Abstract
The article investigates the application of a new type of global quantile-oriented sensitivity analysis (called QSA in the article) and contrasts it with established Sobol’ sensitivity analysis (SSA). Comparison of QSA of the resistance design value (0.1 percentile) with SSA is performed on an example of the analysis of the resistance of a steel IPN 200 beam, which is subjected to lateral-torsional buckling. The resistance is approximated using higher order polynomial metamodels created from advanced non-linear FE models. The main, higher order and total effects are calculated using the Latin Hypercube Sampling method. Noticeable differences between the two methods are found, with QSA apparently revealing higher sensitivity of the resistance design value to random input second and higher order interactions (compared to SSA). SSA cannot identify certain reliability aspects of structural design as comprehensively as QSA, particularly in relation to higher order interactions effects of input imperfections. In order to better understand the reasons for the differences between QSA and SSA, two simple examples are presented, where QSA (median) and SSA show a general agreement in the calculation of certain sensitivity indices.
Highlights
The fundamental characteristic of safety and reliability of the design of load bearing structures is the design value of resistance (Galambos, 1998)
This study focuses on issues that are related to the accuracy in the tails, as it influences the quantile-oriented sensitivity analysis (QSA) results
Proportions QTi differ from proportions Qi (Figure 7) or Si (Figure 5), it can be concluded that the random variability of imperfection has a different effect on the uncertainty of the design quantile Rd than it has on the random resistance R of the structure
Summary
The fundamental characteristic of safety and reliability of the design of load bearing structures is the design value of resistance (Galambos, 1998). Common sensitivity analysis methods (Saltelli, Chan, & Scott, 2004) monitor the correlation between model inputs and the output or the effects of random inputs on the variance of model output These methods may not be suitable for analysing the effects of random imperfections on the design quantile Rd. In accordance with the classical utility theory, variance is not sufficient for the determination of the decision-maker state of knowledge. The first and higher order quantile contrast indices have been compared with Sobol’ global sensitivity analysis in an example using Ishigami Function in Kala (2018). The aim of the present article is global quantile-oriented sensitivity analysis (QSA) of Rd performed by numerical estimation of all first and higher order quantile contrast indices. QSA revealing higher sensitivity of the resistance design value to random input higher order interactions (compared to SSA)
Sign-in/Register to view highlights & summary 
Published Version (
Free)
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call