An efficient approach for computing analytical non-parametric fragility curves. (July 2020)

• Record Type: Journal Article
• Authors: Altieri, Domenico; Patelli, Edoardo
• Is Part Of: Structural safety. Volume 85(2020)
• Publication Date: 2020-07
• Subjects: Seismic fragility curves -- Conditional failure probability -- Reliability analysis -- Performance based earthquake engineering; Structural stability -- Periodicals; Safety factor in engineering -- Periodicals; Reliability (Engineering) -- Periodicals; Constructions -- Stabilité -- Périodiques; Coefficient de sécurité en ingénierie -- Périodiques; Fiabilité -- Périodiques; 620.86
• DOI: 10.1016/j.strusafe.2020.101956
• Languages: English
• ISSNs: 0167-4730

Abstract

Highlights: A non-parametric methodology is developed for seismic analytical fragility analysis. No assumptions are required on the final conditional cumulative density function. Surrogate probability density functions coupled with Multinomial Logistic Regression. Complex and noise failure regions can be identified. The approach overcomes the performance of existing methods with rare failure domains. Abstract: Fragility curves are used in earthquake engineering for assessing the seismic vulnerability of structures or systems. Direct estimations of fragility curves by means of simulation-based approaches lead generally to relevant computational costs, especially when the failure region is characterized by small probabilities of occurrence. Simplified hypotheses are therefore introduced in the common practice to approximate the dependency between the structural response and the associated seismic intensity level. The study proposes a non-parametric methodology to estimate analytical fragility curves without specific assumptions on their final shape. The approach starts by identifying all the subsets characterized by the same values of the chosen seismic intensity measure parameter. Then, the failure region is mapped by means of a classification algorithm coupled with a polynomial kernel. Finally, the conditional failure probability is computed by associating the samples generated in each subset to the corresponding classification score. A stochastic earthquake model is employed to define the seismic dataset and the uncertainty associated with the ground motion records. Two case studies are analyzed in which the non-parametric methodology is compared against three popular parametric approaches and a reference solution. The proposed approach shows an overall higher accuracy and efficiency, especially in case of rare failure domains.