A Tightly Coopled Approach to Fuzzy Syntatic Parsing and Neural Networks for Event-Synchronous Signal Inspection

Abstract

The article presents a hybrid knowledge-based approach to solve the complex problems encountered in a nondestructive signal inspection domain. We propose to combine syntactic pattern recognition and neural network concepts to extract and classify event patterns from the time-varying signals. The proposed method consists of the following steps: first, the target signals are transformed into fuzzy symbols. Second, the fuzzy symbol strings are monitored and event patterns are captured by a syntactic parser. Third, a neural network classifier synchronously decides whether the event pattern does or does not have harmful flaw characteristics. A special knowledge representation and processing architecture is designed and implemented to integrate these steps into a stand-alone process. The proposed method enables implementing an event-synchronous knowledge-based system to inspect signals. A prototype of the proposed method has been implemented for monitoring the health of steam generator tubes using eddy current signals in the nuclear power plant and evaluating it based on experiments using field data. (C) 1995 John Wiley and Sons, Inc.