When is a bridge not an aeroplane?

One of the main problems in data-based SHM is the scarcity of labelled data from damage states of structures. This issue limits applications of machine learning technologies to supervised learning and cuts off access to higher-level diagnostic systems. In recent years, Population-Based SHM has been proposed as a means of alleviating the problem by allowing transfer of knowledge between structures in populations, in such a way that damage-state data for individual structures can inform diagnostics across the population. However, transfer is only possible between structures or substructures which are sufficiently similar that the inference is meaningful and positive. In order to measure similarity between structures, the ideas of irreducible-element (IE) and attributed-graph (AG) models have been introduced, such that structures are represented in an abstract metric space which allows principled comparisons. This framework will work, in principle, even for heterogeneous populations of disparate structures. This paper gives a high-level overview of the concepts involved, and illustrates them via two example comparisons.