A review of artificial intelligence applied to ultrasonic defect evaluation

The main objective in ultrasonic defect evaluation is to locate and classify suspect flaw indications quickly and accurately. Since the volume of data to be assessed can be very large, traditional forms of defect evaluation involving a skilled human interpreter are often unsuitable. The progress in the automated evaluation of ultrasonic data has been considerable in recent years and this paper outlines some of the approaches adopted in this area. Traditional pattern recognition techniques and the currently popular neural network approaches have been widely employed to process feature sets, extracted from A-scan signals. Knowledge-based system techniques, although not so widespread, are also considered. A number of authors have taken the approach that such AI techniques should be embedded in an integrated software framework for defect evaluation, and this is also discussed