On Similarity in Case-based Reasoning for Structural Health Monitoring (Extended Abstracts)

Case-based Reasoning (CBR) is a problem solving method, whereby already known knowledge in the form of cases is used to solve new problems. New problems are compared to the knowledge in a case base; the solution usually is adopted from the "nearest" of these cases. It is well known that the comparison of cases and consequently the provision of indicators for similarity between them are main principles especially of CBR’s retrieve phase and mostly are sticking points in the design of a Case-based system. Searching in general always was a very popular data processing operation. While at first models (algorithms, index structures) for exact-match searches were developed, the search paradigms soon changed and requirements like the (time-)efficient calculation of similarities between more complex objects were arising. Similar objects can be described as a set of objects being in some sense “near” to each other. The metric space notation satisfies the needs of representations and abstractions in this connection. Similarity search in metric spaces can be seen as a kind of ranking of objects in respect to a query object and to certain similarities. Nevertheless in many cases it is not an easy task or even impossible to transform data to make it representable in a metric space. This contribution should try to evaluate methods for knowledge representation, similarity calculation and data indexing especially for the field "Structural Health Monitoring". Measurements from a certain structure have to be interpreted and conclusions to its condition are drawn by engineers of this domain. Due to the individuality of each structure and the complexity of the measurement interpretation process our goal is to aid the engineer using a case-based decision support system.