This paper proposes a novel method for structural damage identification that integrates the power spectral density (PSD) of structural acceleration responses with densely connected convolutional networks (DenseNet). The method transforms the training object of the DenseNet into a numerical matrix (PSD matrix) for structural damage identification. Leveraging transfer learning, the DenseNet models are initially trained on simulated data and further fine‐tuned using experimental data to enhance robustness and generalization. Results demonstrate that frequency‐domain signals processed by PSD significantly enhance model performance, achieving lower mean squared error (MSE), higher Pearson’s correlation coefficient ( R value), and reduced mean absolute error (MAE) compared to time‐domain signals. The effectiveness of this method was verified on a six‐story frame structure. This study underscores the efficacy of transfer learning in bridging the gap between simulated and real‐world data, thereby facilitating effective structural health monitoring and damage identification.

Tuned liquid damper (TLD) is one of the main technologies for passive control. The fundamental modal parameters of a TLD contain natural frequency and damping ratio, which are primarily related to the liquid height in the TLD device. Although the liquid height of the TLD is calibrated before installation, it is still necessary to identify the physics parameters of the TLD and the main structure during the service period to prevent the detuning of the TLD. A physics parameter identification method for the structure‐damper‐coupled system based on mechanics‐enhanced stochastic subspace identification (SSI) is proposed in this paper, illustrated through annular TLD (ATLD). By extracting the state matrix of the first controlled mode of the main structure, the natural frequency and damping ratio of the ATLD are identified, thereby determining the liquid height of the ATLD. Numerical models of structure‐ATLD‐coupled systems with different degrees of freedom are constructed, and their simulated acceleration responses under different excitations are obtained. Sensitivity analysis of environmental noise is performed to verify the accuracy and robustness of the proposed parameter identification method. A dynamic test was designed for a steel chimney model to further verify the practicality of this method. The results show that when the noise level of the measurement noise is below 5.0%, the average relative error in identifying the ATLD liquid height does not exceed 10%. The identification error in the damping ratio of the structure‐ATLD‐coupled system will lead to a decrease in the accuracy of ATLD liquid height estimation. The proposed method can effectively identify changes in the ATLD liquid height within the optimal frequency ratio range by analyzing the experimental data from the chimney model. The proposed method can effectively estimate the modal parameters of the coupled system, providing reliable data support for evaluating the working condition of the ATLD during its service period.

Shake tables are useful earthquake simulation tools for structural seismic experiment, but they may also inadvertently induce vibrations to nearby buildings while in operation. Accelerating the comfort level quantification process of these vibrations before conducting a shake table test is necessary. To this end, this paper focuses on the influence of vibration introduced by a 6 × 9 m large‐scale shake table at Southeast University and presents a one‐dimensional convolutional neural network–based deep learning approach to efficiently estimate the vibration comfort of the shake table laboratory and surrounding buildings. Based on the on‐site structural vibration monitoring of shake table test, a three‐dimensional numerical model of the shake table–soil–surrounding building system is established and validated through the finite element method, and thus a dataset comprising 12,215 groups of input (i.e., peak acceleration values and time‐history of the triaxial ground motion) and output (i.e., three‐directional acceleration response for nine measuring points of surrounding buildings) data is simulated. Thereafter, the deep learning network is trained with 80% of the dataset and tested with the remaining 20%. The test results indicate that the approach enables the network to directly extract dynamic features from triaxial ground motion accelerations and to accurately estimate the weighted acceleration level (WAL) of nine different locations at the surrounding buildings. Finally, the optimized network is verified through an actual shake table experimental test on a self‐centering concrete structure, which confirms the superior performance of the proposed approach on shake table–induced vibration comfort estimation. The approach is also beneficial for researchers to design reasonable loading scenarios before conducting shake table tests.

