Quantifying Uncertainty in Steel Member Reliability via Global Sensitivity Indices

Abstract

The study presents a comprehensive global sensitivity investigation focusing on the failure probability in torsion-loaded thin-walled steel members with uncertainties in material, geometric, and loading parameters. Two distinct sensitivity measures are scrutinised: one formulated through variance analysis of a binary failure outcome and the other based on discrete entropy. Total-effect indices, encompassing all interaction orders, are employed to assess variable importance. The results indicate that the long-term variable load significantly influences the uncertainty in structural reliability, while material and geometric parameters play a secondary role. The entropy-based measure accentuates the distinction between dominant and less influential inputs, while the variance-based approach captures a broader sensitivity structure. The findings underscore the impact of the selected sensitivity metric on input ranking and its implications for reliability assessment.The study presents a comprehensive global sensitivity investigation focusing on the failure probability in torsion-loaded thin-walled steel members with uncertainties in material, geometric, and loading parameters. Two distinct sensitivity measures are scrutinised: one formulated through variance analysis of a binary failure outcome and the other based on discrete entropy. Total-effect indices, encompassing all interaction orders, are employed to assess variable importance. The results indicate that the long-term variable load significantly influences the uncertainty in structural reliability, while material and geometric parameters play a secondary role. The entropy-based measure accentuates the distinction between dominant and less influential inputs, while the variance-based approach captures a broader sensitivity structure. The findings underscore the impact of the selected sensitivity metric on input ranking and its implications for reliability assessment.