Data-Driven Physics-Based Reconstruction of Structural Defects Using Multi-Modes Signals

Abstract

As compared with traditional ultrasonic bulk waves testing methods, the ultrasonic guided waves technique has been used in nondestructive evaluation and health monitoring of large complex elastic waveguide structures such as aircraft wing plates, railway tracks and oil pipelines owing to the advantage of wave traveling along the axial length with low attenuation. Meanwhile, with the rapid developments of computing power (in particular, Graphic Processing Unit-GPU acceleration) and artificial intelligence technologies in image processing, natural language processing and other industries, using data-driven methods to achieve a quantitative reconstruction of structural defects with high-efficiency and high-precision has become a very hot research topic in the field of structural integrity and operational monitoring. In this paper, a novel data-driven defect reconstruction method, called Net-guide, has been proposed to realize the end-end mapping between the guided wave scattering signals and the profiles of structural defects by intelligent learning. It is worth noting that Net-guide has the ability to comprehensively utilize the multi-modes of guided waves signals to achieve high-precision reconstruction of defects. Throughout the numerical verification, the superiority of Net-guide has been demonstrated by successfully reconstructing complex defects with different shapes, as compared with the results by traditional knowledge-driven reconstruction approach. Results show that the proposed Net-guide can achieve high-precision reconstructions with a high level of efficiency and provide a valuable insight into the development of data-driven quantitative nondestructive testing methods.