J. Jarmulak, “Case-Based Reasoning for Automatic Interpretation of Data from Eddy-Current Inspection”, in: Review of Progress in Quantitative Nondestructive Evaluation, Vol. 17, D.O. Thompson and D.E. Chimenti (eds.), Plenum Publishing, New York, 1998, pp. 767-774. (pdf)
In the design of the eddy-current inspection systems that have been reported to be able to interpret EC data automatically one can distinguish use of two methodologies. One is the use of classifiers to assign the signals to several predefined defect classes. Another is the use of expert systems to reason about the shape and other parameters of the signals in order to determine the defect types they represent. Both sorts of systems are usually designed with a specific inspection type in mind (e.g. steam generators of nuclear power plants). Adapting these systems to a different inspection type requires a considerable effort; therefore, they are generally not suitable for application in (petro-) chemical industry where heat-exchanger types vary from one inspection to another. This paper suggests case-based reasoning (CBR) as a methodology which is well suited for such applications. In this respect, one of the most important advantages of CBR systems is their ability to learn during use.
The paper begins with a presentation of existing automatic systems using either classifiers or expert systems for signal interpretation. Some of the disadvantages of these two methodologies are discussed. Then, the main ideas and advantages of case-based reasoning are presented. Next, the LISSA system which implements the ideas of CBR for interpretation of EC data is described. Finally, results of tests on real inspection data are presented and discussed.