TY - GEN
T1 - On the accuracy of kriging model in active subspaces
AU - Palar, Pramudita Satria
AU - Shimoyama, Koji
N1 - Funding Information:
We would like to express our gratitude to Jouke De Baar from The University of New South Wales Canberra, Australian Defence Force Academy, who provide us the aerodynamic data for the FFAST airfoil.
Publisher Copyright:
© 2018, American Institute of Aeronautics and Astronautics Inc, AIAA. All rights reserved.
PY - 2018
Y1 - 2018
N2 - Investigation on the accuracy of Kriging models in active subspaces was carried out. Our goal is to investigate how much a gain in the accuracy of Kriging model can be obtained by rotating and/or reducing the coordinate set through the active subspaces method procedure. The Kriging model in the transformed coordinate can then be used for various purposes such as uncertainty quantification and optimization. In this paper, we estimate the active subspaces structure with true and estimated gradient information directly from the Kriging model. Two test problems (i.e., borehole and inviscid FFAST airfoil problem) were employed as benchmark problems for our investigation. The results show that one can construct a more accurate Kriging model when a suitable new coordinate is discovered. We observe that using the estimated gradient is a far from optimal approach to find the proper coordinate rotation compared to that of, obviously, true gradient information. When compared to gradient-enhanced Kriging, the potential lowest accuracy that can be achieved by Kriging in the rotated space is almost comparable to that of the gradientenhanced one. This means that there is a potential for the Kriging model to achieve an accuracy that almost matches to that of the gradient-enhanced Kriging without utilizing gradient information.
AB - Investigation on the accuracy of Kriging models in active subspaces was carried out. Our goal is to investigate how much a gain in the accuracy of Kriging model can be obtained by rotating and/or reducing the coordinate set through the active subspaces method procedure. The Kriging model in the transformed coordinate can then be used for various purposes such as uncertainty quantification and optimization. In this paper, we estimate the active subspaces structure with true and estimated gradient information directly from the Kriging model. Two test problems (i.e., borehole and inviscid FFAST airfoil problem) were employed as benchmark problems for our investigation. The results show that one can construct a more accurate Kriging model when a suitable new coordinate is discovered. We observe that using the estimated gradient is a far from optimal approach to find the proper coordinate rotation compared to that of, obviously, true gradient information. When compared to gradient-enhanced Kriging, the potential lowest accuracy that can be achieved by Kriging in the rotated space is almost comparable to that of the gradientenhanced one. This means that there is a potential for the Kriging model to achieve an accuracy that almost matches to that of the gradient-enhanced Kriging without utilizing gradient information.
UR - http://www.scopus.com/inward/record.url?scp=85044617684&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85044617684&partnerID=8YFLogxK
U2 - 10.2514/6.2018-0913
DO - 10.2514/6.2018-0913
M3 - Conference contribution
AN - SCOPUS:85044617684
SN - 9781624105326
T3 - AIAA/ASCE/AHS/ASC Structures, Structural Dynamics, and Materials Conference, 2018
BT - AIAA/ASCE/AHS/ASC Structures, Structural Dynamics, and Materials
PB - American Institute of Aeronautics and Astronautics Inc, AIAA
T2 - AIAA/ASCE/AHS/ASC Structures, Structural Dynamics, and Materials Conference, 2018
Y2 - 8 January 2018 through 12 January 2018
ER -