Structural Health Monitoring

The diagnosis of bearings is greatly difficult due to strong background noise and complex transmission paths. So, we designed an aviation gas turbine engine bearings fault diagnosis method. It is based on the combination of Wavelet packet decomposition (WPD) and correlation coefficient-energy ratio-kurtosis criterion judgments with AO-PNN, a probabilistic neural network (PNN) optimized by introducing the Aquila optimizer (AO). The vibration signal is firstly decomposed by WPD and reconstructed by screening Node components by correlation coefficient-energy ratio-kurtosis criterion judgment. The overlapping segmentation of reconstructed signals and the multi-scale permutation entropy of each sample are calculated as the feature vector and reduced by Kernel principal component analysis. AO-PNN is used for fault classification of fault patterns. The experimental results show that this method can effectively eliminate the interference of background noise as to improve the accuracy of fault diagnosis. Compared with the non-optimized PNN, the accuracy is improved by 11.25%.

Modal frequencies are widely utilized as damage-sensitive features in vibration-based bridge damage warning. However, nonlinear and nonstationary modal variabilities induced by environmental fluctuations, compounded by improper selection of modal orders, may obscure damage sensitivity and degrade warning accuracy. Therefore, a subdomain cointegration (Sub-CI) method for environment-tolerant and early damage warning of bridges is proposed. The proposed method initially partitions the multimodal training data into local subsets that capture local variation information using the Dirichlet process K-means cluster. Subsequently, an implicit cointegration model with immunity to environmental effects within each cluster is established. Then, two types of damage indexes are considered separately, namely the robust Mahalanobis squared distance for multivariate cointegration and the Johnson transformation-based one-dimensional residual for bivariate cointegration. After that, the dynamic thresholds are derived from each Sub-CI model to assess the health status of in-service bridges. Results demonstrate that the proposed method mitigates modal variabilities induced by latent periodic temperature changes and overcomes larger damage prediction errors and higher false alarm rates in comparison to classical linear cointegration.

The demand for advanced monitoring and fault diagnosis technologies for critical mechanical components is growing rapidly. Early detection of rolling bearing faults is essential for preventing performance degradation, unplanned downtime, and safety risks. This article presents a novel fault diagnosis method that leverages digital twin technology and transfer learning to address the limitations of existing approaches in terms of data dependency and cross-domain effectiveness. Initially, a precise digital twin model is developed using finite element analysis to accurately simulate bearing dynamics under various operating conditions, generating extensive simulation data. These data compensate for the scarcity of fault data and are valuable for training diagnostic models. To reduce the noise level in real-world data, the snow ablation optimizer algorithm is employed to optimize variational mode decomposition for noise reduction. Subsequently, transfer learning techniques are utilized to treat the simulation data as the source domain and the actual vibration signals as the target domain, enabling domain-adaptive transfer learning. This approach facilitates cross-domain feature alignment and knowledge transfer, further optimized through adversarial loss and the maximum kernel mean discrepancy metric. Moreover, a deep learning model that combines residual convolutional neural networks with a Transformer is developed, significantly enhancing feature extraction and classification accuracy. Experimental validation conducted on the XJTU-SY dataset demonstrates that the proposed diagnostic method exhibits superior diagnostic performance under small sample conditions, outperforming existing diagnostic methods.

This article introduces a novel self-supervised learning framework for structural health monitoring (SHM) that utilizes power spectral density (PSD) data to detect structural damage. The approach leverages PSD data to monitor structural health comprehensively, without relying on engineered features or domain-specific expertise. Designed to accommodate data from multiple similar structures, the framework trains a model using data from a population of similar structures, rather than focusing on domain-specific adaptation. This method effectively handles environmental variability and the complexity inherent in high-dimensional PSD data. A key innovation in this work is the introduction of a method for evaluating and tuning the model without needing actual damage data. The framework follows a two-stage training and evaluation process: in the first stage, a neural network is trained to classify PSD data based on the structures from which the data originates. During inference, the model’s classification confidence serves as an anomaly index, which forms the foundation of the proposed damage detection system (DDS). In the second stage, virtual anomalies (VAs) are introduced as data augmentation, modifying healthy data to evaluate the model’s ability to detect deviations that suggest potential damage. A key finding is that the DDS’s ability to detect VAs correlates with its ability to detect real structural damage, allowing for future validation without real damage data. Heatmap visualizations derived from VAs highlight the frequency regions where the model is most sensitive. The DDS is applicable in scenarios where a single accelerometer is used on similar structures (e.g., wind turbines, transmission towers) or multiple accelerometers are deployed on a single structure. The framework’s effectiveness is demonstrated in two case studies: a simulated population of 8-degree-of-freedom mechanical systems and real-world data from a transmission tower instrumented with four accelerometers under controlled damage conditions.

