Skip to main content
OSTI.GOVU.S. Department of Energy Office of Scientific and Technical Information
U.S. Department of Energy
Office of Scientific and Technical Information
 

Multiscale graph neural network autoencoders for interpretable scientific machine learning

Journal Article · · Journal of Computational Physics
 [1];  [2];  [2];  [3]
  1. Argonne National Laboratory (ANL), Argonne, IL (United States). Argonne Leadership Computing Facility (ALCF)
  2. Carnegie Mellon Univ., Pittsburgh, PA (United States)
  3. Pennsylvania State Univ., University Park, PA (United States)
+ Show Author Affiliations
The goal of this work is to address two limitations in autoencoder-based models: latent space interpretability and compatibility with unstructured meshes. This is accomplished here with the development of a novel graph neural network (GNN) autoencoding architecture with demonstrations on complex fluid flow applications. To address the first goal of interpretability, the GNN autoencoder achieves reduction in the number nodes in the encoding stage through an adaptive graph reduction procedure. Further, this reduction procedure essentially amounts to flowfieldconditioned node sampling and sensor identification, and produces interpretable latent graph representations tailored to the flowfield reconstruction task in the form of so-called masked fields. These masked fields allow the user to (a) visualize where in physical space a given latent graph is active, and (b) interpret the time-evolution of the latent graph connectivity in accordance with the time-evolution of unsteady flow features (e.g. recirculation zones, shear layers) in the domain. To address the goal of unstructured mesh compatibility, the autoencoding architecture utilizes a series of multi-scale message passing (MMP) layers, each of which models information exchange among node neighborhoods at various lengthscales. The MMP layer, which augments standard single-scale message passing with learnable coarsening operations, allows the decoder to more efficiently reconstruct the flowfield from the identified regions in the masked fields. Analysis of latent graphs produced by the autoencoder for various model settings are conducted using unstructured snapshot data sourced from large-eddy simulations in a backward-facing step (BFS) flow configuration with an OpenFOAM-based flow solver at high Reynolds numbers.
Research Organization:
Argonne National Laboratory (ANL), Argonne, IL (United States). Argonne Leadership Computing Facility (ALCF)
Sponsoring Organization:
USDOE Office of Science (SC), Advanced Scientific Computing Research (ASCR); USDOE Office of Science (SC), Basic Energy Sciences (BES). Scientific User Facilities (SUF)
Grant/Contract Number:
AC02-06CH11357
OSTI ID:
2571430
Alternate ID(s):
OSTI ID: 2202891
Journal Information:
Journal of Computational Physics, Journal Name: Journal of Computational Physics Vol. 495; ISSN 0021-9991
Publisher:
ElsevierCopyright Statement
Country of Publication:
United States
Language:
English

References (39)

Review of computational fluid dynamics for wind turbine wake aerodynamics journal February 2011
Experimental and Numerical Investigation of 2-d Backward-Facing step flow journal August 1998
Particle image velocimetry measurements of a backward-facing step flow journal December 2002
Solution of the implicitly discretised fluid flow equations by operator-splitting journal January 1986
Fast and Flexible Protein Design Using Deep Graph Neural Networks journal October 2020
Multi-level convolutional autoencoder networks for parametric prediction of spatio-temporal dynamics journal December 2020
A comparison of neural network architectures for data-driven reduced-order modeling journal April 2022
The role of strain rate, local extinction, and hydrodynamic instability on transition between attached and lifted swirl flames journal January 2019
Ten iterative steps for model development and evaluation applied to Computational Fluid Dynamics for Environmental Fluid Mechanics journal July 2012
Model reduction of dynamical systems on nonlinear manifolds using deep convolutional autoencoders journal November 2019
Adversarial sampling of unknown and high-dimensional conditional distributions journal February 2022
Neural-network learning of SPOD latent dynamics journal November 2022
Time-series learning of latent-space dynamics for reduced-order model closure journal April 2020
Emerging trends in numerical simulations of combustion systems journal January 2019
Extracting information overlap in simultaneous OH-PLIF and PIV fields with neural networks journal January 2021
Dynamic mode decomposition of numerical and experimental data journal July 2010
Experimental and theoretical investigation of backward-facing step flow journal February 1983
Cluster-based reduced-order modelling of a mixing layer journal August 2014
Super-resolution reconstruction of turbulent flows with machine learning journal May 2019
Nonlinear mode decomposition with convolutional neural networks for fluid dynamics journal November 2019
Self-sustained oscillations and vortex shedding in backward-facing step flows: Simulation and linear instability analysis journal July 2004
Scalar imaging velocimetry measurements of the velocity gradient tensor field in turbulent flows. I. Assessment of errors journal July 1996
Reduced-order modeling of advection-dominated systems with recurrent neural networks and convolutional autoencoders journal March 2021
Multi-scale rotation-equivariant graph neural networks for unsteady Eulerian fluid dynamics journal August 2022
Grid adaptive reduced-order model of fluid flow based on graph convolutional neural network journal August 2022
Machine learning–accelerated computational fluid dynamics journal May 2021
Using Machine Learning to Construct Velocity Fields from OH-PLIF Images journal October 2019
AMGNET: multi-scale graph neural networks for flow field prediction journal October 2022
Experimental data-based reduced-order model for analysis and prediction of flame transition in gas turbine combustors journal April 2019
Data-driven reduction and decomposition with time-axis clustering journal June 2023
Supervised neural networks for the classification of structures journal May 1997
XAI—Explainable artificial intelligence journal December 2019
A review of a new generation of wildfire–atmosphere modeling journal June 2019
Turbulence Modeling in the Age of Data journal January 2019
The Proper Orthogonal Decomposition in the Analysis of Turbulent Flows journal January 1993
General Circulation Experiments with the Primitive Equations: i. the Basic Experiment* journal March 1963
Modal Analysis of Fluid Flows: Applications and Outlook journal March 2020
Multigrid solution of the Navier-Stokes equations on triangular meshes journal August 1990
Disentangling Generative Factors of Physical Fields Using Variational Autoencoders journal June 2022

Figures / Tables (24)

Figure 1(p. 5)figure Figure 1
Figure 2(p. 6)figure Figure 2
Table 1(p. 6)table Table 1
Figure 3(p. 7)figure Figure 3
Figure 4(p. 7)figure Figure 4
Figure 5(p. 8)figure Figure 5
Figure 6(p. 8)figure Figure 6
Figure 7(p. 9)figure Figure 7
Figure 8(p. 12)figure Figure 8
Figure 9(p. 13)figure Figure 9
Figure 10(p. 14)figure Figure 10
Figure 11(p. 14)figure Figure 11
Figure 12(p. 17)figure Figure 12
Figure 13(p. 18)figure Figure 13
Figure 14(p. 20)figure Figure 14
Figure 15(p. 21)figure Figure 15
Figure 16(p. 23)figure Figure 16
Figure 17(p. 23)figure Figure 17
Figure A.18(p. 25)figure Figure A.18
Figure A.19(p. 26)figure Figure A.19
Figure A.20(p. 27)figure Figure A.20
Figure B.21(p. 27)figure Figure B.21
Figure B.22(p. 28)figure Figure B.22
Figure C.23(p. 29)figure Figure C.23

Similar Records

Mesh-based super-resolution of fluid flows with multiscale graph neural networks
Journal Article · 2025 · Computer Methods in Applied Mechanics and Engineering · OSTI ID:2587579

Related Subjects

97 MATHEMATICS AND COMPUTING
Computational fluid dynamics
Graph neural networks
Interpretable machine learning
Scientific machine learning