Characterization of Lamb waves behavior in CFRP pseudo-damaged configurations for SHM applications

Kurzfassung

The structural framework of modern aircraft is subject to some of the most demanding engineering constraints, balancing the imperative of uncompromising safety with the need for high operational readiness. In pursuit of lighter, more efficient airframes, aerospace designers have turned extensively to carbon fiber-reinforced polymers (CFRPs), whose superior mechanical properties per unit weight have revolutionized structural design. However, this advancement comes at a cost: CFRPs do not fail like conventional metals. Their layered, anisotropic nature gives rise to subtle and often hidden damage mechanisms. Chief among these is barely visible impact damage (BVID)-a phenomenon where an apparently intact surface conceals internal disruption that may critically undermine the structural integrity over time. To meet the growing demand for continuous, reliable assessment of such complex materials, Structural Health Monitoring (SHM) has emerged as a frontier field, with guided ultrasonic waves-particularly Lamb waves-offering a non-intrusive route to damage interrogation. These waves, by virtue of their ability to travel long distances and remain sensitive to internal inhomogeneities, are well-suited for inspecting CFRP components. Yet a persistent obstacle hinders progress: the high cost and destructive nature of creating true damage scenarios, which are essential for validating SHM methodologies. As a response to this constraint, the idea of pseudo-damages has taken shape-engineered disturbances designed to mimic real structural defects while preserving the functionality of the material. This thesis investigates the interaction of Lamb waves throught such controlled, reversible damage proxies in order to assess the possibility to reliably emulate the ultrasonic signature of BVID, thereby enabling more scalable and cost-effective SHM development by validating pseudo-damages as faithful surrogates for real impact-induced damage. This is achieved within a controlled experimental framework in which Lamb wave acquisitions are systematically performed over a designed set of pseudo-damage configurations-varying material type, defect size, and adhesive technique-while maintaining consistent excitation and sensing conditions. The resulting dataset is then processed using two complementary approaches: classical inference methods, which rely on the explicit extraction of physics-based features from the measured signals and convolutional neural networks (CNNs), which take the minimally pre-processed acquisition data as input, learning discriminative representations of the pseudo-damage response directly from the measurements without manual feature engineering.