In environments with strong noise and varying loads, accurate fault diagnosis remains a significant challenge, especially in industrial applications where early fault detection is crucial for ensuring equipment reliability and operational safety. Traditional convolutional neural networks (CNNs) and graph convolutional networks (GCNs) often perform poorly under such complex conditions. This paper proposes a novel multibranch graph convolutional neural network (MBGCN) integrated with Gaussian filters to diagnose rotating machinery faults in noisy and variable load environments. The framework combines CNNs, Gaussian filter branches, and denoising autoencoders for efficient feature extraction and fusion. The enhanced multi-receptive field GCN captures information from different receptive fields optimizes node representations, and effectively mitigates noise interference. Comparative experimental results demonstrate that MBGCN outperforms traditional methods in fault diagnosis under complex industrial conditions, highlighting its potential real-world applications in predictive maintenance and intelligent monitoring systems.

A characterization technique is proposed for polymer coating structures immersed in water based on the ultrasonic interference and critical attenuation phenomenon. When a coating layer is modeled as a linear viscoelastic material and is subjected to the normal incidence of an ultrasonic wave, the theoretical analysis shows that the reflection spectrum takes local minima at multiple frequencies, which correspond to the resonance frequencies of the coating layer depending on the wave velocity and the thickness under the assumption of relatively low loss factors. Furthermore, the lower envelope of the reflection spectrum yields the critical attenuation condition, which is closely related to the loss factor of the coating. These theoretical findings are applied to the viscoelastic characterization of the coating layers. Two bonded specimens with different substrate thicknesses validate the proposed method. As a result, the experimental data demonstrate that the estimated wave velocities and loss factors of the coatings agree well with the results reported in the previous paper. After the experimental validation, the proposed method is applied to the curing monitoring of a resin coating on a metal substrate. The ultrasonic responses provide the temporal variation of the layer resonance frequencies and the critical attenuation frequency by the resin curing. These results can be used to estimate the viscoelastic properties of the coating, which imply the curing state of the resin.

In bridge engineering, void damage in rubber bearings refers to the phenomenon where gaps occur between the rubber bearing and the bridge structure due to factors such as aging, overloading, and earthquakes. Void damage can lead to abnormal stress distribution, bearing deformation failure, and uneven load transfer, thereby affecting the overall stability and safety of the bridge. Consequently, long-term monitoring of void damage is essential. However, methods such as manual inspection and computer vision have disadvantages, including high costs, lengthy processes, and limited monitoring scope. To address these challenges in void damage detection, this paper proposes a novel method that combines the active sensing approach with the Bayesian optimization (BO)-optimized Extreme Gradient Boosting (XGBoost) algorithm. A total of 1200 sets of experimental data under different void conditions were obtained through the active sensing method, and a high-precision void damage detection model based on the BO-XGBoost algorithm was proposed. The study found that the proposed void damage detection model achieved an Accuracy value of 95.0% on the test set, indicating that the combination of the active sensing method with the BO-XGBoost algorithm can effectively solve the problem of void damage detection in laminated rubber bearings.

Crack identification is an essential yet challenging task in structural health monitoring. This study presents a novel method for automatic crack detection based on optical frequency-domain reflectometry (OFDR) sensors, tackling the challenge of extracting crack-induced information from sensor measurements. Initially, a damage index is created by removing the substrate strain field obtained via the inverse finite-element method (iFEM) from the strain distribution recorded by OFDR sensors, allowing for the automatic identification of crack locations with an error margin of 7.68 mm. Next, using the established anomaly index, the support vector regression model, optimized by particle swarm optimization, predicts both the depth and width of the cracks simultaneously. The effectiveness of this crack detection method is confirmed through concentrated loading tests performed on four simply supported H-steel beams. Finally, the study examines and discusses how the size and diversity of training sets, as well as noise, affect prediction accuracy. This method enables the automated identification of crack depth and width, which is essential for structural safety evaluation and practical engineering applications.

