P. van’t Hoff. Ontwerp van case-base voor eddy current CBR system: Optimal case representation and retrieval, M.Sc. Thesis, TU Delft, the Netherlands, 1996. (pdf)
Summary: For the support of operators working with eddy current inspection of heat exchangers
a system is built at TNO-TPD to automatically determine the seriousness of defects. To do this the type of defect has to be determined first. The type of defect can be determined by looking at the shape of the curve. A case-based reasoning technique is used to build a system which can identify the type of a curve which is measured using the curves of defect of which the type is known.
This research has focused on the part of the system that find the best case in the case-base. Herefore a representation method and a corresponding distance measure has been developed. Because the shape of the curves determines the type of defect this distance measure should be invariant for the position, size and orientation of the curves. A number of possible algorithms for the calculation of a distance between curves have been tested. Furthermore a number of classification methods to determine the type of defect using the distances between the curves.
The best algorithm gives good results. Depending on the criteria 75 to 95 percent of the defects is correctly classified. This is much better than the results of algorithms that were used before this research.