Where is the end of a Bridge (model)?

Abstract

Abstract Bridge SHM solutions have been developed to assist with the assessment and monitoring of bridges. State of the art bridge SHM solutions tend to be data based, where machine learning algorithms are trained using large, historical bridge datasets, and outlier analysis is subsequently used for anomaly detection. However, most bridges lack the required healthy state data for machine learning approaches to be considered, and many bridges are not in a healthy condition to collect the required data from. A population based structural health monitoring (PBSHM) approach has recently been proposed that seeks to facilitate knowledge transfer between similar structures. The approach proposes that if two structures are similar enough, there could be scope to make SHM inferences between the structures. The ability to make inferences between bridges, which are currently monitored in isolation, would be highly valuable as a bridge management tool, particularly if datasets could be leveraged between bridges through transfer learning. However, before knowledge can be shared between bridges, there is first a need to identify bridges that are similar enough for inferences to be made. The PBSHM approach proposes the use of Irreducible Element (IE) models to describe structures, which allows Attributed Graphs (AG) to generated and compared for similarity using graph theory techniques. The general method for comparing structures was trialled on bridges previously, however the resulting similarity metrics were for the whole bridge as opposed to particular common zones of interest e.g. the deck. This paper instead proposes that bridges be modelled as subsections of structures that interact via shared boundaries (i.e., points of articulation such as bearings), as opposed to whole structures. Bridge datasets are often limited to the part of the bridge that was investigated, i.e., datasets particular to bridge decks, abutments etc. Therefore, the extents of the IE models proposed in this paper are set to only include elements that would pertain to a particular dataset. In particular, two beam and slab bridges are each described with bridge deck, abutment and pier IE models to trial the concept. The revised extents of the bridge IE models reduced the number of elements being compared, resulting in increased resolution graph comparisons being carried out and more meaningful similarity metrics between the bridge parts being achieved.