Traditional signal decomposition methods, such as ensemble empirical mode decomposition (EEMD), variational mode decomposition (VMD) and singular spectrum decomposition (SSD), can decompose a multicomponent signal into several simple components. However, when the signals are very complicated and crossover instantaneous frequencies exist, traditional methods often lose their effects. Because of the complexity of the mechanical structure and the interferences of noise, the vibration signals of the rotary machine are very complicated, and may contain crossover instantaneous frequencies. Thus, traditional methods may fail to get accurate results. In order to solve the problem above, a new signal decomposition method named as adaptive synchrosqueezing chirplet group decomposition (ASSCGD), is proposed and applied to gear fault diagnosis. ASSCGD firstly use adaptive chirplet transform to get the time–frequency-chirp-rate representation of the signal. Then, synchrosqueezing transform under time-varying parameter is used to further concentrate the energy of the representation. Next, ASSCGD extract 3-d ridge lines of all components through the ridge search algorithm. The decomposition results are finally obtained through group decomposition method. The analysis of simulation signals and experimental bevel gear signals verifies the effectiveness of ASSCGD in gear fault diagnosis.

This article proposes a method for detecting the dynamic balance of brake discs based on deep convolutional neural networks (DCNN) and GraphSAGE networks, aimed at addressing the issue of dynamic balance detection of automotive brake discs under real-world operating conditions. As a key component of vehicle safety, the braking system’s performance directly affects the vehicle’s braking efficiency and driving safety. During use, the brake disc may develop an imbalanced state due to factors such as uneven thickness, warping, and surface dents. However, the number of normal state data samples is often much higher than that of the imbalanced state, and traditional detection methods frequently face the challenge of low recognition accuracy in such scenarios. To overcome this challenge, this article utilizes multisource data, including vibration signals, sound signals, and brake pressure and performs feature extraction and compression through DCNN. It then combines the graph structure characteristics of the GraphSAGE network to aggregate information from adjacent nodes, significantly improving detection performance in imbalanced data scenarios. Experimental results show that the proposed method achieves an effective detection accuracy of at least 93.98% on imbalanced datasets. The average area under the curve value obtained from the ROC curve is 0.9483, indicating good classification capability. t-Distributed Stochastic Neighbor Embedding (t-SNE) visualization further confirms the method’s effectiveness and robustness in distinguishing between different brake disc states. The study demonstrates that the proposed method solves the data imbalance issue without the need for data augmentation or significant modifications to the network structure. It offers an efficient and feasible solution for dynamic balance detection of automotive braking systems under complex operating conditions, with important engineering application value.

Under the conditions of variable speed and load for the motor, there are some problems such as complex operation data, difficult fault feature extraction and insufficient model generalization ability. A feature fusion method based on adaptive chirp mode decomposition (ACMD) and maximum absolute value rule (MAVR) and a fault diagnosis model based on improved convolutional neural network are proposed, which provides an efficient anti-noise fault diagnosis solution for the motor under variable conditions. First, the instantaneous frequencies (IFs) of vibration and current signals are extracted by ACMD, and then the MAVR is used to fuse the two IFs as new samples and input into the CNN model based on convolution kernel width and depth optimization. Based on a self-built platform, the experimental data of three-phase induction motor from static state to operation at 1800 rpm and from no load to heavy load under normal state, bearing and rotor mechanical faults, stator and rotor electrical faults are obtained, and the fault recognition accuracy of the proposed fault diagnosis method on the training set is more than 97%. In the model test, Gaussian white noise, colored noise, and random uniform distribution noise are added to the test set in a single and mixed way, respectively. The results show that the accuracy of the method is more than 71% when the noise intensity is greater than the signal strength, the accuracy is more than 84% when the noise intensity is equal to the signal strength, and the accuracy is more than 89% when the noise intensity is less than the signal strength, which proves that the proposed fault diagnosis method has strong anti-noise capability and robustness.

