Development of an intelligent system for ultrasonic flaw classification in weldments

Even though ultrasonic pattern recognition is considered as the most effective and promising approach to flaw classification in weldments, its application to the realistic field inspection is still very limited due to the crucial barriers in cost, time and reliability. To reduce such barriers, previously we have proposed an intelligent system approach that consisted of the following four ingredients: (1) a PC-based ultrasonic testing (PC-UT) system; (2) an effective invariant ultrasonic flaw classification algorithm; (3) an intelligent flaw classification software; and (4) a database with abundant experimental flaw signals. In the present work, for performing the ultrasonic flaw classification in weldments in a real-time fashion in many real word situations, we develop an intelligent system, which is called the ‘Intelligent Ultrasonic Evaluation System (IUES)’ by the integration of the above four ingredients into a single, unified system. In addition, for the improvement of classification accuracy of flaws, especially slag inclusions, we expand the feature set by adding new informative features, and demonstrate the enhanced performance of the IUES with flaw signals in the database constructed previously. And then, to take care of the increased redundancy in the feature set due to the addition of features, we also propose two efficient schemes for feature selection: the forward selection with trial and error, and the forward selection with criteria of the error probability and the linear correlation coefficients of individual features.