Projection-based model reduction: Formulations for physics-based machine learning

Renee Swischuk, Laura Mainini, Benjamin Peherstorfer, Karen Willcox

Research output: Contribution to journalArticle

Abstract

This paper considers the creation of parametric surrogate models for applications in science and engineering where the goal is to predict high-dimensional output quantities of interest, such as pressure, temperature and strain fields. The proposed methodology develops a low-dimensional parametrization of these quantities of interest using the proper orthogonal decomposition (POD), and combines this parametrization with machine learning methods to learn the map between the input parameters and the POD expansion coefficients. The use of particular solutions in the POD expansion provides a way to embed physical constraints, such as boundary conditions and other features of the solution that must be preserved. The relative costs and effectiveness of four different machine learning techniques—neural networks, multivariate polynomial regression, k-nearest-neighbors and decision trees—are explored through two engineering examples. The first example considers prediction of the pressure field around an airfoil, while the second considers prediction of the strain field over a damaged composite panel. The case studies demonstrate the importance of embedding physical constraints within learned models, and also highlight the important point that the amount of model training data available in an engineering setting is often much less than it is in other machine learning applications, making it essential to incorporate knowledge from physical models.

Original languageEnglish (US)
JournalComputers and Fluids
DOIs
StateAccepted/In press - Jan 1 2018

Fingerprint

Learning systems
Physics
Decomposition
Airfoils
Polynomials
Boundary conditions
Composite materials
Costs
Temperature

Keywords

  • Data-driven reduced models
  • Model reduction
  • Physics-based machine learning
  • Proper orthogonal decomposition
  • Surrogate models

ASJC Scopus subject areas

  • Computer Science(all)
  • Engineering(all)

Cite this

Projection-based model reduction : Formulations for physics-based machine learning. / Swischuk, Renee; Mainini, Laura; Peherstorfer, Benjamin; Willcox, Karen.

In: Computers and Fluids, 01.01.2018.

Research output: Contribution to journalArticle

Swischuk, Renee ; Mainini, Laura ; Peherstorfer, Benjamin ; Willcox, Karen. / Projection-based model reduction : Formulations for physics-based machine learning. In: Computers and Fluids. 2018.
@article{528e6480914547f49d06c357045ac5c9,
title = "Projection-based model reduction: Formulations for physics-based machine learning",
abstract = "This paper considers the creation of parametric surrogate models for applications in science and engineering where the goal is to predict high-dimensional output quantities of interest, such as pressure, temperature and strain fields. The proposed methodology develops a low-dimensional parametrization of these quantities of interest using the proper orthogonal decomposition (POD), and combines this parametrization with machine learning methods to learn the map between the input parameters and the POD expansion coefficients. The use of particular solutions in the POD expansion provides a way to embed physical constraints, such as boundary conditions and other features of the solution that must be preserved. The relative costs and effectiveness of four different machine learning techniques—neural networks, multivariate polynomial regression, k-nearest-neighbors and decision trees—are explored through two engineering examples. The first example considers prediction of the pressure field around an airfoil, while the second considers prediction of the strain field over a damaged composite panel. The case studies demonstrate the importance of embedding physical constraints within learned models, and also highlight the important point that the amount of model training data available in an engineering setting is often much less than it is in other machine learning applications, making it essential to incorporate knowledge from physical models.",
keywords = "Data-driven reduced models, Model reduction, Physics-based machine learning, Proper orthogonal decomposition, Surrogate models",
author = "Renee Swischuk and Laura Mainini and Benjamin Peherstorfer and Karen Willcox",
year = "2018",
month = "1",
day = "1",
doi = "10.1016/j.compfluid.2018.07.021",
language = "English (US)",
journal = "Computers and Fluids",
issn = "0045-7930",
publisher = "Elsevier Limited",

}

TY - JOUR

T1 - Projection-based model reduction

T2 - Formulations for physics-based machine learning

AU - Swischuk, Renee

AU - Mainini, Laura

AU - Peherstorfer, Benjamin

AU - Willcox, Karen

PY - 2018/1/1

Y1 - 2018/1/1

N2 - This paper considers the creation of parametric surrogate models for applications in science and engineering where the goal is to predict high-dimensional output quantities of interest, such as pressure, temperature and strain fields. The proposed methodology develops a low-dimensional parametrization of these quantities of interest using the proper orthogonal decomposition (POD), and combines this parametrization with machine learning methods to learn the map between the input parameters and the POD expansion coefficients. The use of particular solutions in the POD expansion provides a way to embed physical constraints, such as boundary conditions and other features of the solution that must be preserved. The relative costs and effectiveness of four different machine learning techniques—neural networks, multivariate polynomial regression, k-nearest-neighbors and decision trees—are explored through two engineering examples. The first example considers prediction of the pressure field around an airfoil, while the second considers prediction of the strain field over a damaged composite panel. The case studies demonstrate the importance of embedding physical constraints within learned models, and also highlight the important point that the amount of model training data available in an engineering setting is often much less than it is in other machine learning applications, making it essential to incorporate knowledge from physical models.

AB - This paper considers the creation of parametric surrogate models for applications in science and engineering where the goal is to predict high-dimensional output quantities of interest, such as pressure, temperature and strain fields. The proposed methodology develops a low-dimensional parametrization of these quantities of interest using the proper orthogonal decomposition (POD), and combines this parametrization with machine learning methods to learn the map between the input parameters and the POD expansion coefficients. The use of particular solutions in the POD expansion provides a way to embed physical constraints, such as boundary conditions and other features of the solution that must be preserved. The relative costs and effectiveness of four different machine learning techniques—neural networks, multivariate polynomial regression, k-nearest-neighbors and decision trees—are explored through two engineering examples. The first example considers prediction of the pressure field around an airfoil, while the second considers prediction of the strain field over a damaged composite panel. The case studies demonstrate the importance of embedding physical constraints within learned models, and also highlight the important point that the amount of model training data available in an engineering setting is often much less than it is in other machine learning applications, making it essential to incorporate knowledge from physical models.

KW - Data-driven reduced models

KW - Model reduction

KW - Physics-based machine learning

KW - Proper orthogonal decomposition

KW - Surrogate models

UR - http://www.scopus.com/inward/record.url?scp=85051393788&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85051393788&partnerID=8YFLogxK

U2 - 10.1016/j.compfluid.2018.07.021

DO - 10.1016/j.compfluid.2018.07.021

M3 - Article

JO - Computers and Fluids

JF - Computers and Fluids

SN - 0045-7930

ER -