Type-3 composite overwrapped pressure vessels (COPVs) are increasingly being validated through research and practical applications because of their lightweight and high-strength properties. Based on these structural strengths, they are widely applied to the aerospace industry for the storage and transportation of fuel and high-pressure systems. Type-3 COPVs are manufactured by filament winding continuous carbon fiber layers onto the outer surface of a metal liner to withstand internal pressure and external structural loads. The composite manufacturing process makes these vessels prone to defects during winding. This emphasizes the need for precise non-destructive testing technologies to assess their structural integrity. This paper demonstrates the feasibility of using a pulse-echo (PE) mode inspection with a rotational scanning method based on laser ultrasonic testing. The depth-wise defect detection performance of PE mode with high-power laser integration was evaluated. Furthermore, PE mode was applied to inspect a type-3 COPV with artificial defects, and its applicability and practicality were validated through comparison with the through-transmission mode.

Corrosion presents significant threats to structural integrity within the oil and gas industry, increasing maintenance demands and associated costs. However, traditional inspection techniques often face limitations in detecting early-stage corrosion, particularly under challenging environmental and operational conditions. This study investigates structural health monitoring using low computational cost data-driven approaches, specifically principal component analysis, t-distributed stochastic neighbor embedding, and locally linear embedding to enable early and autonomous corrosion detection capabilities in a scalable manner for embedded systems. Corrosion was induced in a steel pipe using an ionic solution, with nearly 400 GB of guided wave signal data recorded across well-defined corrosion stages via an array of piezoelectric transducers. The results demonstrate that the proposed approach enables the detection of corrosion with a sensitivity of up to 0.39% reduction in the cross-sectional area. Furthermore, the results validate the feasibility of establishing autonomous damage detection thresholds, mitigating the need for periodic inspections. The methodology proved robust, effectively isolating environmental effects without the need for environmental and operational condition compensation techniques. Unsupervised dimensionality reduction effectively detects early structural changes, such as corrosion and transduction loss, with guided waves and optimized parameters, offering strong support for traditional inspection methods.

Rolling bearings are critical components in rotating machinery. Throughout their operational life, they endure periodic loading cycles that could lead to the formation of spalls. While current capabilities enable early detection of incipient spalls, which helps prevent catastrophic failure of the machine, to utilize the entire operational life of the bearings, it is essential to estimate their spall severity and remaining useful life. Using physics-based models and experimental results, this article introduces an integrative approach. We develop a new conceptual framework for monitoring bearing health, assessing defect severity by identifying physical processes that govern defect evolution, and predicting bearing failure in real-world applications. The framework incorporates four models: the dynamic model, oil debris monitoring (ODM) model, damage model, and finite element model, along with experimental work, including vibration analysis, ODM data, and strain measurements using fiber Bragg grating sensors. The integration of experimental work with these models provides condition and health indicators for both diagnosis and prognosis. By using this model, the research community can gain a deeper understanding of spall propagation mechanisms, which will result in better predictions regarding the remaining useful life of rolling bearings.

Images are widely employed to analyze vibration signals and extract information for damage detection in noninvasive structural health monitoring (SHM) techniques. However, video quality must often be high in resolution to accurately extract patterns associated with damage in classification processes. In this context, this article introduces a robust approach that enables classifying a structure’s health state using noisy, low-resolution images, which are much cheaper and easier to obtain. The approach involves decomposing these videos using optDMD—dynamic mode decomposition to extract patterns and assess changes over time, proposing a metric for SHM based on video data that detects and localizes damage while providing qualitative information on its extent. To illustrate the formulation, numerical tests on an Euler–Bernoulli beam and experimental validation on a beam with damage caused by varying crack sizes are conducted. Videos of different quality and resolution are used to demonstrate the extraction of both modal characteristics and the contributions of the identified modes to image reconstruction and the detection of the beam’s structural states.

Preoperational inspections of oil and gas pipelines are critical for ensuring their operational safety and integrity before commissioning. Given the complexities of pipeline environments and the wide range of potential defects, a comprehensive inspection methodology is essential. To address these challenges, we propose a novel composite perception fusion detection framework that offers comprehensive detection and localization of both environmental and defect anomalies through multisensor fusion. The proposed deep localization and classification decoupling (DLCD) network employed as the base detector simplifies the high-dimensional detection problem by decoupling the tasks of localization and classification, allowing for efficient defect detection with few-shot learning. The forward multispectral fusion detection system integrates infrared thermal testing (IRT) and visual testing (VT) to mitigate their respective limitations. Additionally, the incorporation of prior pipeline environment knowledge allows for efficient object-level registration of infrared and visible image pairs. The probability-based fusion strategy is employed to leverage the redundant information from both IR and visible modalities, significantly enhancing detection accuracy. Furthermore, by incorporating spatial relationships between forward and circumferential views, the circumferential defect detection system can efficiently detect weld defects based on the pipeline environment while achieving a 96.7% reduction in computational complexity. The proposed system is experimentally validated on a preoperational pipeline as well as a standard pipeline with artificial defects. Comparative experiments with state-of-the-art algorithms are performed to further verify the effectiveness and superiority of the framework.

