A Bayesian approach for sparse flaw detection from noisy signals for ultrasonic NDT

Ultrasonic pulse-echo methods for flaw detection have been widely employed as an effective strategy for nondestructive evaluation, and flaw detection plays an important role due to its ability to detect localized damage in structures. In practice, flaw damage typically occurs in a few areas in the material, resulting in only a few echoes that exist in a received signal, which motivates us to detect flaws using sparse representation methods. In this study, the noisy signal is modelled by a linear combination of modulated Gaussian pulses, which form an over-complete dictionary. The over-complete dictionary is designed such that the sparseness of the representation is expected. A robust sparse Bayesian learning framework is employed with the goal of enforcing model sparseness and reducing the source of ill-conditioning in the inversion problem for flaw detection. Useful information, including the range of frequency and bandwidth parameters of the flaw echoes, is also estimated. Based on this information, we propose a post-processing scheme for structure noise elimination and flaw detection. The capability of the proposed method is quantitatively evaluated by simulation studies and is further validated by the experimental data.