions were implemented significantly influenced the results and the convergence rate. The model successfully predicted the free surface profile and the gradients of potential distribution in various directions. In addition, the results proved that wave angular frequency or wave number can be set as a field constant unknown to be found by the PINN.
Presenting Author: Mohammad Sheikholeslami Chalmers University of Technology
Presenting Author Biography: Mohammad Sheikholeslami is currently pursuing his doctoral studies at Chalmers University of Technology in Sweden. His doctoral research primarily focuses on the utilization of physics-informed neural networks to address challenges associated with water wave phenomena and hydropower applications.
Authors:
Mohammad Sheikholeslami Chalmers University of TechnologySaeed Salehi Chalmers University of Technology
Wengang Mao Chalmers University of Techonology
Arash Eslamdoost Chalmers University of Technology
Håkan Nilsson Chalmers University of Technology
Physics-Informed Neural Networks for Modeling Linear Waves
Submission Type
Technical Paper Publication