Delivery drones have become increasingly important in recent years. It is advantageous for commercialization that suppliers are able to deliver orders autonomously and directly to their customers via air transport. However, the safety aspect must be considered. Real-time inspection of delivery drones during operation helps preventing accidents and a threat to civilians. For continuous monitoring, the sensors must be installed on the drone throughout the flight. A promising approach uses the inherent excitation of the servomotors for vibration-based Structural Health Monitoring (SHM). Vibrations can be recorded using triaxial acceleration sensors and analyzed using suitable methods such as stochastic subspace-based fault detection or histogram difference. In comparison to nondestructive testing, reference measurements of the intact structure are necessary in SHM. As soon as laboratory conditions no longer exist and environmental parameters, for example, wind, influence the vibration spectrum, classification methods are necessary for compensation. The recording of comprehensive reference datasets with different environmental conditions is limited by the battery life. This work focuses on diagnosing irreversible rotor blade damages of an 8.5 kg delivery drone. A parametric analysis taking into account systematic fusion of damage indicators and a specific number of considered references were determined for this purpose in order to assess the severity of the existing damage. Onboard SHM to evaluate the airworthiness in real time for linear and hovering flights was achieved.

Rotating machinery fault diagnostics has received a lot of attention in recent years. As a result, there has been an increase in research interest in rotating machine intelligent fault detection, especially when using measurement from multi-sensors. However, an accurate fault diagnosis is still challenged based on nonlinear and non-stationary vibration signals. On the other hand, not enough research has been done on structural information fusion from multi-sensor measurement due to the complexity of spatial–temporal correlation. This paper explores the use of vibration signals in multi-sensor measurement fusion for rotating machinery fault diagnostics, and a method using a cross-attention-based dual-branch graph neural network (CA-GNN) is proposed. First, improved complete ensemble empirical mode decomposition with adaptive noise (ICEEMDAN) is used to decompose vibration signals. Variational mode decomposition is then used to further deconstruct the highest-frequency subsequence after its parameters have been optimized using the whale optimization algorithm. Next, we build the CA-GNN, which contains two space branches for high- and low-frequency signals after initially creating the graph using multiple sensor data sets. Information across both branches for high- and low-frequency components can then be efficiently fused. Lastly, two experimental scenarios are used to illustrate the suggested technique and assess its viability and accuracy for rotating machinery fault diagnosis. Results indicate that the proposed method can diagnose rotating machinery health issues with an average accuracy of up to 99%, indicating that the method’s performance can fulfill real-world requirements.

In the context of escalating global water security concerns, effective monitoring of water pipeline leakages remains a critical challenge, with existing distributed sensing technologies falling short in sensitivity and accuracy. This study introduces a novel application of piezoelectric-distributed acoustic sensing (PDAS) technology using sensor-enabled piezoelectric geocable (SPGC) to address these limitations. Our approach enhances both the sensitivity and the positional accuracy of leakage detection. Through a series of controlled experiments, we demonstrate that the SPGC technology can discern leakage signals with high precision, with a remarkable increase in effective voltage from 2.263 to 19.636 mV, marking an 867% increase. This advancement allows for the accurate location of leakage points and the derivation of a voltage–pressure leakage equation. The research results not only validate the effectiveness of SPGC in monitoring pipeline leakages but also contribute to the development of a new distributed monitoring method, offering significant potential for long-term operation and maintenance of water pipelines. Combined with the voltage vibration signal of the SPGC, a method for pipeline leak identification and location based on the flow velocity equation was proposed. The research results are expected to provide a new distributed monitoring method for long-term operation and maintenance of water pipelines.

