The abnormal tension change in one cable in a cable-stayed bridge indicates cable damage, so it is necessary to obtain the benchmark of the cable tension. After establishing the regression model of the mapping between the temperature-induced cable tension and the bridge temperature field or other data, the regression value can be used as the benchmark. To improve the regression model, the geometric compatibility and mechanical equilibrium must be jointly considered. Therefore, two data groups, which contain the bridge temperature field and the regression values of the temperature-induced deflection of the main girder, are input into the deep learning neural networks. Time lags exist between the temperature features and the temperature-induced cable tension, but are not significant between the temperature-induced deflection and tension. So one neural network module, which receiving the regression values of the temperature-induced deflection, is composed of Convolutional Neural Networks (CNNs). The other neural network module, which receives the temperature features, is composed of stacked CNN and Long Short-Term Memory (LSTM). Finally, several convolution kernels will integrate the array output from the two modules into one regression value of the temperature-induced cable tension. By combining the input data and the composite neural networks, the R² of the regression models of the temperature-induced cable tension is more than 0.95, and the error of the regression values is less than 0.3 kN. In the future, if the nonlinearity at the curve inflection point and the complexity in data distribution could be solved, the stability of the model may be improved.

Half-through truss bridges exhibit significantly different mechanical characteristics due to their open tops, necessitating special treatment for calculating their free torsional moment of inertia. This study proposes a novel method: considering the constraints of vertical web members and crossbeams on the top chord, the top chord is equivalently modeled as a continuous beam on elastic supports. An equivalent horizontal bending moment of inertia of the top chord is derived by converting the top chord to the height of the top crossbeam while maintaining equivalent stiffness based on the equivalence principle. According to the analytical formula for the torsional moment of inertia and detailed parametric analysis, the main dimensional parameters affecting the torsional stiffness of half-through truss bridges include bridge length, bridge width, and main truss height. These parameters primarily enhance the bridge’s torsional stiffness by influencing the constrained torsional moment of inertia. However, due to scale limitations and aesthetic requirements, these dimensions cannot be increased indefinitely. In such cases, besides considering weight and aesthetics, increasing the size of the chords may be considered to enhance torsional stiffness. The interactions among the various factors affecting torsional behavior are relatively complex, and more systematic research is recommended for future study.

Among routine wind turbine maintenance activities, the timely and accurate detection of blade surface damage is crucial for maintaining structural integrity and ensuring operational safety. In practical on-site wind farm inspection workflows, aerial images captured via unmanned aerial vehicles (UAVs) frequently suffer from miniature, geometrically irregular defect targets, highly cluttered natural backgrounds, fluctuating lighting intensity, heavy shadow occlusion, and persistent surface dirt contamination, which creates an intractable performance conflict that conventional detection frameworks struggle to balance high localization precision and low inference latency simultaneously. To address these challenges, this study proposes an improved detection model named FCD-YOLO. Within this architecture, lightweight FasterNet Blocks are fully embedded into both the backbone feature extraction pipeline and the neck multi-scale fusion branch to reduce computational redundancy and improve inference efficiency. In parallel, a Channel Prior Convolutional Attention (CPCA) module is embedded to selectively enhance damage-sensitive features by jointly considering channel-wise and spatial information and suppressing irrelevant background feature activations. Moreover, an auxiliary detection branch (DetectAux) is incorporated to introduce supplementary supervisory cues during training, which facilitates more robust learning of features across different scales and alleviates missed detection risks for tiny, low-contrast blade defects. Comprehensive experimental validation indicates that the proposed model achieves mAP@0.5 and mAP@0.5:0.95 scores of 80.98% and 57.77%, respectively, surpassing the corresponding results of the YOLO11s baseline (79.22% and 56.08%). Side-by-side benchmark tests against multiple state-of-the-art mainstream detectors further validate the comprehensive competitive superiority of the proposed FCD-YOLO framework. Overall, this work provides a promising approach for workstation-based offline or near-real-time wind turbine blade damage inspection with remarkable lightweight characteristics and robust anti-interference capability for complex field environments.

Accurate extraction of temperature-induced structural responses from bridge monitoring data is the core prerequisite for constructing a more precise regression model linking the temperature field to temperature-induced responses, whose outputs can provide reliable baseline benchmarks for the safety evaluation of bridges. Traditional wavelet multiscale decomposition methods for temperature-induced response extraction typically rely on empirical parameter selection, leading to inadequate signal separation accuracy and limited model generalization. To eliminate the black-box effect and poor generalization caused by empirical parameter selection, physically meaningful indices can be established to quantitatively assess the physical consistency between temperature-induced responses obtained from wavelet analysis under different parameters and the temperature effect. This makes the parameter selection process physically interpretable rather than empirical. Accordingly, this study proposes an optimized wavelet multiscale decomposition method based on time–frequency similarity evaluation to physically extract temperature-induced response signals. Specifically, two quantitative evaluation indices are established to characterize the time-domain and frequency-domain similarity between extracted response signals and original temperature data, respectively. The optimal parameter combination for wavelet decomposition is automatically determined by integrating the dual time–frequency similarity results, which avoids the randomness and subjectivity of artificial parameter tuning. For method validation, deep learning regression models are constructed to map temperature fields to structural responses processed via different wavelet parameter schemes. The experimental results show that the regression model corresponding to the proposed optimized method achieves the optimal predictive performance, with a coefficient of determination (R²) reaching 0.90. This outcome verifies that the time–frequency similarity maximization strategy can significantly improve the extraction accuracy of temperature-induced responses in wavelet multiscale analysis. The proposed method provides a robust and reliable data processing framework for monitoring data purification, offering a positive basis for subsequent temperature-induced response extraction and baseline modelling.