Effects of disparate information levels on bridge management and safety

Thumbnail Image
CiaranHanleyThesis_Final.pdf(4.66 MB)
Full Text E-thesis
Hanley, Ciaran
Journal Title
Journal ISSN
Volume Title
University College Cork
Published Version
Research Projects
Organizational Units
Journal Issue
Maintenance planning and life-cycle assessment methods for bridge networks have received large research interest for many years; with modern emphasis often based on probabilistic approaches, due to their ability to handle uncertainty. This allows for risk-based approaches to quantify structural safety, which is largely seen as a superior approach than the deterministic methods typically used in practice. Structural safety can broadly be defined as an acceptable level of chance/probability that the structure will not fail in its function; i.e. to resist the loads/actions to which it is subjected to. The structural reliability method provides for the computation of structural safety by accounting for probabilistically uncertain load models and uncertainties around the return period of extreme load events, as well as the uncertainty in the resistance capacity of the structural system. In addition to a single-point-in-time evaluation of structural safety, the life-cycle performance can be evaluated using physical models for future deterioration; again, constrained under uncertain information about future deterioration. However, being a probabilistic method, it can be somewhat subjective in nature, based on the availability of accurate data and the reliance on expert knowledge, and thus sensitive to the parameters of the input model which rely on the level of information available for the problem at hand. While structural safety is the apex of modern maintenance planning and lifecycle assessment, the most prevalent performance indicator for which future maintenance and intervention decisions are made come from visual inspection based condition ratings. These visual inspections are used to evaluate the extent of deterioration present and assign a condition rating based on a predefined scale of damage, after which bridge managers trigger further assessment or intervention actions based on acceptable damage levels. Again, in evaluating a single or small number of bridges, there is a degree of subjectivity and reliance on expert knowledge that is also seen with probabilistic assessment methods. Unlike structural reliability, which often suffers from a lack of available or accurate information, condition rating data for large bridge networks generate a large repository of data which provides an excellent opportunity to look at data on a larger scale than is currently implemented in practice. This results in disparate levels of information being available for bridge networks, with large amounts of lower level information and small amounts of detailed information. In this thesis, how disparate information levels affect these two assessment methods will be explored and efforts to mitigate against the uncertainty in the information will be suggested. It will be shown that: Reliability-based calibrations of bridges are possible through observed clustering of parametric importance and sensitivity measures, based on uncertainty in relation to the available information for probabilistic modelling (Hanley and Pakrashi 2016) Existing bridges assessed under code-defined traffic load are sensitive to safety reclassification due to evolving definitions, leading to misinterpretation of the actual state of the structure and, thus, a misallocation of resources (Hanley et al. 2017a) Bridges designed under modern, more conservative code-defined traffic load models and assessed under probabilistic load models can expect a longer projected service life before intervention is required, and that the initial construction cost of this conservatism is largely offset when lifecycle cost is considered (Hanley et al. 2016a) The use of multivariate analysis methods are applicable to modern bridge management systems that store large amounts of data, and that these methods can provide for clustering of bridges based on their structural forms and states of disrepair (Hanley et al. 2015) Large groups of specific bridge types have well-defined, consistent factor structures, whereby a bespoke linear combination of individual elemental condition ratings provide an accurate assessment of the bridge’s overall condition rating; improving on currently implemented decision tools in existing bridge management systems (Hanley et al. 2016b, 2017b,c) This work provides a basis for which further research can be undertaken into developing an information-driven probabilistic decision making framework, leading to the quantification of the value of disparate information levels. The potential future applications include incorporating the derived underlying factor structure of large data-sets of bridges directly into structural reliability methods through probabilistic graphical models, such as Bayesian Belief Networks; thus providing a more robust, information driven framework from which to make decisions under uncertainty.
Bridge engineering , Structural reliability , Multivariate analysis
Hanley, C. 2017. Effects of disparate information levels on bridge management and safety. PhD Thesis, University College Cork.