Seepage is one of the main causes of dam accidents, characterized by long latency periods and spatiotemporal randomness. In this study, an innovative combined algorithm model (AT‐LSTM‐CUSUM) is proposed to predict such leakage hazards. First, a long short‐term memory (LSTM) network model based on an attention mechanism is established to focus on key influencing factors in predicting the time series data. Following the time series prediction, an improved Cumulative Sum (CUSUM) change‐point monitoring algorithm is introduced. Within a sliding window period, a control function collects cumulative residuals, and a threshold test is performed to determine whether a potential hazard trend exists. Using monitoring data from a pressure measuring pipe in a concrete dam as the experimental subject, five related influencing factors were collected (upstream and downstream water levels, temperature, precipitation, and structural aging). These data were fed into the AT‐LSTM model for iterative parameter tuning, yielding optimal prediction results. These results were compared with those of the LSTM, GRU, ARIMA, and Prophet models, validating the superior performance of the AT‐LSTM model. In addition, by simulating the seepage hazard occurrence process, the change‐point monitoring effectiveness of the improved CUSUM algorithm was tested. A parameter sensitivity analysis of the window period and threshold values revealed that the algorithm performed effectively in detecting seepage hazards. The innovative algorithm proposed in this paper exhibits strong early warning capabilities and holds significant value for dam safety monitoring and maintenance.

Domain adaptation (DA) techniques have recently been developed as a promising approach to enhance the performance of structural damage classification algorithms. Unlike traditional methods, DA imposes fewer constraints on the nature and completeness of datasets, although its effectiveness largely depends on the similarity between the datasets used for knowledge transfer. This paper proposes a novel approach for assessing structural similarity to improve DA in structural health monitoring (SHM). The identification of suitable source data for knowledge transfer in damage detection is an open issue in SHM, especially when dealing with important geometric, mechanical, and topological differences between the structures. To address this issue, damage detection accuracy is increased by investigating similarity in the modal features of different framed structures, with the aim of understanding their dynamic behavior through a similarity index based on divergence measures. In detail, this work proposes a novel modal sensitivity‐based similarity index which relies on the Kullback–Leibler divergence computed from vibration‐based dynamic features. This similarity index effectively reveals how structures differing in highly sensitive parameters exhibit greater divergence. When DA is applied, source datasets with higher similarity lead to improved multiclass damage classification accuracy on the target framed structure. The proposed index can be used to systematically rank candidate source structures before applying DA, allowing a more efficient selection process. Its applicability extends to large‐scale structures, where managing heterogeneous structural datasets is essential, supporting data‐driven SHM strategies with enhanced transferability and reliability in real‐world monitoring scenarios.

Structural damage identification based on metaheuristic algorithms is an important part of structural health monitoring with great potential. However, the metaheuristic intelligent algorithms probably have flaws of slow convergence speed and low calculation accuracy, which need to be improved to address engineering optimization problems. In this paper, the black widow optimization (BWO) algorithm is used for structural damage identification. In addition, a multistrategy fusion–improved BWO (IBWO) algorithm is proposed by introducing the tent chaotic mapping, the golden sine equation, the gazelle wandering equation, and the boundary treatments. First, in the population initialization stage, tent chaotic mapping is introduced to improve the quality of the initial solution. Second, the golden sine strategy is used to acquire the optimal solution quickly in local search. Then, the motion equation of the gazelle algorithm is employed to enhance the global search ability and avoid the algorithm falling into the local optimal solution. Finally, the boundary processing strategy is presented to reduce the calculation of solutions and improve the optimization efficiency. A novel damage identification objective function is redefined by combining the modal assurance criterion and the modal flexibility. Then, a two‐story rigid frame structure is utilized for numerical simulations. Moreover, experimental studies with a simply supported beam were carried out to verify the performance of the proposed damage identification method. Simulation results and experimental studies demonstrate that, even with the interference of strong noise, the IBWO algorithm has a higher accuracy and efficiency in damage identification compared to the BWO algorithm.

