Rapid prediction of seismic damage of existing buildings is of paramount importance for post-earthquake emergency response and accelerating rescue and recovery in urban regions. To achieve rapid and quantitative diagnosis of the safety of existing buildings, a multi-feature driven method for rapid evaluation and prediction of seismic damage is proposed. For reinforced concrete structures, through measured dynamic properties in-situ and nonlinear performance parameters obtained from HAZUS manual, a numerical model of existing building is developed. Using the ATC-63 ground motion record sets and a data-driven nonlinear damage index, a large database is generated. The feature engineering is adopted to reveal the correlations between various input features. A random forest machine learning model is employed to predict the structural seismic damage quantitatively based on design features, measured structural features, and ground motion features. The coefficient of determination for the test set is 0.99, and the proportion of samples with a relative error within ±20% is 99.23%. Model interpretability analyses reveal the importance of various input features on the output results, with structural modal periods being one of the most critical features. Finally, using the data from existing seismic stations, the model successfully predicts the damage condition of the examined existing buildings under a specific ground motion. The result indicates that the proposed framework of seismic damage prediction for existing building structures exhibits ideal predictive accuracy and efficiency.

Artificial intelligence algorithm is a key technology to realize the automated and intelligent design of building structures. However, due to the lack of physical rules constraint, artificial intelligence algorithms in practical engineering applications tend to provide unreasonable results. Therefore, in this study, by integrating structural design rules into the generative adversarial network (GAN) in the form of the neural network module, a novel intelligent layout design method for steel frame-brace structures, FrameGAN-sym, was proposed. The basic principles and ideas of this method were first introduced, and then the design results of FrameGAN-sym were compared and analyzed in detail with those of FrameGAN, which proves that FrameGAN-sym can synthesize more symmetric structural drawings according to the requirements of the proposed symmetry constraint network module. The mechanical properties of the design of FrameGAN, FrameGAN-sym and the engineers were compared through three engineering cases of steel frame-brace structures with different heights. The results show that the design of FrameGAN-sym is closer to that of engineers in terms of mechanical properties, and the torsion effect of the FrameGAN-sym-designed structure is reduced compared with the FrameGAN-designed structures.

The design of shear wall structures currently relies primarily on the experience of designers and continuous trial and error process. To achieve rapid and optimal design, a method integrating generative design with intelligent optimization has been proposed in this study. Initially, the Stable Diffusion (SD) model is used for the generative design of shear wall structures. Subsequently, based on the results of the generative design, taboo search is employed for intelligent optimization. The SD-based generative design mainly includes fine-tuning of SD by low-rank adaptation methods using small sample data, as well as image processing techniques such as vector pixelation and pixel vectorization. The intelligent optimization based on taboo search includes the definition of parameter space, the definition of domain actions, the optimization objectives and the design of the optimization process. The proposed method is compared and verified with two actual project cases. The results show that the generative design method for shear wall structures can provide designs that meet basic design requirements in approximately 30 seconds. The designs optimized by the intelligent system achieve a similarity of up to 85% compared to the solutions given by designers, fulfilling the purpose of aiding the design process.

The core columns can improve the bearing capacity of ordinary masonry walls. To explore the feasibility of applying core columns in 3D printed concrete walls, composite walls of 3D printed concrete reinforced hollow walls and cast-in-place reinforced concrete core columns were designed. The effects of cross-section form (i.e.,vertical ribs, diagonal ribs and composite ribs ) and the number of core columns (i.e., two-core columns, three-core columns and five-core columns ) on the axial compression performance of 3D printed concrete walls were studied. The results show that the failure modes of the wall can be categorized into three types, that is, the bottom crush of one side of the wall, the top half crush and spalling of one side of the wall, and the overall damage of the wall. The cross-section form of the vertical and diagonal combination of ribs has the best mechanical performance, which can improve the integrity and bearing capacity of the wall. The increase of the number of core columns can significantly improve the bearing capacity of the wall, by 16.3%-19.1%, but the dispersed arrangement of core columns will weaken the integrity of the wall and reduce the cracking load of the wall by 34.3%-48.5%. By comparing the bearing capacity calculation formulas of different codes, it is found that the bearing capacity calculated according to the reinforced block masonry wall is in good agreement with the experimental value, and the error is less than 8.5%. ABAQUS was used to simulate and analyze the specimens, and the mechanical properties of the weak plane between the material layers were reduced by setting the defect layer. The difference between the simulation results and the experimental results is less than 10%, indicating that the model is accurate and can be used to estimate the bearing capacity of 3D printed concrete walls with core columns.

