Modeling of 3D Blood Flows with Physics-Informed Neural Networks: Comparison of Network Architectures
Machine learning-based modeling of physical systems has attracted significant interest in recent years. Based solely on the underlying physical equations and initial and boundary conditions, these new approaches allow to approximate, for example, the complex flow of blood in the case of fluid dynami...
Main Authors: | , , , , |
---|---|
Format: | Article |
Language: | English |
Published: |
MDPI AG
2023-01-01
|
Series: | Fluids |
Subjects: | |
Online Access: | https://www.mdpi.com/2311-5521/8/2/46 |
_version_ | 1797621046168256512 |
---|---|
author | Philipp Moser Wolfgang Fenz Stefan Thumfart Isabell Ganitzer Michael Giretzlehner |
author_facet | Philipp Moser Wolfgang Fenz Stefan Thumfart Isabell Ganitzer Michael Giretzlehner |
author_sort | Philipp Moser |
collection | DOAJ |
description | Machine learning-based modeling of physical systems has attracted significant interest in recent years. Based solely on the underlying physical equations and initial and boundary conditions, these new approaches allow to approximate, for example, the complex flow of blood in the case of fluid dynamics. Physics-informed neural networks offer certain advantages compared to conventional computational fluid dynamics methods as they avoid the need for discretized meshes and allow to readily solve inverse problems and integrate additional data into the algorithms. Today, the majority of published reports on learning-based flow modeling relies on fully-connected neural networks. However, many different network architectures are introduced into deep learning each year, each with specific benefits for certain applications. In this paper, we present the first comprehensive comparison of various state-of-the-art networks and evaluate their performance in terms of computational cost and accuracy relative to numerical references. We found that while fully-connected networks offer an attractive balance between training time and accuracy, more elaborate architectures (e.g., Deep Galerkin Method) generated superior results. Moreover, we observed high accuracy in simple cylindrical geometries, but slightly poorer estimates in complex aneurysms. This paper provides quantitative guidance for practitioners interested in complex flow modeling using physics-based deep learning. |
first_indexed | 2024-03-11T08:51:06Z |
format | Article |
id | doaj.art-af557815319c41c7bf32fdb1b62e5678 |
institution | Directory Open Access Journal |
issn | 2311-5521 |
language | English |
last_indexed | 2024-03-11T08:51:06Z |
publishDate | 2023-01-01 |
publisher | MDPI AG |
record_format | Article |
series | Fluids |
spelling | doaj.art-af557815319c41c7bf32fdb1b62e56782023-11-16T20:28:39ZengMDPI AGFluids2311-55212023-01-01824610.3390/fluids8020046Modeling of 3D Blood Flows with Physics-Informed Neural Networks: Comparison of Network ArchitecturesPhilipp Moser0Wolfgang Fenz1Stefan Thumfart2Isabell Ganitzer3Michael Giretzlehner4Research Unit Medical Informatics, RISC Software GmbH, 4232 Hagenberg, AustriaResearch Unit Medical Informatics, RISC Software GmbH, 4232 Hagenberg, AustriaResearch Unit Medical Informatics, RISC Software GmbH, 4232 Hagenberg, AustriaResearch Unit Medical Informatics, RISC Software GmbH, 4232 Hagenberg, AustriaResearch Unit Medical Informatics, RISC Software GmbH, 4232 Hagenberg, AustriaMachine learning-based modeling of physical systems has attracted significant interest in recent years. Based solely on the underlying physical equations and initial and boundary conditions, these new approaches allow to approximate, for example, the complex flow of blood in the case of fluid dynamics. Physics-informed neural networks offer certain advantages compared to conventional computational fluid dynamics methods as they avoid the need for discretized meshes and allow to readily solve inverse problems and integrate additional data into the algorithms. Today, the majority of published reports on learning-based flow modeling relies on fully-connected neural networks. However, many different network architectures are introduced into deep learning each year, each with specific benefits for certain applications. In this paper, we present the first comprehensive comparison of various state-of-the-art networks and evaluate their performance in terms of computational cost and accuracy relative to numerical references. We found that while fully-connected networks offer an attractive balance between training time and accuracy, more elaborate architectures (e.g., Deep Galerkin Method) generated superior results. Moreover, we observed high accuracy in simple cylindrical geometries, but slightly poorer estimates in complex aneurysms. This paper provides quantitative guidance for practitioners interested in complex flow modeling using physics-based deep learning.https://www.mdpi.com/2311-5521/8/2/46physics-informed neural networksfluidsblood flowNavier-Stokesdeep learningpinn |
spellingShingle | Philipp Moser Wolfgang Fenz Stefan Thumfart Isabell Ganitzer Michael Giretzlehner Modeling of 3D Blood Flows with Physics-Informed Neural Networks: Comparison of Network Architectures Fluids physics-informed neural networks fluids blood flow Navier-Stokes deep learning pinn |
title | Modeling of 3D Blood Flows with Physics-Informed Neural Networks: Comparison of Network Architectures |
title_full | Modeling of 3D Blood Flows with Physics-Informed Neural Networks: Comparison of Network Architectures |
title_fullStr | Modeling of 3D Blood Flows with Physics-Informed Neural Networks: Comparison of Network Architectures |
title_full_unstemmed | Modeling of 3D Blood Flows with Physics-Informed Neural Networks: Comparison of Network Architectures |
title_short | Modeling of 3D Blood Flows with Physics-Informed Neural Networks: Comparison of Network Architectures |
title_sort | modeling of 3d blood flows with physics informed neural networks comparison of network architectures |
topic | physics-informed neural networks fluids blood flow Navier-Stokes deep learning pinn |
url | https://www.mdpi.com/2311-5521/8/2/46 |
work_keys_str_mv | AT philippmoser modelingof3dbloodflowswithphysicsinformedneuralnetworkscomparisonofnetworkarchitectures AT wolfgangfenz modelingof3dbloodflowswithphysicsinformedneuralnetworkscomparisonofnetworkarchitectures AT stefanthumfart modelingof3dbloodflowswithphysicsinformedneuralnetworkscomparisonofnetworkarchitectures AT isabellganitzer modelingof3dbloodflowswithphysicsinformedneuralnetworkscomparisonofnetworkarchitectures AT michaelgiretzlehner modelingof3dbloodflowswithphysicsinformedneuralnetworkscomparisonofnetworkarchitectures |