Response-surface-based Embankment reliability under incomplete probability information

Fan Wang, Heng Li, Qi Ling Zhang

Research output: Journal article publicationJournal articleAcademic researchpeer-review

8 Citations (Scopus)

Abstract

Reliability evaluations can be challenging if the limit-state surface (LSS) is implicit and the probability information is incomplete in that only marginal distributions and correlations are given. To address the problem, this study adopted the response-surface method based on an adaptive relevance vector machine (aRVM) to approximate the implicit LSS, and the copula approach was used to reconstruct the joint distributions based on incomplete probability information. The Rosenblatt transformation was used to transform the random variables from the original random space into the independent standard normal space for the first-/second-order reliability method (FORM/SORM) approximations. Five different copulas-the normal, Frank, Clayton, CClayton, and t copulas-were adopted to represent different dependence structures and examine their impacts on the failure probability. Results from the numerical example show that the copula effect was negligible if the shear strength parameters were uncorrelated or fully correlated. However, when the correlation coefficient was 0.6 or 0.8, the probabilistic result corresponding to the commonly used normal copula was 5.6% higher or 3.1% lower if the CClayton or the Frank copula was used to model the dependence structure, respectively.
Original languageEnglish
Article number06017021
JournalInternational Journal of Geomechanics
Volume17
Issue number12
DOIs
Publication statusPublished - 1 Dec 2017

Keywords

  • Adaptive relevance vector machine
  • Copula
  • Embankment reliability
  • Incomplete probability information
  • Rosenblatt transformation

ASJC Scopus subject areas

  • Soil Science

Fingerprint

Dive into the research topics of 'Response-surface-based Embankment reliability under incomplete probability information'. Together they form a unique fingerprint.

Cite this