Please use this identifier to cite or link to this item: http://hdl.handle.net/10397/81928
Title: Parametric and nonparametric bayesian mixture models for bridge condition assessment
Authors: Chen, Ran
Advisors: Ni, Yi-qing (CEE)
Keywords: Long-span bridges
Structural health monitoring
Issue Date: 2019
Publisher: The Hong Kong Polytechnic University
Abstract: Long-span bridges are vital components in the public transportation network, while ensuring the serviceability and integrity of these assets are of great significance to a modern sustainable city. Bridge condition diagnosis based on the long-term structural health monitoring (SHM) technology has been recognised as a promising approach for achieving the condition-based preventive maintenance. In the real situation, in-service long-span bridges are normally subject to combined execution of multiple stochastic loads such as highway traffic, railway traffic, wind and temperature, which cause heterogeneous structural responses with multimodality. Conventional probabilistic assumptions for modelling the monitoring data could be quite restrictive and unverifiable, leading to high bias in characterisation of structural behaviours. More importantly, multiple sources of uncertainties are inevitably encountered in the process of data interpretation, including the intrinsic randomness, uncertain model parameter, uncertain model order, and among others. Prediction of structural performance under severe uncertainties remains as the most challenging task in the monitoring-based bridge condition assessment. The present Ph.D. thesis dedicates to develop two classes of Bayesian mixture models for condition assessment of long-span bridges that are able to better address the above scientific issues. The suspension Tsing Ma Bridge serves as the testbed for this research. A parametric Bayesian mixture model is first established to accommodate the multimodal structural responses with consideration of parametric uncertainty. Efficient Markov chain Monte Carlo (MCMC) simulation based Gibbs sampler is devised to pursue the joint posterior of the mixture parameters. Convergence of the MCMC simulation is ensured through a quantitative procedure. A Bayes factor based method is employed to find the optimal model order of the mixture model. The parametric Bayesian mixture model is utilised to identify neutral axis positions of the Tsing Ma Bridge under stochastic traffic loads. A novel neutral axis position based damage identification method is proposed for real-time alert of abnormal bridge condition. Single and multiple damages of the bridge deck are confidently detected by the proposed damage indexes based on neutral axis change. Subsequently, a nonparametric Bayesian mixture model is further developed to allow the model complexity automatically adapts to the monitoring data with the joint consideration of the parametric and model order uncertainties. A collapsed Gibbs sampler is devised to pursue the nonparametric estimation of the mixture density samples. Convergence diagnosis of the MCMC simulation is achieved based on a quantitative procedure. Both the parametric and nonparametric Bayesian mixture models are used to characterise the live load effects of the bridge under multi-load condition. An updatable conditional reliability index is formulated based on the first-order reliability method (FORM) that is able to account for both the aleatory and epistemic uncertainties arising from load effect characterisation. Bayesian updating of the reliability for the bridge deck is carried out based on the accumulation of monitoring data. A clear vision on the safety risk can be learnt by bridge authorities through reporting not only the average structural reliability but also its associated uncertain range.
Description: xxii, 236 pages : color illustrations
PolyU Library Call No.: [THS] LG51 .H577P CEE 2019 ChenR
URI: http://hdl.handle.net/10397/81928
Rights: All rights reserved.
Appears in Collections:Thesis

Files in This Item:
File Description SizeFormat 
991022347055403411_link.htmFor PolyU Users168 BHTMLView/Open
991022347055403411.pdfFor All Users7.92 MBAdobe PDFView/Open
Show full item record
PIRA download icon_1.1View/Download Contents

Page view(s)

7
Citations as of May 6, 2020

Download(s)

1
Citations as of May 6, 2020

Google ScholarTM

Check


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.