Seismic response simulation of mega-scale urban building clusters is crucial for assessing urban resilience, but is hindered by data scarcity, prohibitive computational costs, and the limited scalability of conventional analytical methods. This paper presents an intelligent computational framework designed to overcome these limitations, enabling efficient and accurate seismic simulation for urban areas with hundreds of thousands to millions of buildings. The framework is composed of three integrated modules: (i) a parameter calibration module that infers key structural dynamic properties from readily available building information; (ii) an adaptive dynamic K-means clustering module to drastically reduce computational workload; (iii) a response prediction module that employs a novel graph neural network to learn from inter-entity relationships, enabling accurate regression from a minute fraction of known data. The framework's efficacy is validated through a mega-scale case study, yielding 4.416 billion seismic response predictions for 1.15 million buildings in Shanghai under 3840 ground motions. The results demonstrate exceptional performance. The framework's predictions of engineering demand parameters achieved a response error of just 4.19%. Significantly, by learning from just 0.0087% of the total response space, the proposed framework reduces runtime from 12.2 years (estimated) with time history analysis to just 59.8 h, offering a robust and scalable solution for quantitative seismic resilience assessment and disaster risk management in modern mega-cities.

Earthquake ground motions demonstrate inherent non-stationarity and significant intensity variability due to the complex interaction of source mechanisms, propagation paths, and site conditions. This presents dual challenges for existing modeling approaches, including data scarcity in extreme scenarios and coupling of time-frequency-amplitude features. This paper proposes an innovative decoupled generative framework by hierarchically separating three stages: time-frequency generation, amplitude prediction, and time history reconstruction. First, leveraging a conditional generative model, peak-normalized acceleration time-frequency spectra are generated randomly based on key physical parameters as constraints, revealing temporal variations in frequency and amplitude. Then, a cross-attention mechanism is employed to fuse parameters and spectral representations, enabling precise prediction of peak ground acceleration (PGA). Lastly, a phase recovery algorithm is applied to reconstruct the time-frequency spectra into time-domain signals, with amplitude scaled by PGA to produce realistic waveforms. Additionally, data augmentation strategies are proposed to improve physical consistency, alongside a gradient field analysis method to quantify the impact of physical constraints and improve interpretability. To validate its extrapolation capabilities, comparative experiments with the end-to-end simulation method were conducted in cross-regional earthquake scenarios. The results demonstrate that the proposed framework achieves better performance in both the accuracy of response spectra prediction and the fidelity of time-frequency detail restoration. This study provides a novel approach for stochastic simulation of strong earthquakes in data-scarce scenarios.

The generation of realistic wind speed time histories is essential for wind engineering analysis but remains challenging due to the scarcity of high-quality measured data and the non-stationary, stochastic nature of atmospheric turbulence. While existing artificial intelligence-based data-driven methods in wind engineering mainly focus on conditional forecasting tasks, the unconditional generation of independent wind speed time histories has received limited attention. To address this gap, a novel spectrogram-based generative framework, called DiffWind, is proposed for wind speed time history generation based on denoising diffusion probabilistic models (DDPM). Wind speed time histories are transformed into magnitude spectrograms using the short-time Fourier transform (STFT), modeled in the spectral domain using a U-Net-based diffusion model, and reconstructed through the Griffin–Lim algorithm (GLA). Field-measured wind speed records were employed to tune the STFT-GLA hyperparameters and train the DDPM. The effectiveness and accuracy of the STFT-GLA combination were validated through numerical experiments, while the diffusion-based spectrogram generation was evaluated using quantitative metrics. The results indicate that the proposed framework can generate high-fidelity wind speed time histories that reproduce key statistical, temporal, and spectral characteristics of measured wind data while demonstrating the capability to generate longer-duration records, highlighting its potential for wind engineering applications.