In recent years, in order to improve the work efficiency of drawing review and reduce the rework rate, BIM-based intelligent drawing review and automatic drawing review technologies have become a hot topic in the field of BIM research. However, the joint model, which is crucial in  structures, has not yet been studied for automatic compliance check. Against this background, an intelligent drawing review framework based on Revit secondary development and deep learning is proposed, which includes three parts: information extraction, semantic enrichment, as well as compliance reasoning and suggestions. The intelligent drawing review system is developed for four kinds of beam-column joints in steel frame structures. The one-dimensional convolutional neural network trained with joint eigenvalues as samples fully considers the geometric features of the model, and the accuracy of beam-column joint classification reaches 98.59% after optimization, which is higher than that of other commonly used machine learning classification algorithms. The developed compliance reasoning algorithm can complete the construction rule checking and strength checking of the joint model, and put forward optimization suggestions. The developed intelligent drawing review system has completed the intelligent drawing review for a four-story steel frame structure model with a total of 136 beam-column joints. The accuracy rate is 97.79% and the time-consuming is 86 s, which improves the accuracy and efficiency of drawing review compared with manual drawing review.

Efficient construction safety managements facilitate healthy and high-quality development of the construction industry. To ensure safe construction, it is crucial to prevent human-machine collision accidents. To accurately recognize the safety risks of human-machine operation, an automated method for human-machine collision risk recognition and warning was proposed based on self-calibration of dual-scale monocular cameras. This method recovered the three-dimensional scale of monocular vision based on the geometric characteristics analysis of the construction site and the extraction of target features, leading to precise measurement of the spatial distance between humans and machines. Furthermore, an approach for human-machine collision warning and visual simulation  was proposed based on the kinematic characteristics of construction machinery. This method can trigger multi-level collision warnings based on a human-machine distance threshold.A construction project in Shanghai was selected as a test case and achieved accurate object detections (with average accuracy of 91.2%), spatial distance measurements (with accuracy above 98%) and collision event assessments, with the algorithm frame rate meeting real-time monitoring requirements.

The failure of large reinforced concrete support systems (high formwork systems) may lead to instantaneous collapse and overturning of structures, thus the accuracy and real-time performance of the displacement monitoring of the formwork system are particularly important. Accordingly, a lightweight multi-target visual perception method for safety monitoring of high formwork system is proposed. For target occlusion and poor light in the monitoring process, a Pyramid-Histogram-Otsu multilevel threshold segmentation method is proposed, based on the area and shape consistency constraints, combined with a passive infrared target to achieve a robust extraction of the center. For multiple measurement points and real-time monitoring of the formwork system, based on the initial positioning and update optimization of the target tracking window, a Camshift-Gaussian-Centroid lightweight multi-target detection algorithm is developed to realize real-time monitoring of multi-targets in the formwork system. By integrating the above innovations, an online camera software-hardware integrated monitoring and warning system with edge computing capability is further developed. The developed method and system are successfully applied to the displacement monitoring of the high formwork system of a nuclear power casting platform, realizing real-time data acquisition (2 fps) and automatic processing (with accuracy of 0.35 mm), and demonstrating the effectiveness of the lightweight multi-target visual perception method in the safety monitoring of the formwork system.

AI-based computation in civil engineering exhibits high accuracy and efficiency. However, due to its black-box nature, the results are difficult for researchers and engineers to comprehend, impeding its application in practical engineering projects that prioritize safety. To address this issue, a design formula intelligent discovery method based on dimensional analysis and engineering prior knowledge is proposed. This method utilizes intelligent computing technology to automatically identify the key features affecting the performance of materials and components from experimental data and generate design formulas that are dimensionally balanced, physically meaningful, and mechanically interpretable. A formula intelligent generation model considering dimensional constraints is established based on symbolic regression expression trees, ensuring the mechanical rationality of the formulas. Normalization methods for scenarios with multiple mechanical-geometric variables and engineering feature segmentation algorithms based on spectral clustering and decision trees are developed to further improve the stability and accuracy of the model. The effectiveness of the method is verified using the shear bearing capacity of reinforced cementitious materials as an example. The results show that the intelligent-generated formulas improve the accuracy by 61.3% and the fitting correlation by 23.3% compared to empirical formulas generated manually, with R2 value of 0.90, demonstrating excellent performance. Moreover, compared to traditional symbolic regression methods, the intelligent-generated formulas are not only more accurate but also dimensionally correct, with stronger engineering generalization capabilities. Furthermore, the proposed method contributes to revealing the mechanical mechanisms and accelerating the translation process from experimental testing to design methods for new materials and structures.

