Structural Health Monitoring (SHM) in aerospace applications requires accurate detection and characterization of impact events to ensure operational safety and reduce maintenance costs. This work presents an experimental framework for developing physically informed machine learning models that leverage embedded PVDF-TrFE transducers for acoustic emission (AE) monitoring. Representative composite coupons were instrumented with PVDF-TrFE sensors and subjected to controlled impact tests at varying energy levels to generate AE datasets under pristine and damaged conditions. The experimental campaign was designed to capture key features associated with impact occurrence, localization, and energy quantification in critical regions of aerostructures.
To enhance model robustness and scalability, experimental data were fused with synthetic datasets derived from high-fidelity numerical simulations, enabling the integration of physical constraints into the learning process. This hybrid approach aims to improve generalization across different geometries and loading scenarios, paving the way for future deployment on full-scale aerospace components. Preliminary results indicate that the proposed methodology can identify impact events and estimate their severity with promising accuracy, while ongoing work focuses on refining localization and energy quantification metrics.
The study demonstrates the potential of combining advanced sensing technologies with physics-informed data-driven models for next-generation SHM systems in aerospace applications.

Gas transport and storage structures are critical infrastructure that require continuous monitoring throughout their lifetime. One employed method involves using the acoustic waves generated when the structure is being damaged to characterize and locate the associated defects. Conventional localization methods for acoustic emission rely on the identification of time of arrivals between the acoustic emission and an array of sensors bonded on the structure. They typically require a relatively large number of sensors and a theoretical or numerical propagation model of an unbounded structure to estimate the position of the acoustic emission. To overcome these limitations, we propose to apply Time Reversal (TR) method to localize impacts on two industrial structures: a thermoplastic composite gas tank and an aluminum pipe. To apply this localization method, a model of propagation is necessary. For the tank, this model is purely experimental. It corresponds to a set of transfer functions between a grid of points previously defined on the monitored area and a sparse network composed of three piezoelectric copolymer transducers made of P(VDF-TrFE). These transfer functions will integrate information about all the reverberated field in the tank, allowing the use of few transducers and will be used at low frequency to have a modal behavior of the structure and so improve the robustness of the model to changes in environmental conditions. A different approach is used for the aluminum pipe, taking advantage of the waveguide properties of this kind of structure. Based on a set of dispersion curves and modal displacement vectors over a certain frequency band the Normal Mode Expansion method is used to construct the propagation model and apply TR using flexible linear transducer arrays forming rings when glued around the pipe. Localization of impacts and pencil lead breaks are successfully carried out on the two studied structures. The performances of the whole setups are evaluated based on multiple experiments and quantitative indicators.

Aircraft structures are susceptible to low-velocity foreign object impacts, which may occur during manufacturing, maintenance, and in service. In metallic structures, damage due to impacts appears as dents and tears. In composite structures, impacts can create internal damage that may not be visible on the outer impacted surfaces. Common ways to detect such impacts are based on flight or ground crew observations and reports leading to close examinations of structures using non-destructive evaluation (NDE) techniques. If such damage is not detected and repaired, it may worsen, potentially compromising the structural integrity of the aircraft. Thus, it is essential to promptly identify any signs of impact damage to facilitate timely maintenance actions. Therefore, the aim of this research is to create methods that employ Structural Health Monitoring (SHM) techniques to automatically detect and evaluate foreign object impact.

In this experiment, a cut-out measuring 51.25 by 28.5 inches [130.2 by 72.4 cm], from a vertical stabilizer of an A320-211 aircraft was used. The vertical stabilizer consists of an outer skin reinforced by ribs and stringers, all of which are made from carbon/epoxy prepreg. To detect and measure the impact events, three different sets of sensors were used. The sensors used were: dragonfly dynamic strain measuring sensors from Worm Sensing, acoustic emission (AE) sensors from Vallen, and Lead Zirconate Titanate (PZT) sensors from Acellent. To acquire the impact event data, the dragonfly and PZT sensors were directly connected to a digital oscilloscope (Picoscope 4824) without any amplification or filtering. For the AE sensors, the Vallen AMSY-6 data acquisition system, along with Vallen’s AE software, was used to collect the data. A multi-use tap-hammer with force feedback capability was used to inflict the impact.

Features extracted from the input (tap-hammer signal) were compared to the output (sensor signal) to discern any relationships, with most signals exhibiting linear correlations between input and output, indicating that regression techniques could be used to predict previously unseen signals. A Genetic Algorithm (GA), employing the gamma value as the fitness function, was utilized to identify the most effective subset of features. Various machine learning models were then applied to predict the position and energy level of the impact using the optimal features extracted by the GA, with XGBoost delivering the best results for position prediction and Multi-layer Perceptron (MLP) excelling in energy prediction.

Structures operating in complex marine environments are exposed to potential impacts from various external sources. Such events can result in significant structural or environmental consequences and may interrupt normal operations. Therefore, accurate impact identification is essential to ensure safety, minimize environmental risks, protect assets, and support informed decisions in structural integrity assessment.

Impacts release part of their energy in the form of stress waves that in thin-walled structures such as ship hulls, propagate long distances as Guided Waves (GWs). The time reversibility and spatial reciprocity properties of the GUW allow the use of Time Reversal (TR) process of the recorded wave signals to localize impacts.

The objective of this paper is to investigate impact localization and force reconstruction on steel plates and stiffened panels by means of monitoring impact induced GWs. To achieve this, an analytical framework has been developed and experimentally evaluated. The experimental part consists of generation of controlled impacts with an instrumented hammer. The stress waves emitted during impact were subsequently measured using surface mounted piezoelectric sensors with a resonant frequency of 60 kHz. Impact localization was performed through a virtual TR process which was implemented in the frequency domain using an analytical propagation formulation that accounts for dispersion and wave amplitude decay due to geometric spreading. Force reconstruction is achieved using a parametric half-sine model combined with a transfer function obtained from a single calibration impact.

The approach was validated through small-scale experiments on steel plates and stiffened plates measuring 400x400 mm². The average localization error for the plate ranged from 11 to 15 mm, while stiffened panel tests showed slightly higher errors in the order of 12 to 24 mm due to wave scattering at the stiffener. These results show that accurate localization is achievable. The reconstructed forces using the parametric model provide a good estimate of the applied impact force, with a mean relative error of approximately 24%.

The results demonstrate that the proposed framework can enable combined impact localization and force estimation in thin-walled structures, with and without the presence of structural stiffeners.

Ships are often subjected to impact loads, and it is difficult to measure the impact loads directly. It is feasible to invert the impact loads by structural responses, but there are relatively few studies on the inversion of the impact loads of ship panels. Taking the stiffened panels of ships as the research object, this paper established a time-history inversion method of impact load based on Green kernel function, adopted regularization method to deal with the ill-posedness of inversion problems, quantitatively analyzed the impact load inversion errors with displacement response and stress response as the input respectively, and further proposed the impact load location identification method based on Maxwell-Betti theorem. The interpolation function is used to improve the inversion accuracy of the impact load position. The results show that the identification error of impact load amplitude is less than 5%. The error of impact position is less than 7% when choosing appropriately the calibrated points. The conclusions in this paper will provide a reference for the identification of impact load of real ship.