The vibration serviceability of large-span footbridges under crowd loading has become a governing design criterion. However, the significant divergence among existing load models introduces substantial uncertainty into response prediction. This study presents a comparative evaluation of ten representative crowd walking load models from international codes and the relevant literature. It objectively evaluates their theoretical mechanisms regarding crowd synchronization and structural damping. Initial parametric sensitivity analyses are conducted utilizing single-degree-of-freedom systems. Subsequently, the predictive capabilities of these models are evaluated against field measurements from four footbridges under resonant and off-resonant conditions. The investigation reveals that response-based amplification models (e.g., M1–M3) assume high synchronization and thus overestimate accelerations under natural unrestricted resonant flows. However, these models perform reasonably well under off-resonant high-frequency conditions. In contrast, load-based models that incorporate the square-root growth law and explicit damping terms (e.g., M8–M10) better represent uncorrelated crowd flows under resonant conditions. These observations, while derived from a limited set of validation cases, provide indicative guidance and illustrate that accounting for phase randomness and structural damping is important for serviceability assessment.

The seismic engineering demand parameters (EDPs) of building clusters exhibit significant spatial correlations and need full consideration in regional risk and reliability assessments. This study presents an efficient scheme to determine the joint distribution of multi-structure EDPs, which captures all EDP correlations and enables direct calculation of system reliability for building clusters. This scheme generates spatially correlated random ground motion fields through ground motion cross power spectrum density (PSD) models with stochastic harmonic function simulations. Subsequently, the decoupled multi-probability density evolution method (M-PDEM) is integrated to conduct seismic analysis of building clusters under random ground motion fields to determine their EDP joint distribution. An example of three linear single-degree-of-freedom (SDOF) models shows that the proposed scheme requires only hundreds of analyses to achieve the same accuracy as 10(5) Monte Carlo Simulation (MCS) analyses, while also capturing the nonlinear correlations among EDPs. Finally, an engineering application of three reinforced concrete (RC) frame shear-wall buildings under a rare earthquake scenario is investigated, and the joint collapse probability by the scheme is compared with that by commonly-adopted assumptions of mutual independence and linear correlation. The results reveal that relative errors by the two assumptions can reach up to 39 % and 22 %, respectively.

The effective visualization of urban-scale earthquake simulations is pivotal for disaster assessment but presents significant challenges in terms of computational performance and accessibility. This paper introduces a lightweight, browser-based visualization framework that leverages Web Graphics Library (WebGL) to provide real-time, interactive 3D rendering without requiring specialized software. The proposed framework implements a novel dual multi-level-of-detail (LOD) strategy that optimizes both data representation and rendering performance. At the data level, urban buildings are classified into simplified or detailed geometric and computational models based on structural importance. At the rendering level, a dynamic graphics LOD approach adjusts visual complexity based on camera proximity. To realistically reproduce dynamic behaviors of complex structures, skeletal animation is introduced, while a quad tree-based spatial index ensures efficient object culling. The framework’s scalability and efficacy were validated by successfully visualizing the seismic response of approximately 100,000 buildings in New York City. Experimental results demonstrate that the proposed strategy maintains interactive frame rates (>24 frames per second) for views containing up to 4000 detailed buildings undergoing simultaneous and dynamic seismic behaviors. This approach significantly reduces rendering latency and proves extensible to other urban regions. The source code supporting this study is available from the corresponding author upon reasonable request.

Aftershocks (ASs) following strong mainshocks (MSs) can exacerbate structural damage or lead to collapse. However, the scarcity of recorded data necessitates reliance on artificial sequences, which have difficulty in characterizing the time-frequency correlation between MSs and ASs. This study innovatively converts the AS time history prediction into an image translation task, exploiting the invertible transformation between accelerograms and time-frequency representations. An encoder-decoder neural network is developed to encode the MS information into the latent space of a pre-trained generative adversarial network, enabling accurate AS predictions through the decoder. The integration of seismic parameters further improves the AS prediction performance. Comparative analyses demonstrate that the proposed method outperforms the traditional ones on accuracy and robustness and reproduces the non-stationarity of ASs.

