Ultrasonic defect detection in noisy signals by a nonconvex sparse regularization approach

Abstract

Ultrasonic testing has been an important tool for the detection of hidden defects in materials yet its effectiveness is usually compromised by the noise originated from both the testing system and the material being evaluated. Motivated by the prior knowledge that only a small number of defects or reflectors exist within the tested material thus the measured signal can be viewed as a linear combination of a few echoes, a signal processing method is proposed in this study where a nonconvex sparse regularization method based on lp-norm (0 < p < 1) penalty is employed to suppress noises in ultrasonic NDT signals thus enhancing the effectiveness and accuracy of flaw detection. Based on a specially designed over-complete Gabor dictionary, the nonconvex sparse regularization method is introduced to sparsely represent the noisy signal. After signal representation, noise is suppressed by a pruning operator, facilitating the identification and reconstruction of flaw-reflected echoes. The performance of the proposed method is quantitatively evaluated and compared with competing algorithms using simulated noisy ultrasonic signals in a statistical manner. Experimental results are also presented to demonstrate the effectiveness of our method.