Application of Decision trees for the identification of weld central line in austenitic stainless steel weld joints

Austenitic stainless steels (ASS) are preferred in chemicaland nuclear industries mainly due to their high corrosion resistance and unique high temperature creep properties. Austenitic stainless steel welding isan integral part of the Indian nuclear components and ultrasonic non-destructive testing technique (NDT) plays a major role in testing the integrity of the weld joints. The concept of manual ultrasonic testing (UT) of defects/flaws/discontinuities has now been replaced by computerization, automation and mechanization concepts. Remote ultrasonic NDT inspection assumes great dimension in the industrial system and in particular testing of pressure vessels made of several weld joints. Identification of weld centre line is very importantis very important in flaw evaluation in the weld jointsparticularly while carrying out remote ultrasonic testingof pressure vessels. Recently, successful attempts are being made in applying machine learning techniques for accurate flaw detection, sizing and location of weld joints. In this work, a 42 mm thick single “V” butt weld joint was fabricated and A-scan ultrasonic signals (time domain signals) were acquired at the weld centre and across the weld joint at the 5 mm distance interval and stored for further analysis using Decision tree algorithm. Critically refracted longitudinal (Lcr) wave probe at 2 MHz was used for this purpose. Decision tree algorithm which is an artificial Intelligence technique, classified under supervised machine learning algorithms, was used for training theacquired A-scan data and to reliably identify the centre line in the weld region for the purpose finding flaw location during remote ultrasonic testing. The developed procedure/ technique is first of its kind, simple to use and straight forward and useful for identifying the weld centre line and for accurate flaw location in the weld regions during ultrasonic testing of ASS weld joints.