Abstract
In recent years, data-driven operator approximation techniques have been explored as a means of solving physical problems described by ordinary and partial differential equations. In this paper, solutions to the linear 2D acoustic wave equation predicted by Fourier neural operator (FNO) networks are investigated in a square, free-field domain. The network's ability to generalise over variable excitation source positions in unseen locations is investigated. Furthermore, the network is tasked with learning progressively longer solutions in time to assess how the ratio of input to output data affects network prediction accuracy. Error between ground truth and predicted simulations is quantified and examined in an acoustics context.
| Original language | English |
|---|---|
| Title of host publication | Forum Acusticum 2023 - 10th Convention of the European Acoustics Association, EAA 2023 |
| Publisher | European Acoustics Association, EAA |
| ISBN (Electronic) | 9788888942674 |
| Publication status | Published - 15 Sept 2023 |
| Event | 10th Convention of the European Acoustics Association, EAA 2023 - Torino, Italy Duration: 11 Sept 2023 → 15 Sept 2023 |
Publication series
| Name | Proceedings of Forum Acusticum |
|---|---|
| ISSN (Print) | 2221-3767 |
Conference
| Conference | 10th Convention of the European Acoustics Association, EAA 2023 |
|---|---|
| Country/Territory | Italy |
| City | Torino |
| Period | 11/09/23 → 15/09/23 |
Bibliographical note
Publisher Copyright:© 2023 Michael Middleton This is an open-access article distributed under the terms of the Creative Commons Attribution 3.0 Unported License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Keywords
- acoustic simulation
- deep learning
- FDTD
- Fourier neural operator
- physics-informed neural network