Strong earthquake disasters can cause noticeable damage correlations among regional buildings with similar mechanical properties, termed structure-to-structure seismic damage correlation, which has a pronounced impact on the regional seismic risk assessment and therefore needs to be properly quantified. This study first introduces a time-domain analytical and equivalent frequency-domain analysis based on random vibration theory for calculating structure-to-structure seismic damage correlation coefficients. Subsequently, by employing the spatially consistent white noise excitation and the spatially varying white noise excitation with the Luco-Wong coherence function, the structural filtering effect and the ground motion spatial correlation effect are progressively incorporated, and an analytical interstructural damage correlation model incorporating structural dynamic properties and spatial distance is derived. Comparations with Monte Carlo simulations and existing empirical models demonstrate that the proposed analytical model possesses a clear physical basis and high reliability. Finally, a case study was conducted on a district having 29,461 buildings in Shanghai, China to illustrate the influence of interstructural damage correlation on the regional seismic risk. Results show that disregarding the interstructural seismic damage correlation can lead to underestimation of overall loss uncertainty.

Data-driven deep learning application in earthquake engineering highlights the insufficient quantity and the imbalanced feature distribution of measured ground motions, which can be mitigated with artificial ones. Traditional ground motion generation techniques tend to extend the catalogs conditioning on existing records, while current deep learning-based methods such as generative adversarial networks (GANs) only provide limited duration or sampling rate, thus obstructing further applications. In this paper, an invertible time–frequency transformation process is employed, based on which the transformed earthquake representation is implicitly modeled by advanced GANs for high-resolution and unconditional generation. Moreover, leveraging the disentangling property of the GAN’s latent space, the newly developed latent space walking method is adopted to assure the generations with controllable time–frequency features. A feature-balanced generated ground motion dataset has been constructed in combination with the proposed methods, and the application potential was demonstrated through comparative experiments of different datasets.

Compressive membrane/arch action at small deformations and resistance evolution path at large deformations are two key factors in determining whether an RC frame structure can resist progressive collapse. In this study, two deep learning models based on deep and cross network (DCN) are developed to predict the progressive collapse resistance of RC frames. DCN model I is constructed to predict the compressive membrane/arch action resistance, while DCN model II is developed to predict the resistance displacement curve considering the dynamic effect. 464 records regarding structural collapse were collected through extensive literature review and stringent data filtering. They were randomly divided into 80% and 20% for training and testing the two models, to which the column removal scenario, with or without slab, boundary condition, beam net span, beam section, beam reinforcement ratio, and material property were selected as input features. Moreover, Shapley additive explanations (SHAP) method was introduced to interpret the proposed DCN model. The results indicated that the performance of DCN model I in predicting the compressive membrane/arch action resistance was satisfactory with an MAPE value of 10.66% and an R2 value of 0.9799, respectively, while being more accurate than the existing yield line theory and Park's compressive arch calculation model. The proposed DCN model II can further capture the evolution of the dynamic resistance of RC frames accurately. The two DCN models have been deployed online to public for application.

Urban seismic damage assessment has recently become an emerging research topic due to the accelerating global urbanization trend, to which the seismic responses of building clusters to various earthquakes are a prerequisite. Traditional methods for this task, including vulnerability analysis and time-consuming time history analysis, may suffer from accuracy or efficiency problems especially for nonlinear response calculation. Machine learning methods allow for rapid and accurate response prediction, but current applications still lack scalability (on the size of structures or earthquakes) and the corresponding real datasets. To tackle this issue, this paper proposes an artificial neural network framework for simultaneously predicting nonlinear seismic responses of all buildings in a cluster subjected to multi-earthquake inputs. Inspired by the advanced collaborative filtering techniques, the framework converts the regional response prediction into a matrix completion problem, thereby aggregating information extracted from historical response records and physical characteristics to improve performance. The framework is used to assess nonlinear responses of a real urban region consisting of 2788 buildings subjected to 3798 measured ground motions. The results clearly demonstrated that the proposed framework achieves orders of magnitude faster than time history analysis and average errors below 3 % on several response metrics, showing high computational efficiency and accuracy.

