Weld Flaw characterization in Ultrasonic Testing from Discrete Walsh Hadamard Transform Co-efficient

Abstract

Ultrasonic Testing is one of the most commonly used Non-Destructive Testing Techniques for assessing the quality of welds. It is an indirect technique, in which the signals are analyzed for flaw characterization. Though various methods are cited in literature, probability of detection and sensitivity of detection are yet to be improved. Hence in this work, a novel flaw characterization technique is attempted by decomposing the ultrasonic signals into Walsh Hadamard transformed co-efficient and by determining statistical parameters. These parameters are then used for characterizing the defects.  In  order  to  improve  the  performance,  a  pseudorandom  signal  is  also  added  to  the  ultrasonic  signal  and  its performance is measured. It is found that Probability of Detection for the processed signals is better than that of unprocessed signals.  Also  these  features  are  used  to  develop  a  Back  Propagation  Network  based  classifier  for  ultrasonic  signals. Probability of Detection has significant increase using BPN based characterization.