Efficient probabilistic approaches for the seismic reliability analysis of structures subject to seismic pounding or equipped with non-linear viscous dampers

Altieri, Domenico
(2019) Efficient probabilistic approaches for the seismic reliability analysis of structures subject to seismic pounding or equipped with non-linear viscous dampers. PhD thesis, University of Liverpool.

[img] Text
Access to this file is embargoed until 1 February 2022.

Download (10MB)


Seismic reliability analysis represents a critical stage in the structural design process. The response of structures and systems under the action of dynamic loads can lead to both material (e.g. structural damage) and immaterial losses (e.g. consequences on the human lives). A robust assessment of the seismic reliability can be obtained through probabilistic approaches in order to account for the uncertainty that characterize both the dynamic load and the model definition. Different simulation-based methodologies can be employed to propagate the input variability over the structural response and their efficiency plays a relevant role, especially in case of complex models or rare failure regions. Similarly, for computational expensive models, machine learning approaches can represent an alternative powerful tool to approximate the original model through mathematical formulae or dimensionality reductions. The thesis work aims at presenting a series of case studies and simulation-based approaches to study, assess, predict and analyze the seismic performance within a probabilistic framework. Different uncertainty propagation techniques are employed and in turn coupled with optimization, machine learning and sensitivity algorithms to capture more insights from the physical problem by increasing at the same time the final efficiency and accuracy. Seismic vulnerability studies are therefore performed over three different case studies. A system of contiguous nuclear fuel assemblies prone to seismic induced impacts is analyzed both in terms of fragility curves and surrogate models applicability. In particular, the adopted approximated models are calibrated and tested to predict the maximum permanent deformation in case of seismic events, showing a good final accuracy. A simplified physical model is than analyzed by means of a nondimensionalization of the equation of motion to provide more insights of the seismic pounding phenomenon in bridges. Multiple nondimensionalized response parameters are therefore identified and studied with both parametric and sensitivity analyses to reach a more comprehensive assessment of the structural seismic reliability. The problems arising from the fragility curves estimation are also discussed. In general, seismic fragility analyses return the evolution of the structural failure probability for increasing levels of seismic intensity measures and they can be used, for instance, to compute the mean annual failure rate when combined with the local seismic hazard. An approximated non-parametric approach is proposed to overcome the limitations of a lognormal-based approach by removing any preliminary assumption on the final distribution of the system failure dataset. The new methodology is then validated against classic approaches showing promising performances in terms of efficiency and accuracy. Finally, a probabilistic-based reliability analysis that account for the full record-to record variability is efficiently coupled with an optimization algorithm to reach a more robust final structural configuration. The methodology is applied to reach the optimal design of non-linear viscous dampers by guaranteeing predefined seismic performances.

Item Type: Thesis (PhD)
Divisions: Fac of Science & Engineering > School of Engineering
Depositing User: Symplectic Admin
Date Deposited: 18 Aug 2020 11:13
Last Modified: 09 Jan 2021 02:35
DOI: 10.17638/03080923
URI: https://livrepository.liverpool.ac.uk/id/eprint/3080923