Development of an Expert System for Ultrasonic Flaw Classification

The complete characterization of a flaw requires information about the flaw type (crack, void, inclusion, etc.), flaw size, and orientation. Here we are only concerned with the determination of the flaw type so that the appropriate sizing algorithms can be chosen. This type of classification problem using ultrasonic waves is very suitable for employing the tools and techniques of artificial intelligence [1,2]. Adaptive learning methods, for example, have in the past been employed to train a flaw classification module so that it can distinguish between cracks and volumetric flaws [3]. Some of the limitations of this approach, however, have been due to the empirical nature of the features used for classification and the difficulty of understanding and adjusting the decision-making process when errors occur.