Hybrid knowledge-based architecture for building an intelligent nondestructive signal inspection system

The paper deals with the methods of implementing an intelligent inspection system for monitoring the health of any device or material using a nondestructive signal. A hybrid knowledge representation and processing architecture is proposed as a solution to problems encountered in the processes of the nondestructive inspection. In the proposed system, to modularize knowledge elicitation and to make knowledge processing reliable, the task of inspection is delegated to two subsystems each of which has a proper knowledge processing scheme to match the properties of its own task. The front-end subsystem which detects the signal patterns (events) for any harmful flaw is built by integrating a fuzzified syntactic pattern recognition concept and a neural network concept. The back-end subsystem which evaluates the characteristics of the events is based upon object-oriented rule base system concepts. The methods of integrating the approaches in the proposed architecture are also proposed in the paper. The proposed architecture is verified by developing and evaluating a prototype which automatically interprets eddy current Lissajous signals to inspect the state of tubes used in nuclear power plants.