Acoustic emission (AE) techniques are increasingly employed in structural health monitoring (SHM) and non-destructive testing (NDT) to assess the integrity of engineering structures. AE sources are typically associated with deformation and damage mechanisms such as cracking, dislocations, corrosion, inclusions, and delamination. Their localization enables real-time monitoring and supports timely, targeted maintenance to ensure structural safety.

AE-based localization relies on a network of spatially distributed sensors that record AE signals. The localization accuracy depends critically on the information captured by these sensors, which in turn is governed by their spatial placement. In practical scenarios, deploying sensors across an entire structure is infeasible, making optimal sensor placement essential. The challenge is to determine a sensor configuration that yields the most informative signals for accurate localization under practical constraints.

In this paper, we propose a Bayesian optimal sensor placement strategy to determine the optimal locations of sensor clusters that yield the most accurate AE source localization in isotropic plates with unknown properties. Each cluster consists of three sensors arranged in a right-angled triangular configuration. The source location parameters and their uncertainties are estimated using a Bayesian optimal design framework for parameter estimation. Although AE source locations are typically unknown, critical areas where damage is likely to initiate — referred to as hot-spots — are often known. To localize AE sources within these hot-spot regions, at least two sensor clusters are required.

The proposed Bayesian framework was employed to identify both optimal and worst cluster placements for different hot-spot regions using configurations of 2, 3, and 4 sensor clusters. The Bayesian strategy requires solving an optimization problem; in this study, five different optimization algorithms were employed and compared in terms of the resulting cluster locations and computational efficiency.

To validate the approach, experiments were conducted on an aluminum plate. The best and worst cluster locations predicted by the Bayesian design, along with an arbitrary cluster configuration, were positioned on the plate, and AE signals were generated at 10 distinct locations using pencil lead breaks. It was observed that the localization error was smallest for the optimal cluster placements and largest for the worst placements, with the arbitrary configuration yielding intermediate accuracy, as shown in the figure.

The proposed Bayesian optimal sensor placement strategy for AE source localization is directly applicable to practical scenarios across mechanical, aerospace, space, and other engineering domains. Since AE-based monitoring of metallic plate-like structures (e.g., aircraft fuselages, bridges, wings) commonly relies on pre-defined hot-spot regions rather than known defect locations, the presented methodology enables sensor networks to be deployed in a manner that maximizes localization accuracy and thereby enhances structural safety.

Acoustic emission (AE) source localization is a key technique in structural health monitoring (SHM) of aerospace structures, including components for space applications. AE enables real-time detection and localization of damage events such as cracking, delamination, or material degradation, often before they propagate to critical failure. For space-bound structures, early damage detection is particularly vital due to extreme operational conditions, high safety requirements, and the prohibitive cost of repair or replacement after launch. Unlike other non-destructive evaluation methods, AE allows continuous, in-situ monitoring under operational loads, providing immediate feedback on structural integrity and supporting predictive maintenance and risk mitigation strategies critical to mission success.

Classical AE localization approaches, such as time-of-arrival (TOA) triangulation, typically require prior knowledge of wave velocity and rely on assumptions regarding structural symmetry. These requirements may limit their applicability in real spacecraft components, where material properties, structural layout, and directional stiffness are often unknown or spatially varying, potentially reducing localization accuracy. In this work, we present a comparative study between (i) a classical TOA-based localization method implemented in the Vallen Systeme GmbH AE system, and (ii) a wavefront-shape based localization approach that does not require knowledge of wave speed, material properties, or orientation of the axes of symmetry. The approaches were employed to localize AE sources on a cylindrical spacecraft structure intended for space applications, as shown in the figure.

Experimental AE signals were generated using pencil lead breaks at multiple locations on the spacecraft structure. Two sensor configurations were considered: a conventional spatial arrangement for classical localization, and L-shaped sensor clusters for the wavefront-shape based method. Each L-shaped cluster consists of three sensors arranged at right angles, enabling the extraction of time-difference-of-arrival (TDOA) measurements and, consequently, the angle of arrival of the AE signals. The L-shaped cluster provides directional sensitivity, allowing the wavefront-shape method to infer the source location based on local wavefront curvature and propagation characteristics, independent of material properties or structural assumptions.

The study focuses on comparing both methods in terms of source localization accuracy across multiple excitation points. The wavefront-shape based method assumes that the AE-generated wavefront is elliptical and solves an optimization problem using only the angle of wave arrival (derived from the TDOA) and the positions of the L-shaped sensor clusters as inputs. This enables source localization without prior calibration or knowledge of wave propagation characteristics. In contrast, the classical Vallen-based approach relies on TOA measurements and user-provided wave velocity inputs.

