Local reduced order modeling and iterative sampling for parametric analyses of aero-icing problems

Zhao Zhan, Wagdi G. Habashi, Marco Fossati

Research output: Contribution to journalArticle

11 Citations (Scopus)

Abstract

A framework of local reduced-order modeling using machine learning algorithms is presented together with an approach to optimally select the snapshots for strongly nonlinear problems. By using an unsupervised learning algorithm, solutions are grouped into clusters of similar features. The input parameter space is divided into subregions by decision boundaries based on a supervised learning algorithm. Local reduced-order bases are extracted on each cluster, for which the solutions are represented as a linear combination of the basis vectors from their corresponding subregion. The proposed methodology is employed to conduct a comprehensive, exploration of the in-flight icing certification envelopes.
LanguageEnglish
Pages2174 - 2185
Number of pages10
JournalAIAA Journal
Volume53
Issue number8
Early online date24 Apr 2015
DOIs
Publication statusPublished - 31 Aug 2015

Fingerprint

Learning algorithms
Sampling
Unsupervised learning
Supervised learning
Learning systems

Keywords

  • machine learning algorithms
  • in-flight icing
  • vectors

Cite this

@article{761bbe70451c4c3bb0f56a08292fb55c,
title = "Local reduced order modeling and iterative sampling for parametric analyses of aero-icing problems",
abstract = "A framework of local reduced-order modeling using machine learning algorithms is presented together with an approach to optimally select the snapshots for strongly nonlinear problems. By using an unsupervised learning algorithm, solutions are grouped into clusters of similar features. The input parameter space is divided into subregions by decision boundaries based on a supervised learning algorithm. Local reduced-order bases are extracted on each cluster, for which the solutions are represented as a linear combination of the basis vectors from their corresponding subregion. The proposed methodology is employed to conduct a comprehensive, exploration of the in-flight icing certification envelopes.",
keywords = "machine learning algorithms, in-flight icing, vectors",
author = "Zhao Zhan and Habashi, {Wagdi G.} and Marco Fossati",
year = "2015",
month = "8",
day = "31",
doi = "10.2514/1.J053654",
language = "English",
volume = "53",
pages = "2174 -- 2185",
journal = "AIAA Journal",
issn = "0001-1452",
number = "8",

}

Local reduced order modeling and iterative sampling for parametric analyses of aero-icing problems. / Zhan, Zhao; Habashi, Wagdi G.; Fossati, Marco.

In: AIAA Journal, Vol. 53, No. 8, 31.08.2015, p. 2174 - 2185.

Research output: Contribution to journalArticle

TY - JOUR

T1 - Local reduced order modeling and iterative sampling for parametric analyses of aero-icing problems

AU - Zhan, Zhao

AU - Habashi, Wagdi G.

AU - Fossati, Marco

PY - 2015/8/31

Y1 - 2015/8/31

N2 - A framework of local reduced-order modeling using machine learning algorithms is presented together with an approach to optimally select the snapshots for strongly nonlinear problems. By using an unsupervised learning algorithm, solutions are grouped into clusters of similar features. The input parameter space is divided into subregions by decision boundaries based on a supervised learning algorithm. Local reduced-order bases are extracted on each cluster, for which the solutions are represented as a linear combination of the basis vectors from their corresponding subregion. The proposed methodology is employed to conduct a comprehensive, exploration of the in-flight icing certification envelopes.

AB - A framework of local reduced-order modeling using machine learning algorithms is presented together with an approach to optimally select the snapshots for strongly nonlinear problems. By using an unsupervised learning algorithm, solutions are grouped into clusters of similar features. The input parameter space is divided into subregions by decision boundaries based on a supervised learning algorithm. Local reduced-order bases are extracted on each cluster, for which the solutions are represented as a linear combination of the basis vectors from their corresponding subregion. The proposed methodology is employed to conduct a comprehensive, exploration of the in-flight icing certification envelopes.

KW - machine learning algorithms

KW - in-flight icing

KW - vectors

U2 - 10.2514/1.J053654

DO - 10.2514/1.J053654

M3 - Article

VL - 53

SP - 2174

EP - 2185

JO - AIAA Journal

T2 - AIAA Journal

JF - AIAA Journal

SN - 0001-1452

IS - 8

ER -