In the context of Industry 4.0, new sensing and communication technologies have unlocked vast amounts of process data, offering significant potential for its transformation into actionable insights to support manufacturing decisions. The reliable detection and diagnosis of faults in rolling element bearings pose a significant challenge for condition-based maintenance and fault detection and diagnosis (FDD), which are critical strategies for enhancing equipment reliability and reducing operational costs. Deep learning methods, such as convolutional neural networks (CNNs), can extract features from vibration signals compared to traditional signal processing. However, these methods in isolation are insufficient to reliably detect novel fault conditions and faults in variable working environments. Also, existing novelty and anomaly detection criteria are not accurate enough to correctly distinguish novel or unseen faults. This study introduces a multi-fault detection framework leveraging a variational autoencoder with Mahalanobis distance (MD) novelty scores for unknown condition detection and a hybrid CNN-Swin transformer (Swin-T) model for incremental learning and fault classification. Using frequency-domain transformation and image-based representation of vibration signals, a hybrid model with a CNN-based feature extractor after projecting to the patch embedding layer of a simplified Swin-T model is trained incrementally with novel conditions to allow continuous learning and adaptation. Extensive validation with three separate datasets from fault simulation test rigs demonstrates the superior performance of the method over traditional and cutting-edge models in FDD and novelty detection (ND), achieving near-perfect accuracy (99.7%), precision (99.8%), recall (99.6%), and F1 score (99.7%). ND outperformed traditional approaches with an MD novelty score threshold yielding a true-positive rate of 98.9% and a false-positive rate of 1.2%. Additionally, incremental learning improved classification accuracy by up to 5.4% for newly introduced fault types, highlighting its adaptability. These results demonstrate the framework’s ability to enhance reliability and efficiency in industrial machinery maintenance by identifying both known and novel fault conditions with high precision.

Existing bridge monitoring methods face high monitoring costs, and the processing and forecasting of monitoring data often rely on machine learning which lacks interpretability in the prediction results. Based on the Neural Basis Expansion Analysis for Time Series Forecasting (N-Beats) model, this study proposes an SAR-Nbeats (S-N) model for extracting, decomposition, and predicting bridge deformation. The input of S-N model is the bridge deformation data, which are obtained by Persistent Scatterer Interferometric Synthetic Aperture Radar (PS-InSAR) technique. At first, using Sentinel-1A imagery from 2018 to 2023 as the data source, bridge deformation results are obtained through PS-InSAR technology. Then, extremum symmetric mode decomposition and seasonal-trend decomposition are employed to decompose bridge deformation into trend, seasonal, and random components effectively, conducting adaptive classification based on different periods. Finally, based on the periodic and trend characteristics of bridge deformation, improvements and parameter adjustments are made to conventional N-Beats algorithms. The time-series data after decomposed are used as input to train the improved N-Beats model and obtain prediction results. Compared with the original algorithm, the main improvements include transforming the input data into modal decomposed data and associating the parameters of the fitting function with the deformation composition of the bridge. The bridge deformation patterns were evaluated based on climatic rules and InSAR time-series prediction results, yielding the following findings: by comparing the prediction results, the performance of SAR-Nbeats model is better than Long Short-Term Memory (LSTM) and Autoregressive Integrated Moving Average Model (ARIMA), which has highest R ² 0.8605. The SAR-Nbeats model improves the accuracy and interpretability of bridge deformation predictions by refining the input of the data-driven forecasting model. It can achieve the goal of monitoring and early warning for bridges without ground-based monitoring systems with lower computational effort and faster processing speeds.

Current vibration-based damage identification methods face difficulties in accurately identify damage features due to the low richness of data feature for wind turbine blade. This article introduces convolutional long short-term memory (ConvLSTM) that can better characterize the spatiotemporal characteristics in deep learning and explores the damage identification method combining stacked ConvLSTM network with structural state space model. A state space model enhanced stacked ConvLSTM for blade damage identification is proposed. First, the vibration signals of the blades at different damage states are converted to time–frequency images through the preprocessing of normalization and wavelet transform. The preprocessing operation improved the damage characteristics of the original vibration signals. Then, the designed stacked ConvLSTM is used to train and test time–frequency images at different damage states and output damage states and corresponding probability values through Softmax component. During the training, the different between the state equation of blades and cell state of stacked ConvLSTM is taken as loss function. Finally, specific parameter of the proposed state space model enhanced stacked ConvLSTM are set using the displacement data of blades in OpenFast software, and the recognition results are compared and validated with the mainstream networks convolutional neural network (CNN), LSTM, and ConvLSTM. The results show that, among these networks, 1D CNN, 2D CNN, LSTM, ConvLSTM, and BConvLSTM, the proposed state space model enhanced stacked ConvLSTM network achieves the best recognition results. Compared with the standard ConvLSTM network, the accuracy and mean intersection over union of the proposed state space model enhanced stacked ConvLSTM network are improved by 1.69% and 4.4%, respectively. Moreover, the proposed state space model enhanced stacked ConvLSTM achieved recognition accuracy of over 97% at different wind turbine blades working conditions. This indicates that the proposed state space model enhanced stacked ConvLSTM for blade damage identification has high accuracy and robustness. The effectiveness of the proposed state space model enhanced stacked ConvLSTM in blade damage identification has been validated again through laboratory scale wind turbine blade damage test.