In existing studies and practical nondestructive testing applications, ultrasonic tomography images were usually utilized for manual qualitative interpretation but hardly used for accurate quantitative detection purposes of internal defects for reinforced concrete (RC) structures. To this end, a deep learning method based on array ultrasound and feature fusion neural network was proposed in this study for pixel-wise nondestructive recognition of internal cracks in RC structures. RC components with preset artificial internal cracks were manufactured. Array ultrasonic B-scan images were then acquired by testing the RC components with shear-wave low-frequency transducer array, and the dataset was setup. A deep neural network with the basic encoder-decoder architecture was developed, which was optimized by feature fusion strategy and residual modules to improve the compatibility with the semantic structure of ultrasonic B-scans. Moreover, individual local predicted images were combined with global representations by registration to indicate global information such as crack location and distribution of the entire section. The results indicate that F-scores of the training, validation, and testing sets are higher than 70%. The cracks as small as 1mm in width can be recognized by the proposed feature fusion neural network, and the mean absolute percentage error of quantified crack length is 6.22%, substantiating the effectiveness of the proposed method.

To establish a method for evaluating the distribution characteristics of prestress in concrete structures and calculating the evaluating characteristic value based on measured data, the concept of prestress ratio was introduced to represent the probability distribution of prestress in three levels，including single-bar level, component level and structure level. Through the combination of theoretical analysis and numerical simulation, the probability distribution of prestress ratio of each level was studied, and a Gaussian mixture model classification simplified analysis theory considering the prestress system layout features and the influence of design variances was established. Additionally, by introducing the sequential estimation theory and setting the maximum estimation error of the guarantee rate feature parameter at 95% as the constraint condition, a sampling stopping criterion combined with the bootstrap method, EM algorithm, and pivot method was established. A sequential sampling estimation and evaluating characteristic value calculation method for prestress ratios of the structure, single-bar, and component levels was proposed. The effectiveness of this method was verified by a numerical example of a prestressed concrete frame structure without bond. The research shows that compared to traditional sampling methods using fixed sample sizes, sequential sampling methods can effectively quantify estimation accuracy, reduce the estimated sample size and reduce detection costs. The maximum estimation error of the evaluating characteristic value of components decreases gradually with the increase of the number of prestressing bars, and the estimation accuracy can be higher than 95%.

In order to prevent the occurrence of communication tower structural safety accidents and warn the possible damages in time, it is of vital significance to carry out the communication tower structural safety monitoring work. In this paper, a damage identification method for the three-tube tower structure was proposed by taking the changing rate of the statistical moment of strain as the damage index. The relationship between the strain response statistical moments and structural stiffness of multi-degree-of-freedom structural system was theoretically deduced, and the damage index of strain fourth-order statistical moment changing rate was proposed. By establishing the relationship between the fourth-order statistical moments of strain and the strain energy density, the sensor deployment was optimized based on the principle of maximum strain energy density. By setting different signal-to-noise ratios, the effectiveness of the method was verified based on numerical modeling under the condition of varied wind speed and direction, and the identification effect was compared with other damage identification methods. At the same time, it was analyzed in combination with the measured strain data of the three-tube towers on site. The results show that the method can optimize the sensor deployment,  reduce the cost and improve the monitoring efficiency, and identify the damage of the three-tube tower in the local area range with 20 dB ambient noise.

Large-span spatial structures are typical wind-sensitive structures, and their wind field and wind pressure have strong randomness and spatiotemporal distribution characteristics. Therefore, it is necessary to conduct a comprehensive and systematic study on synchronous measurements of structural wind field and wind pressure characteristics of such structures. Based on the wind speed and wind pressure measurement data of Beijing Daxing International Airport terminal, the wind field and wind pressure characteristics of the terminal roof were studied in this paper, involving the wind field characteristics at different positions of the roof and the relationship between different wind field characteristics. Combining the measured wind field characteristics, the spatiotemporal distribution and non-Gaussian characteristics of wind pressure on the roof were investigated, and the power spectral densities of fluctuating wind speeds and fluctuating wind pressures were compared. The results indicate that there are significant differences in wind field characteristics at different locations on the roof. The mean wind speed is negatively correlated with the turbulence intensity and gust factor, while the turbulence intensity is positively correlated with gust factor. The wind pressure on the roof exhibits complex spatiotemporal distribution characteristics at windward edges and corridor locations where characteristic turbulence is likely to occur. The wind pressure on the majority of the roof area exhibits a non-Gaussian distribution with skewness less than 0 and kurtosis greater than 3, and the power spectral densities of fluctuating wind pressures and fluctuating wind speeds are significantly influenced by the roof characteristic turbulence.