Predicting remaining service potential of railway bridges based on visual inspection data

Year

2015

Degree Name

Doctor of Philosophy

Department

School of Mechanical, Materials and Mechatronic Engineering

Abstract

Asset authorities need to predict the future condition of infrastructure, including railway bridges as part of the process of managing asset integrity and to prioritise maintenance, repair and rehabilitation activities given a limited budget. Visual inspection-based qualitative rating of component condition on an ordinal scale generated during periodic inspection cycles may be the primary source of information. Typically this data is sparse and intermittent. A complete sequence of data is not available for most of the individual bridges or bridge components and then only for a limited period of observation relative to the total period of use to date. A review of applicable existing modelling approaches shows that they may inadequately predict the future condition of bridge components. Although many of these approaches typically assume a Markov chain transition process, they do not deal adequately with missing data which is typical of this situation. Furthermore, non-linear optimisation techniques used in many existing methods are known to lead to inaccurate condition prediction, being susceptible to settling at local optima. More accurate prediction would allow more cost effective and certain management of the integrity of structures, specifically railway bridges.

By using the proposed Bayesian inference-based methods, the accuracy of various Markov deterioration models has been improved well particularly when applied to a situation where data is missing. This has been successfully demonstrated for a limited set of data associated with a network of railway bridges. Such techniques may also be applicable to other structures including pipeline networks, pavement systems and others.

The results here are limited by the restricted data sets that were used for validation. Validation on data from a broader range of component groups is required for gaining further evidence of the suitability of the developed models in practical application.

FoR codes (2008)

010206 Operations Research, 010303 Optimisation, 080201 Analysis of Algorithms and Complexity, 090505 Infrastructure Engineering and Asset Management

This thesis is unavailable until Thursday, January 17, 2019

Share

COinS
 

Unless otherwise indicated, the views expressed in this thesis are those of the author and do not necessarily represent the views of the University of Wollongong.