Contribution
Assessing the capabilities of Physics-Informed-Networks for solving the Helmholtz equation using the example of a car horn
* Presenting author
Abstract:
Solving the wave equation is of importance in many different fields of physics (acoustics, electromagnetism, seismology, …) when it comes to modelling wave phenomena. Current approaches to finding solutions to this equation tend to be computationally expensive using traditional solving methods like finite differences or finite elements (FD & FE). In this study, we make use of the recently introduced PINN (Physics Informed Neural Network) framework to solve a specific form of the wave equation: the Helmholtz equation. Using a COMSOL model for simulating the acoustic pressure field of a car horn, we will train a PINN to, in a first place, evaluate the replication capabilities on a given COMSOL simulation, before assessing the predictive capabilities of such a neural network when using different simulation parameters as inputs, such as excitation frequency and shape of the geometry, as well as different types of loss functions for the PINN.