The presence of free water in the concrete slurry significantly influences the crack patterns of concrete. In this study, uniaxial compression tests were conducted on concrete specimens with varying moisture contents under acoustic emission (AE) monitoring. Through parametric analysis and machine learning, the cracking process of water‐containing concrete was studied, signal patterns during the cracking process were identified, and the impact of moisture content on the damage evolution and fracture mechanism of concrete was understood. The results indicate that free water is capable of absorbing high‐frequency signals. With the increase of moisture content, the AE signals decrease. The failure of concrete is mainly of the tensile type, while the shear‐type accounts for a relatively small proportion. The presence of free water decreases the likelihood of diagonal shear failure in concrete structures. The unsupervised learning was used for various moisture content analyses. Three distinct AE signal patterns were identified during the concrete compression tests: frictional motion signals of the compression surface, fracture surface activity signals, and aggregate cracking signals. Based on the moisture content, this study analyzes the variations in signal responses across different modes. A predictive model was established utilizing the BP neural network to differentiate signals of various modes, achieving an accuracy rate of 99%.

The modulus of elasticity of concrete (Ec) is an essential parameter commonly used in concrete design, concrete curing monitoring, and deterioration evaluation and damage detection of concrete structures. A quick and reliable in situ determination of Ec helps in accurate decision‐making for construction and maintenance. The electromechanical impedance (EMI) technique using surface‐bonded lead zirconate titanate (PZT) patches (referred to as SBP) has become a popular nondestructive method for monitoring concrete structures due to ease of operations. The existing research mainly utilized baseline‐dependent approaches to monitor the changes (e.g., hardening or damage evolution) of concrete structures. However, relevant baselines are greatly influenced by the status of PZT sensors and bonding layers, limiting the practical applications of this technique. This paper presents a baseline‐free EMI resonance method for measuring Ec of concrete without using prior baseline data for the first time. Numerical and experimental studies on standard concrete cubes (100 mm in width) with SBP were conducted to examine the proposed method. A dimensionless physical quantity was first proposed to view the EMI signals. Then, resonance peaks highly related to concrete properties but insensitive to the status of the sensor system were selected, physically described, and correlated to the concrete parameters through numerical analyses. Finally, experimental validations, covering the measurement of Ec, repeatability of the chosen resonance peaks, and temperature effects, were conducted to illustrate the proposed method’s stability, accuracy, and sensitivity.

A 1‐dimensional convolutional neural network (1D CNN) model is developed to process deep learning of raw impedance signals for smart aggregate (SA)‐based concrete stress monitoring. First, the framework of the SA‐based stress monitoring using deep learning of raw impedance signals is described. An impedance measurement model is designed for a SA‐embedded concrete body under compression. A 1D CNN model is developed for deep learning of raw impedance signals corresponding to various stress levels. Three approaches for concrete stress monitoring are designed to deal with data availability, signal noises, and untrained stress levels. Second, a few SA‐embedded concrete cylinders are experimented to measure impedance signals under various stress levels. Finally, the performance of the proposed method is extensively evaluated by investigating the feasibility of the K‐fold cross‐validation to deal with the data availability and the effects of signal noises and untrained data on the accuracy of stress estimation in the SA‐embedded concrete cylinders.

Compared with conventional linear viscous damping (LVD), rate‐independent linear damping (RILD) mitigates floor response acceleration in low‐frequency structures more effectively without compromising the control performance of displacement responses during long‐period or low‐frequency earthquakes. Although theoretical and experimental attempts have been made to overcome the noncausality of RILD devices and realize RILD using passive or semiactive devices, only a few studies have highlighted the impulse‐response precursor and modal analysis method of multistory structures incorporated into RILD devices. This study investigated the impulse‐response precursor of a noncausal RILD system and proposed a novel modal decomposition method for the structures equipped with nonproportionally distributed RILD devices. Additionally, real‐time hybrid simulation was conducted to validate the effectiveness of the proposed modal analysis method and the feasibility of realizing ideal RILD using mechanical devices. This study is the first to demonstrate the differences between RILD and LVD devices in terms of controlling the modal responses of low‐frequency structures and how RILD can lower the floor response acceleration more effectively compared to LVD.