J. Jarmulak, “Case-based classification of ultrasonic B-scans: case-base organisation and case retrieval”, in: Advances in Case-Based Reasoning: EWCBR-98 Proceedings, B. Smyth and P. Cunningham (eds.), Springer, pp. 100-111. (pdf)
Abstract. Dutch Railways use a special train for the ultrasonic inspection of rails. The output of the ultrasonic scanning system installed on the train consists of echo images – so-called B-scans. The B-scans are classified according to the images of rail constructions, noise artifacts, and/or defects that they contain. Currently, at TNO Institute of Applied Physics a hybrid rule-based / case-based classification system is being built for the train. This paper describes the way the cases are represented and matched in the system, the way the case-base is organised, and the way the case retrieval is done. Interesting aspects of the system are: use of fuzzy membership functions for case matching, decision tree (C4.5) based top-level case-base organisation, and clustered bottom-level case-base organisation.