Efficient global sensitivity analysis framework and approach for structures with hybrid uncertainties

Research output: Contribution to journalArticleResearchpeer review

Authors

Research Organisations

External Research Organisations

  • University of Liverpool
  • Tongji University
View graph of relations

Details

Original languageEnglish
Article number117726
Number of pages31
JournalComputer Methods in Applied Mechanics and Engineering
Volume436
Early online date8 Jan 2025
Publication statusE-pub ahead of print - 8 Jan 2025

Abstract

Engineering structures are often subject to various types of uncertainties, including random, interval, and fuzzy uncertainties. When dealing with hybrid uncertainties, global sensitivity analysis (GSA) becomes particularly challenging due to the computational complexity associated with double-loop procedures in numerical simulations. In this paper, an efficient framework for GSA with hybrid uncertainties is proposed. Generally, surrogate models, such as the radial basis function neural network (RBFNN), are used to reduce computational efforts by replacing real response functions. Then global sensitivity indices can be obtained efficiently by combining with numerical simulation-based methods. However, this process can introduce additional sources of error due to the stochastic nature of the simulations. This paper presents a general framework for GSA with hybrid uncertainties, where variance-based indices for random, interval and fuzzy inputs are defined. Furthermore, to avoid the error propagation commonly associated with simulation-based techniques and to improve the computational efficiency, analytical solutions for these indices and the gradient of the output variance are derived based on the RBFNN. An additional validation strategy is designed to verify the importance ranking of uncertain inputs. Four applications are introduced to demonstrate the efficiency and accuracy of the proposed method for GSA with hybrid uncertainties.

Keywords

    Analytical solution, Hybrid uncertainties, Radial basis function network, Sobol’ indices, Variance-based global sensitivity analysis

ASJC Scopus subject areas

Cite this

Efficient global sensitivity analysis framework and approach for structures with hybrid uncertainties. / Liu, Jinxing; Shi, Yan; Ding, Chen et al.
In: Computer Methods in Applied Mechanics and Engineering, Vol. 436, 117726, 01.03.2025.

Research output: Contribution to journalArticleResearchpeer review

Download
@article{c03e0a1aa48a42cd8b2eccdab5b0a83b,
title = "Efficient global sensitivity analysis framework and approach for structures with hybrid uncertainties",
abstract = "Engineering structures are often subject to various types of uncertainties, including random, interval, and fuzzy uncertainties. When dealing with hybrid uncertainties, global sensitivity analysis (GSA) becomes particularly challenging due to the computational complexity associated with double-loop procedures in numerical simulations. In this paper, an efficient framework for GSA with hybrid uncertainties is proposed. Generally, surrogate models, such as the radial basis function neural network (RBFNN), are used to reduce computational efforts by replacing real response functions. Then global sensitivity indices can be obtained efficiently by combining with numerical simulation-based methods. However, this process can introduce additional sources of error due to the stochastic nature of the simulations. This paper presents a general framework for GSA with hybrid uncertainties, where variance-based indices for random, interval and fuzzy inputs are defined. Furthermore, to avoid the error propagation commonly associated with simulation-based techniques and to improve the computational efficiency, analytical solutions for these indices and the gradient of the output variance are derived based on the RBFNN. An additional validation strategy is designed to verify the importance ranking of uncertain inputs. Four applications are introduced to demonstrate the efficiency and accuracy of the proposed method for GSA with hybrid uncertainties.",
keywords = "Analytical solution, Hybrid uncertainties, Radial basis function network, Sobol{\textquoteright} indices, Variance-based global sensitivity analysis",
author = "Jinxing Liu and Yan Shi and Chen Ding and Michael Beer",
note = "Publisher Copyright: {\textcopyright} 2025 The Authors",
year = "2025",
month = jan,
day = "8",
doi = "10.1016/j.cma.2024.117726",
language = "English",
volume = "436",
journal = "Computer Methods in Applied Mechanics and Engineering",
issn = "0045-7825",
publisher = "Elsevier",

}

Download

TY - JOUR

T1 - Efficient global sensitivity analysis framework and approach for structures with hybrid uncertainties

AU - Liu, Jinxing

AU - Shi, Yan

AU - Ding, Chen

AU - Beer, Michael

N1 - Publisher Copyright: © 2025 The Authors

PY - 2025/1/8

Y1 - 2025/1/8

N2 - Engineering structures are often subject to various types of uncertainties, including random, interval, and fuzzy uncertainties. When dealing with hybrid uncertainties, global sensitivity analysis (GSA) becomes particularly challenging due to the computational complexity associated with double-loop procedures in numerical simulations. In this paper, an efficient framework for GSA with hybrid uncertainties is proposed. Generally, surrogate models, such as the radial basis function neural network (RBFNN), are used to reduce computational efforts by replacing real response functions. Then global sensitivity indices can be obtained efficiently by combining with numerical simulation-based methods. However, this process can introduce additional sources of error due to the stochastic nature of the simulations. This paper presents a general framework for GSA with hybrid uncertainties, where variance-based indices for random, interval and fuzzy inputs are defined. Furthermore, to avoid the error propagation commonly associated with simulation-based techniques and to improve the computational efficiency, analytical solutions for these indices and the gradient of the output variance are derived based on the RBFNN. An additional validation strategy is designed to verify the importance ranking of uncertain inputs. Four applications are introduced to demonstrate the efficiency and accuracy of the proposed method for GSA with hybrid uncertainties.

AB - Engineering structures are often subject to various types of uncertainties, including random, interval, and fuzzy uncertainties. When dealing with hybrid uncertainties, global sensitivity analysis (GSA) becomes particularly challenging due to the computational complexity associated with double-loop procedures in numerical simulations. In this paper, an efficient framework for GSA with hybrid uncertainties is proposed. Generally, surrogate models, such as the radial basis function neural network (RBFNN), are used to reduce computational efforts by replacing real response functions. Then global sensitivity indices can be obtained efficiently by combining with numerical simulation-based methods. However, this process can introduce additional sources of error due to the stochastic nature of the simulations. This paper presents a general framework for GSA with hybrid uncertainties, where variance-based indices for random, interval and fuzzy inputs are defined. Furthermore, to avoid the error propagation commonly associated with simulation-based techniques and to improve the computational efficiency, analytical solutions for these indices and the gradient of the output variance are derived based on the RBFNN. An additional validation strategy is designed to verify the importance ranking of uncertain inputs. Four applications are introduced to demonstrate the efficiency and accuracy of the proposed method for GSA with hybrid uncertainties.

KW - Analytical solution

KW - Hybrid uncertainties

KW - Radial basis function network

KW - Sobol’ indices

KW - Variance-based global sensitivity analysis

UR - http://www.scopus.com/inward/record.url?scp=85214295672&partnerID=8YFLogxK

U2 - 10.1016/j.cma.2024.117726

DO - 10.1016/j.cma.2024.117726

M3 - Article

AN - SCOPUS:85214295672

VL - 436

JO - Computer Methods in Applied Mechanics and Engineering

JF - Computer Methods in Applied Mechanics and Engineering

SN - 0045-7825

M1 - 117726

ER -

By the same author(s)