ULTRASONIC ARRAY IMAGING FOR TWO-LAYERED MEDIA USING THE INTERFACIAL INTERPOLATION TECHNIQUE BASED ON MACHINE LEARNING

Abstract

Internal flaws in a structural component are reconstructed using the full-wave sampling and processing (FSAP) technique which utilizes the ultrasonic wave transmitted through a couplant. In the first step, the shape of the irregular interface is reconstructed by the conventional FSAP. Using the estimated interfacial shape, the path of the diffracted wave through the interface is calculated for the shape reconstruction of the internal flaw in the second step. In this study, the intelligent interpolation of the incomplete shape of the interface was carried out with the aid of the pix2pix that is a kind of generative adversarial network. The training data was made by massive simulated data in wave propagation analyses. The validation by experimental measurement showed that the interpolation of the interface by the pix2pix learning model could enhance the accuracy of the flaw reconstruction in the FSAP.