CBR systems for interpretation of eddy-current and ultrasonic NDT data

The interpretation of data from nondestructive testing (NDT) is a complex task and is typically performed by experienced operators. However, increasing numbers of inspections being performed and increasing requirements with respect to the reliability and repeatability of inspections result in a demand for automatic NDT data-interpretation systems. Reports of successful automatic interpretation systems mainly come from nuclear-power and aircraft industries. These systems typically use such techniques for data interpretation as: statistical classifiers, ANNs, or expect systems. These techniques, however, have had limited success for inspection in less well-defined environments, like petro-chemical installations or railroads. There, the large variability of the data, as well as the practical problems of economic justifiability of automating data interpretation become important. Case-based reasoning (CBR) is an AI methodology particularly suitable for NDT data interpretation in such environments. CBR systems are characterized by lower construction costs, reliability, adaptability, ease of use, and ease of maintenance. This paper presents two CBR systems which have been developed for eddy-current inspection and for ultrasonic rail inspection. The design of both systems as well as the issues encountered in the development of two prototypes are discussed. The evaluation of the performance of the systems was positive. The system for ultrasonic rail inspection is currently under evaluation by the Dutch Railway's.