Offshore structures, such as oil and gas platforms and offshore wind turbines, are subjected to wind and wave loads simultaneously during their service lifetime. Since the wind and wave states are of significant randomness and dependence, the probabilistic modeling of joint wind and wave conditions plays an essential role in the safety design of offshore structures. Currently, three different methods can be adopted to establish the joint probabilistic model, which, however, are somewhat inconvenient in applications. The recently emerged generative adversarial networks has been demonstrated to be effective in dealing with high-dimensional random variables in several fields. In this study, the implicit joint probabilistic model of joint wind and wave load conditions is developed based on the Wasserstein generative adversarial network with gradient penalty. Long-term metocean reanalysis data of the site in the South China Sea is used to train and validate the model. After one million training steps, high-quality samples that are quite similar to the original data can be generated by the developed model. In addition, statistical comparisons of the generated samples obtained by the C-vine copula approach and the developed generative adversarial network model are performed as well, which demonstrates the effectiveness and superiority of the developed model.

Artificial neural networks have been used to predict nonlinear structural time histories under seismic excitation because they have a significantly lower computational cost than the traditional time-step integration method. However, most existing techniques require simplification procedures such as downsampling to maintain identical length and sampling rates, and they lack sufficient accuracy, generality, or interpretability. In this paper, a recursive long short-term memory (LSTM) network was proposed for predicting nonlinear structural seismic responses for arbitrary lengths and sampling rates. Referring to the traditional integral solution method, the proposed LSTM model uses the recursive prediction principle and is therefore applicable to structures and earthquakes with different spectral characteristics and amplitudes. The measured ground motions and multilayer frame structures were used for model training and validation. The rules of hyperparameter selection for practical applications are herein discussed. The results showed that the proposed recursive LSTM model can adequately reproduce the global and local characteristics of the time history responses on four different structural response datasets, exhibiting good accuracy and generalization capability.

提出一种基于长短期记忆(long?short-term?memory,?LSTM)神经网络模型计算非线性结构地震响应的新方法,采用单向多层堆叠式LSTM架构,并借助滑动时间窗实现递推计算.改进了模型预测效果的评价指标,可考虑响应在不同幅值区间的敏感性差异,避免了传统评价指标的相位敏感问题.利用实测地震动和多层框架结构进行了新方法的验证,给出了网络超参数的取值原则,并讨论了不同工况下模型的泛化能力.结果表明,LSTM模型的计算精度较好、对地震动类型具有鲁棒性.由于神经网络模型便于分布式、云部署的特点,该方法可在城市区域地震响应快速模拟等传统数值方法受限的应用场景发挥作用.

钢螺旋楼梯因造型轻盈,集建筑美与结构美于一身而广泛地应用于现代建筑中,但由于其自振频率较低,行人步行荷载下可能会产生振动舒适度问题而影响使用.当前对螺旋楼梯的人致振动响应计算中,往往只考虑步行荷载的竖向分量,未明确是否应考虑步行荷载多分量的耦合效应.针对一实际钢螺旋楼梯,采用数值模拟方法,对三向加载与单向加载工况下的楼梯动力响应进行了对比.结果表明,在三向耦合作用下钢螺旋楼梯单向峰值加速度增大10％左右、总体变化幅度在5％左右,实际设计中可采用步行荷载单分量加载计算响应再乘以放大系数的方式考虑三向耦合效应.在此基础上进一步验证了已有的直线型楼梯竖向加速度简化预测方法,表明在一阶模态控制的情况下,该方法可直接用于螺旋楼梯.