By benchmarking classical and wavefront-based localization on the same experimental dataset, this work aims to quantify the relative strengths and limitations of each method for realistic rocket SHM scenarios. The findings are intended to inform the design of sensor configurations and provide guidance on selecting appropriate AE localization strategies for space-bound structures, where material properties and structural layouts are uncertain or only partially known. The results are expected to support more reliable in-situ monitoring of critical spacecraft components, contributing to enhanced safety and mission success.

In high-altitude open-pit mining areas, freeze-thaw (F-T) cycles readily induce complex thermo-hydro-mechanical (THM) interactions in host rock masses. The long-term effects of F-T environments and engineering activities pose significant threats to the stability of mining rock masses and infrastructure. Existing monitoring and warning indicators have not achieved deep integration with artificial intelligence technologies, and the implementation of human-machine collaborative engineering systems remains limited. In this study, we calculated the dispersion of Mel-frequency cepstral coefficients (MFCCD) extracted from acoustic emission (AE) waveform signals during the progressive fracture evolution of sandstone under F-T cycles and seepage-stress coupling conditions. An improved sparrow search algorithm (ISSA) was employed for automatic hyperparameter optimization of the temporal convolutional network (TCN) and bidirectional gated recurrent unit (BiGRU) models. The superior performance of this hybrid model validates the effectiveness of the proposed methodology. The results demonstrate that, compared with conventional MFCCs, the sliding window statistical-based MFCCD approach not only enhances prediction accuracy in both optimized and benchmark models but also reduces energy consumption by over 83%. Through quantitative analysis using SHapley Additive exPlanations (SHAP), the underlying factors influencing the output of the ISSA-TCN-BiGRU model were ranked in the following order of significance: confining pressure < osmotic pressure < F-T cycles. This research provides novel mechanistic insights and design frameworks for long-term prevention and control in THM coupling and cold-region rock engineering, thereby facilitating the advanced integration and synergistic development of artificial intelligence and rock mechanics.

As wind energy capacity increases and moves further offshore, effective structural health monitoring of wind turbine blades (WTBs) becomes critical to mitigate turbine failures. Acoustic Emission (AE) monitoring offers a promising solution, in which piezoelectric sensors detect stress waves generated by developing defects. However, the successful deployment of AE systems requires knowledge of the material’s wave attenuation properties to accurately determine sensor coverage. Current literature provides limited data on the direction and frequency dependent attenuation of anisotropic Glass Fibre Reinforced Plastic (GFRP) composites commonly used in WTBs.

This study investigates attenuation properties of unidirectional and triaxial ([0/±45]) GFRP composites that are commonly used for WTB applications. The composites used in this study are excited using 10mm diameter piezoelectric disc over the 100–500 kHz frequency range, producing Lamb waves which were measured using AE sensors positioned at five propagation angles (0° - 90°) and distances (50–200 mm) from the piezoelectric disc. Attenuation coefficients were calculated by comparing signal amplitudes across positions and frequencies, with both fundamental Lamb modes considered.

The study demonstrated a strong dependency of frequency and propagation direction on AE signal attenuation. In the unidirectional composite, S0 attenuation along the 0° fibre direction increased from 2.7 to 9.9 Nepers/m between 100–500 kHz, while A0 rose from 6.8 to 16.2 Nepers/m. Off-fibre propagation (90°) produced much higher values in the region of 30.8 to 41.4 Nepers/m for the S0 and A0 mode respectively, and even a 20° misalignment caused marked increases, confirming strong anisotropy. The triaxial composite also exhibited directional-dependant attenuation, though more complex when compared to unidirectional. Attenuation along the fibre aligned directions ranged from 4.4 to 16.5 Nepers/m for S0 and from 8.9 to 21 Nepers/m for A0, with off-fibre propagation showing further increases upwards of 43.2 Nepers/m. As with the unidirectional composite, attenuation grew with frequency, and A0 consistently exceeded S0, reflecting its flexural character.

To assess the practical implications for AE monitoring, the attenuation coefficients from this study were compared with typical matrix cracking source amplitudes reported in the literature (approximately 62.1 dB). Assuming an AE sensor detection threshold of 40 dB, this corresponds to an allowable amplitude loss of 22.1 dB along the propagation path. Using this threshold, the measured attenuation coefficients yield an estimated detection range of approximately 0.7m for on-fibre propagation directions, which reduced to just 0.2m when considering off-fibre propagation directions. This AE sensor detection range considers both the unidirectional and triaxial composites, broadly agrees with values reported in other experimental studies and also provides new insight into frequency and direction dependent attenuation of unidirectional and triaxial GFRP composites.

In conclusion, this study demonstrates that GFRP attenuation can constrain AE coverage in WTBs, and whilst AE monitoring remains viable, effective deployment requires careful sensor placement with expected coverage distance of 0.2 – 0.7m. This approach is being extended to biaxial layups and sandwich core composites, enabling more realistic sensor network designs for industrial full-scale blades.

