A new moment-independent measure for reliability-sensitivity analysis

Abstract

This thesis aims to investigate the advances in the field of reliability-sensitivity analysis since the introduction of the Importance
Factors in FORM. In sensitivity analysis (SA), great efforts have been made to develop global and moment-independent sensitivity
indices. However, previous work in global reliability-sensitivity analysis (RSA) has not yet fully exploited or adopted the advantages of the global moment-independent measures. To this end, this thesis introduces a new global reliability-sensitivity index based on a modification of the $\delta$-index by Borgonovo, 2007. In addition, a new computational approach, denoted as
Enhanced Adaptive Kriging Monte Carlo Simulation (EAK-MCS), is developed to compute the indices at the lowest possible costs. The new set of indices is tested and compared against other established indices on four reliability benchmark problems. Eventually, the convergence of the new computational approach is compared against the ordinary Adaptive Kriging Monte Carlo Simulation (AK-MCS) technique. It is concluded that the new set of indices delivers numerically stable and equivalently ranked index values compared to
the existing global indices. In combination with either the AK-MCS or the EAK-MCS meta-modelling strategy, the indices may be computed with only a few hundreds of full model evaluations.
 

Keywords

Sensitivity analysis, global reliability-sensitivity analysis,
moment-independent index, Adaptive Kriging

BibTeX cite

MSCTHESIS{FSchmidThesis,
author = {Schmid F.},
title = {A new moment-independent measure for reliability-sensitivity analysis},
school = {ETH Zurich, Zurich, Switzerland},
year = {2018}
}