We propose and test a novel sensor for structural health monitoring (SHM) based on a nanocomposite film consisting of indium tin oxide (ITO) nanowires embedded in a polyvinyl butyral (PVB) matrix to detect cracks. The sensor operates by measuring the electrical resistance between terminals when it is adhered to the surface of a structure. When damage mechanisms are present, the ITO nanowires either break or rearrange, causing a change in the sensor's internal resistance. However, certain limitations must be addressed for broader applications. Among these, the effects of loading, humidity, and temperature fluctuations in the host structure pose significant challenges. Experimental tests were conducted to evaluate the environmental influences on measurements, revealing minimal data variation, indicating that these factors are not critical concerns. Additionally, sensitivity tests were performed to assess the sensor's response to structural changes emulating damage mechanisms, and the sensor demonstrated accurate detection. In crack propagation experiments, the sensor detected cracks during their nucleation, propagation, and ultimate failure stages. The results show this novel sensor has strong potential for crack detection and monitoring applications. Another key advantage is the simplicity of signal collection and analysis, as measurements can be obtained through static readings without excita-tion signals in the structure. Furthermore, the sensor's resistance correlates well with the extent of damage.

This research aims to evaluate the potential of fiber-optic sensing of reinforced concrete (RC) structures for improved understanding of damage progression in laboratory testing and for rapid postearthquake damage assessment of real structures. A testing campaign was conducted on RC specimens with different detailing options for RC arch bridges, which experience high axial load and bending moment during earthquakes. Several types of fiber-optic cables were embedded within the four RC arch rib specimens, which were subjected to varied axial loads followed by cyclic lateral loading. The fiber-optic cables enable detailed distributed strain measurement within the RC specimen throughout the testing, as well as the residual distributed strain after each load cycle. Results provide new insight regarding the progression of cracking and the measurement of plastic hinge formation; results are compared with previous plastic hinge models. The residual strain measurements led to the development of a novel damage index that enables correlation of residual distributed strain with the maximum drift experienced during cyclic loading; this correlation is essential to evaluate potential damage when continuous fiber-optic sensing is not economically feasible and only postearthquake fiber-optic strain data are available. The findings from the testing will not only have a direct impact on the design of RC arch ribs but also pave the way for distributed fiber-optic sensing-driven postearthquake damage assessment.

Autoencoder reconstruction-based unsupervised anomaly detection is widely used in structural health monitoring. However, these methods typically require training on historical data from healthy structures, collected under environmental conditions similar to the test data. This limits their practical use, as it demands a comprehensive dataset of historical guided waves gathered across various environmental and operational conditions. Additionally, these methods fail when the training data contain a significant portion of damage-induced guided waves, as the autoencoder may reconstruct damaged waves just as effectively as normal ones. To overcome these challenges, our anomaly detection model is trained directly on current measurements, eliminating the risk of environmental discrepancies between training and test data. Furthermore, our baseline optimization strategy biases the autoencoder toward reconstructing normal guided waves, enabling reliable anomaly detection even when a large proportion of the training data are damage-induced waves. Additionally, we present a strategy to enhance the model’s practical performance by optimizing the weight factor for baseline loss and the baseline set size, based on guided wave reconstruction performance, without relying on damage labels. The effectiveness of this baseline-optimized autoencoder model, even when the training data contain significant damage-induced guided waves, is validated through measurements from 10 regions, each spanning 80 days of guided wave data collected under uncontrolled and dynamic environmental conditions.