Abstract Summary
Many novel approaches to structural health monitoring use machine learning techniques that require large data sets to train the system effectively. In an effort to enhance available data sets for bridges, for these approaches to become wider applicable, a population-based structural health monitoring (PBSHM) approach is being developed that seeks to facilitate transfer learning between notionally similar structures (i.e. sharing data). PBSHM, whilst in its infancy, has been trialled on several different structures to date. Bridges are some of the largest structures to be described with Irreducible Element (IE) Models for Attributed Graph (AG) comparisons to be carried out. For the first pool of bridges, the Canonical Form AG comparison results were indicatively positive, however, Maximum Common Subgraphs (MCS) required interrogation to tell the relevance of partial matches, highlighting the need for revisions to the bridge IE models to provide more meaningful similarity metrics. Therefore, this paper proposes a new concept for bridge IE Models that treats a bridge as a series of structures that interact via shared boundaries, as opposed to a single structure. Instrumentation of bridges (data collection) is often limited to portions of bridges being monitored/assessed, and datasets that historically exist for bridges are often confined to those portions of the bridge, e.g. Deck, Abutment, Pier. Further, as the distance an element is from measurement instrumentation increases, less information can be extracted about that element. Therefore, the extents of the new bridge IE models are set to mimic the extents of parts of bridges that common datasets pertain to (e.g. deck, abutment, pier). The revised model extents reduce the number of elements describing each structure/part of the bridge, increasing the influence of each matched element and relationship on the resulting similarity metrics. Specifically, for areas of partial matches, the relevance of the partial match is more intuitively reflected in the similarity metrics obtained for both deck and support structures.
