This article deals with the use of unsteady thermographic inspection as a structural health monitoring method for fast and accurate detection of internal defects in wind turbine blades. Temperature transients allow estimation of both the thickness of internal anomalies and the depth at which they are embedded. On the other hand, the use of physics‐informed neural networks to predict heat transients in the framework of the solution to the inverse problem in the proposed methodology results in a very low computational cost of the identification algorithm. As a result, the proposed technique can be easily implemented in an instrumented unmanned aerial vehicle (UAV), where onboard processing resources could be rather limited. The developed methodology is applied to two defect identification scenarios. In the first one, active heating (provided by four 250 W lamps) is used to induce thermal transients on the blade surface. In this case, even deep defects can be very accurately identified by the proposed technique. A second identification scenario considers passive heating (through solar radiation) to save battery consumption in order to extend the UAV flight time. In this case, identification accuracy depends on the rate of change in solar radiation. Although the identification errors are larger than in the previous case (with active heating), both depth and thickness can be identified with rather small errors in the considered examples.
Accurate and effective prediction of seismic response is crucial for seismic resistance of bridges and structural health monitoring (SHM). However, finite element analysis methods for bridge seismic response are time‐consuming. Furthermore, multiscale sequence information from nonstationary seismic time‐series is not effectively extracted by traditional machine learning approaches, and challenges remain in improving the accuracy of response predictions. In this paper, a hybrid deep learning model for predicting bridge responses under seismic actions is proposed to accurately capture dynamic characteristics in nonstationary seismic time series and effectively utilize critical information within the data. In the proposed framework, sample convolution and interaction network (SCINet) is employed to decompose and interactively extract hierarchical temporal features from seismic excitation and structural response sequences, the long short‐term memory (LSTM) is used to characterize sequential dependency, and the attention mechanism adaptively emphasizes informative temporal features associated with critical structural responses. The performance of the proposed SCINet‐LSTM‐attention model is validated through numerical studies of a continuous girder bridge and a cable‐stayed bridge, as well as shaking table test of a bridge tower. The accuracy of the proposed method is further compared to LSTM, attention‐based LSTM (AttLSTM), and CNN‐LSTM models. Additionally, a comparative study for varying noise levels, damping ratios, ground motion amplitudes, and hyperparameters is investigated. The numerical and experimental results demonstrate that the SCINet‐LSTM‐attention model exhibited superior performance in predicting bridge responses compared to the reference models, indicating potential as an efficient surrogate model for monitoring‐oriented seismic response prediction of bridge structures.
Arch dam deformation is a key indicator for evaluating structural safety. Influenced by the coupled effects of multiple factors such as water pressure, temperature, and aging, it exhibits nonlinear and nonstationary characteristics. Traditional statistical models have clear physical meanings but struggle to capture complex nonlinear relationships, while purely data‐driven models offer high predictive accuracy but lack interpretability. This paper proposes an arch dam deformation prediction model that integrates physical mechanisms with Bayesian optimization Gaussian process regression (BGPR). First, based on heat conduction principles, the influence of solar radiation on the dam surface temperature is quantified, and a physically enhanced input feature set is constructed. Second, variational mode decomposition (VMD) is employed to extract the primary trend component of the deformation series, which is combined with historical lagged values and original environmental features to form an augmented feature set. Third, BGPR is utilized for prediction using the augmented features. Finally, through feature grouping contribution analysis, the influence of hydraulic pressure, temperature, and aging components on deformation is quantitatively assessed. Engineering case validation demonstrates that the proposed model outperforms comparative models in both prediction accuracy and interpretability, providing a new technical pathway for arch dam deformation prediction.
Operational modal analysis (OMA) identifies the dynamic characteristics of structures under ambient excitations and plays a crucial role in structural health monitoring (SHM). However, traditional OMA methods often have difficulty accounting for the variability of modal parameters induced by environmental variations (EVs), which can obscure genuine damage-related changes and introduce bias in the identification process, particularly in the case of closely spaced modes. To address this challenge, a novel Bayesian time domain OMA framework integrated with a Gaussian process (GP) model is proposed. The proposed method explicitly quantifies and mitigates the influence of EVs on modal parameters by embedding the GP model into the classical structural dynamic formulation. Within a Bayesian inference framework, the modal parameters, GP hyperparameters, and prediction errors are jointly identified while rigorously accounting for measurement noise, modeling errors, and modal interactions. To effectively handle large-scale monitoring data, a Monte Carlo estimator and a stepwise iterative optimization strategy are developed. Validation on synthetic datasets with known parameters demonstrates that the proposed method provides more accurate and stable modal parameter estimates under varying environmental conditions and exhibits clear advantages in handling closely spaced modes compared with both the conventional OMA method and the established Bayesian frequency domain method incorporating environmental modeling. Finally, the method is successfully applied to long-term monitoring data from a full-scale cable-stayed bridge, demonstrating its capability to effectively compensate for environmental influences and provide robust and reliable modal parameter identification for engineering structures.
Long-span bridges exhibit nonstationary deformation under simultaneous environmental loads, yet existing monitoring approaches commonly predict the overall structural response without adaptively separating load-associated components. This study develops a framework combining eXtreme Gradient Boosting (XGBoost) and optimal sparsity and reconstruction-based variational mode decomposition (OSR-VMD) for real-time deformation prediction and adaptive quantification of load-associated deformation components. XGBoost is first used to map multisource environmental observations to lateral and vertical bridge deformations. A physically interpretable OSR objective then evaluates within-mode bandwidth, mode-location separation, spectral overlap, and reconstruction fidelity, and a genetic algorithm adaptively selects the VMD parameters alpha and K. The framework is evaluated using 10 months of heterogeneous monitoring data from the Forth Road Bridge, UK. The results show that XGBoost reliably reproduces the measured deformation patterns, while OSR-VMD separates the predicted responses into frequency-specific modes associated with wind and temperature variations. By integrating nonlinear prediction with adaptive time-frequency decomposition, the framework provides an interpretable approach for identifying dominant environmental-load associations in bridge deformation and supports structural condition assessment under coupled environmental forcing.
