Automatic Detection of High Temperature Hydrogen Attack Defects from Ultrasonic A-scan Signals.

Abstract

Successful application of the rich collection of classification algorithms to nondestructive testing signals depends heavily on the availability of adequate and representative sets of training examples, whose acquisition can often be very expensive and time consuming. In this paper, an out-of-service pressure vessel known to have lots of high temperature hydrogen attach (HTHA) defects is used to develop in a cost effective manner a database of ultrasonic A-scan signals. To test how adequate and representative these sets of A-scan signals are, a basic feature extraction method, coupled with a primitive classifier is shown to distinguish accurately the hydrogen attack from geometrically similar defects.