Ultrasonic Pattern Recognition Study of Intergranular Stress Corrosion Cracks Vs. Weld Crown Reflectors in SS Piping. Interim Report

摘要

Pattern recognition techniques for discriminating between geometrical and crack reflector signals obtained during ultrasonic inspection of weld zone in Type 304 austenitic stainless steel piping have been applied to one set of data. Seven welds from four different 4-in diameter pipe specimens containing intergranular stress corrosion cracking (supplied by the GE pipe laboratory through SwRI) were examined ultrasonically. Geometrical reflectors considered in this feasibility study were crown type reflectors only, since they were readily available in the pipe specimens. The ultrasonic inspection was conducted in a pulse-echo mode using a 1.5 MHz nominal center frequency, 3/8-in diameter transducer mounted on a plexiglass shoe with a 45 exp 0 refracted transverse wave. A version of the Southwest Research Institute pipe weld examination code was used for recording the data. The ultrasonic signals were digitized and stored for further analysis. One hundred fifty-five indications were recorded from seven different welds, four of which were examined from both sides, and three from only one side. The ultrasonic data were correlated with dye penetration tests and ultrasonic examination conducted by SwRI in order to obtain correct training information. The data naturally fell into two categories, cracks and crowns (geometric reflectors). A total of 107 crown indications and 40 intergranular stress corrosion cracking (IGSCC) indications were further analyzed. The pattern recognition analysis indicated that an IGSCC indication was discriminated from a crown indication in about 98% of the cases. The success of the pattern recognition algorithm employed in this study demonstrates the applicability of this technique for solving such important problems as discrimination between IGSCC and geometric reflectors in 304 stainless steel pipe welds. Additional work on other kinds of geometric reflectors is required to establish confidence level in reflector classification analysis. (ERA citation 04:022982)