City Research Online

Neural networks and Monte Carlo simulation to determine structural system reliability: application to static and blast loadings

Thuy Dung, V. (2020). Neural networks and Monte Carlo simulation to determine structural system reliability: application to static and blast loadings. (Unpublished Doctoral thesis, City, University of London)

Abstract

In assessment of the structure-based reliability, there are two levels of reliability required to consider including (1) structural member reliability and (2) system reliability. The former is originated through the failure of a particular component that partial local reliability in a structural system might possibly cause loss of serviceability. However, it is argued by many researchers that structural system is often designed to possess a high level of redundancy making its collapse to occur most likely because of the combined effect of several different failure modes rather than only one particular member. For this reason, it is important to consider both structural member and system reliability in forming any problems related to the structural failure.

Regardless of this statement, literature in the field of structural reliability are focused on the structural member reliability leaving the system reliability to received very little attentions. Although recently, some researches devoted to considering system reliability, the accuracy of this assessment has been considered as a serious issue, which is driven from the fact that these models developed to assess reliability are often assumed to be in linear or weakly nonlinear performance functions. For this reason, the objective of this paper is to propose the approach employed Monte Carlo Simulation and Neural Network to effectively calculate the system reliability of the structural system.

In order to determine the structural system reliability, the proposed method contains the two main stage. In the first stage, the β-unzipping method is employed to determine reliability analysis of structural systems at different level such as Level 0 (on the basic of a single structural element), Level 1 (considering the structural system as a series system), Level 2 (on the basic of a series system where the elements are parallel systems each - with critical pairs of failure elements), and Level 3 (on the basic of a series system where the elements are parallel systems each - with critical triples of failure elements). In the second stage, the Monte Carlo Simulation with Importance Sampling is first employed to general the sample population, which will be then used to trained, test and predict the system reliability of the structure by Back-Propagation Neural Network Algorithm.

The proposed method was validated against the conventical β-unzipping method to estimate the structural system reliability. The results indicate the closed and yet more accurate reliability index and failure probability of the structural system in consideration of its system reliability analysis. This study is thus moving further by demonstrating the whole process of application of Monte Carlo Simulation with the Importance Sampling Techniques and Neural Network with Back-Propagation Algorithm towards the case study of a 10-bar truss structure. The promising results indicate the potential of employing the proposed method to solve the complex problem of the structural system reliability.

The proposed was then applied to assess the structure system reliability employed for a CFTA girder under blast loading and the obtained results were compared against current Eurocodes guidelines for the structure in the event of extreme loading like explosion. The results prove the possibility of employing the proposed method to solve the complex problem of the structural system reliability assessment under explosion in consideration of the loading uncertainties that the application of the proposed method.

Publication Type: Thesis (Doctoral) T Technology > TA Engineering (General). Civil engineering (General) Doctoral ThesesSchool of Science & Technology > School of Science & Technology Doctoral ThesesSchool of Science & Technology > Engineering
Text - Accepted Version
This document is not freely accessible until 30 November 2024 due to copyright restrictions.

To request a copy, please use the button below.

Request a copy