Dodger, a Diagnostic Expert System for the Evaluation of Nondestructive Test Data

Arthur J. Levy, Jane E. Oppenlander, David M. Brudnoy, James M. Englund, Kent C. Loomis and Arnold M. Barsky, General Electric Company

Identification and characterization of indications in eddy current (ET) signals can be highly subjective in nature, with varying diagnoses made by different analysts or by a single analyst at different times. Consistent analyses of an indication over time are required in order to accurately assess trends in material condition. A rule-based expert system, with a well designed set of interpretation guidelines, can provide the consistent and repeatable analysis that is desired. An expert system was developed that analyzes eddy current signals allowing interactive or unattended operation, or a mixture of the two. Measurements are derived from the data using automatic machine recognition of ET impedance plane patterns. Uncertain and conflicting measurements are treated in a rigorous, probabilistic manner using the Dempster-Shafer theory of evidence. The expert system identifies the nature of an indication and the confidence in that diagnosis. The system is also able to automatically reference past measurements of the same indication and analyze them using the same criteria. Finally, it determines the trend in the indication and allows the analyst to make an informed decision about its severity. This paper describes the expert system (Dodger) and its process of analyzing eddy current signals.


This page is copyrighted by AAAI. All rights reserved. Your use of this site constitutes acceptance of all of AAAI's terms and conditions and privacy policy.