Use of Deep Learning Techniques for Damage Localization in Aeronautical Composite Structures

Damage localization is one of the most challenging topics within Structural Health Monitoring (SHM) in aeronautics, especially when the structure is manufactured out of carbon fiber-reinforced composite materials. Using ultrasonic guided waves (particularly Lamb waves), generated and recorded with piezoelectric transducers, is also challenging in this type of material. Otherwise, traditional methods used for this task are subjected to physics-based knowledge of the problem, such as damage imaging algorithms like delay-and-sum and RAPID. This paper presents an entirely data-driven approach, based on the ability of Deep Learning (DL) techniques (particularly those based on Convolutional Neural Networks – CNNs –) to extract features of interest for damage imaging from a pre-dataset. In this work, the selected feature to be estimated is the normal distance from the propagation path of the guided wave to a simulated damage, which allows, in combination with an especially designed positioning algorithm, to locate with high accuracy defects, even in different positions than the used for the training of the network (a fixed grid of points over the analysis zone). This paper presents the application of the method to a real composite material specimen, as well as the recorded results obtained from additional datasets recorded with the simulated damage (a piece of blu-tack) attached to different random positions other than those of the training grid.