Vibration-based Structural Health Monitoring (SHM) techniques are effective for identifying global dynamic behavior, but local damage detection and source localization remain challenging, particularly when damage initiates at small scales or in confined joint regions. In steel structures, critical deterioration mechanisms often develop at welded or mechanically connected details, where modal vibration signatures can remain largely unaltered during early damage stages, reducing their diagnostic sensitivity. This limitation motivates the integration of monitoring approaches that are driven directly by material-level response.

Acoustic Emissions (AE) provide a non-destructive and high-sensitivity monitoring strategy generated by transient elastic waves released from a material when mechanically stressed. AE events originate at the damage source itself, enabling a direct assessment of local condition variations in structural details and joints. AE monitoring is therefore particularly suited for steel connections, where deterioration may remain undetected by global dynamic metrics.

Current AE monitoring approaches often rely on elementary signal descriptors—including peak amplitude and waveform duration—which are widely reported in the literature as indicators of critical structural states. On the other hand, dominant frequency tracking can be ambiguous to extract reliably due to noise sensitivity, mode mixing, and signal non-stationarity. To address this challenge, the present work explores the use of Cepstral Coefficients (CC) computed from AE waveforms as a stable and straightforward feature for signal characterization. Unlike spectral peaks, CC are directly and uniquely derived from the input signal through a computationally lightweight procedure. These characteristics make CC attractive for rapid laboratory assessment and future SHM integration.

In this context, an experimental study was carried out on welded, full-penetration steel joint specimens tested at University of Trento. The specimens, instrumented with AE sensors, were tested at the University of Trento under controlled low-cycle loading protocols inducing yielding damage. Based on the controlled load history, a comprehensive dataset of AE signals was collected from both undamaged and damaged specimens to assess how local deterioration alters CC features, which were extracted from raw AE waveforms and statistically compared between undamaged and damaged conditions. The analysis highlights differences in CC distributions, showing potential in discriminating between distinct structural damage states in steel joints without requiring complex frequency identification. The results support the CC-based characterization of AE signals as a complementary monitoring layer for SHM systems targeting critical steel joint regions.

Additive remanufacturing has shown great potential in in-situ repair of in-service components and restoration of end-of-life components. However, the additive remanufacturing process often introduces defects such as residual stress, deformation, and heterogeneous microstructural distributions, which significantly affect fatigue performance and pose challenges for health monitoring and fatigue performance prediction during service.

To address fatigue crack propagation monitoring in additively remanufactured components, an online crack length estimation method based on acoustic emission (AE) signals is first developed. AE signals are high-frequency, transient, and noise-contaminated, making direct analysis difficult. In this study, low-dimensional latent variables are extracted from raw AE signals using a one-dimensional convolutional autoencoder. A data-driven causal discovery algorithm is then employed to construct a causal network that characterizes the cause–effect relationships among these latent variables. Based on the constructed causal network, a graph attention network is used to aggregate information from parent nodes to form node embeddings, which are subsequently mapped to crack length labels through a fully connected layer. To address data scarcity in online monitoring scenarios, transfer learning strategies are further introduced by transferring models trained on other specimens. Experimental results show that the proposed crack length estimation method achieves good prediction performance and transferability across different wire and arc additive manufacturing (WAAM) specimens.

In additively repaired components, distinct morphological variations exist among different regions. The WAAM region generally exhibits coarse clustered structures, whereas both the heat-affected zone (HAZ) and the base metal (BM) display banded morphologies with notable differences in grain size, orientation, and volume fraction. Such microstructural heterogeneity leads to pronounced distribution shifts in AE responses across regions, making it difficult for a model to learn a unified discriminative basis for crack length prediction.

Motivated by this limitation, a further study is conducted to establish a microstructural recognition approach based on AE. An in-situ fatigue testing system integrating scanning electron microscopy imaging and AE monitoring is constructed to synchronously record crack path microstructures and the corresponding AE signals. Based on these data, a recognition method combining supervised contrastive learning and domain adversarial learning is adopted to extract weak microstructural responses under specimen variability and environmental noise. The contrastive loss improves compactness within a class and separation between classes, while an adversarial branch with a gradient reversal layer reduces discrepancies across specimens. Silhouette coefficients are introduced to dynamically adjust the relative weights of the two losses. The model is trained on WAAM and HAZ data and evaluated in a zero-shot setting on BM specimens, achieving an accuracy of 76.38%, with predictions showing good agreement with scanning electron microscopy observations. The results demonstrate the feasibility of using AE signals to estimate fatigue crack propagation and recognize microstructural characteristics, providing a non-invasive solution for monitoring fatigue behaviour of additively remanufactured components under service conditions.