Robust Guided Wave Tomography Method for Large and Irregular Defects

Abstract

Finding a fast, robust way to quantitatively measuring the remaining wall thickness of complex structures when multiple defects exist is one of the leading challenges in Nondestructive Testing (NDT). Traditional inversion algorithms like ray tomography and full waveform inversion (FWI) suffered from problems like convergence, limited resolution and slow speed. Diffraction tomography (DT) has speed advantage over the preceding methods and its resolution can be further amplified by integrating with other methods like bent-ray tomography and iteration. However, DT can only detect shallow and small defects. Compared with those methods, convolutional neural network (CNN) opens a new way for quantitative defect imaging, as with pre-trained data it can achieve significant speed and resolution than the traditional methods. In this paper, we investigated the performance of CNN in imaging multiple defects and the inversion results show that when dealing with multiple defects with complex shape on a plate-like structure, CNN can achieve better resolution than other methods with maximum errors below 0.54mm in most regions. This research provides the experimental guidance for future study in finding the possible ways to improve the resolution of the algorithms.