Structural Safety最新文献

In engineering practice, Bayesian model updating using field data is often conducted to reduce the substantial inherent epistemic uncertainties in geomaterial properties resulting from complex geological processes. The Bayesian Updating with Subset simulation (BUS) method is commonly employed for this purpose. However, the wealth of field data available for engineers to interpret can lead to challenges associated with the “curse of dimensionality”. Specifically, the value of the likelihood function in the BUS method can become extremely small as the volume of field data increases, potentially falling below the accuracy threshold of computer floating-point operations. This undermines both the computational efficiency and accuracy of Bayesian model updating. To effectively address this technical challenge, this paper proposes an improved BUS method developed based on parallel system reliability analysis. Leveraging the Cholesky decomposition-based midpoint method, the total failure domain in the original BUS method, which involves a low acceptance rate, is subdivided into several sub-failure domains with a high acceptance rate. Facilitated with an improved Metropolis-Hastings algorithm, the improved BUS method enables the consideration of a large volume of field data and spatial variability of geomaterial properties in the probabilistic back analysis. The results of an illustrative soil slope, involving spatially variable undrained shear strength, demonstrate that the improved BUS method is effective in simultaneously incorporating a substantial volume of field measurements and observations in the model updating process. Through a comparison with the original BUS method, the improved BUS method is shown to be useful for Bayesian model updating of high-dimensional spatially variable geomaterial properties and slope reliability 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 10⁵ 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 life-cycle seismic resilience assessment of sea-crossing highway bridges plays a crucial role in guiding decisions for their long-term operation, maintenance, and rehabilitation. Due to the inherently stochastic nature of marine environments, evaluating the resilience of bridges while considering all possible environmental scenarios throughout their service life necessitates substantial computational efforts and presents practical challenges. Thus, this study develops a three-stage framework for predicting the life-cycle seismic resilience of sea-crossing highway bridges. Stochastic models for marine environmental conditions and bridge durability are developed and validated using experimental measurement data. A modified Good Lattice Point-Partially Stratified Sampling (GLP-PSS) method is employed to generate a uniform and limited number of samples. A typical prestressed concrete sea-crossing highway bridge is selected as the benchmark bridge, and parameterized numerical models are established using 460 representative environmental parameter samples on the OpenSees platform. Leveraging the environmental model and material properties, the durability of the bridge is predicted over its service life. Nonlinear time history analyses are carried out for each bridge model using 120 real ground motion records, which allow the identification of variations in seismic demands, capacities, and system fragilities at different time intervals. Subsequently, the life-cycle seismic resilience of the bridge is predicted utilizing surrogate models based on the response surface method (RSM) and artificial neural networks (ANN), respectively. Finally, the time-dependent probabilistic characteristics of seismic resilience are thoroughly discussed. Results indicate that ANN demonstrates a higher degree of generalization capability in predicting the life-cycle seismic resilience. Focusing solely on changes in mean resilience over the service time may lead to an underestimation of bridge resilience, as it may ignore the tails of its distribution, potentially resulting in an overestimation of bridge resilience. Furthermore, global warming may expedite the decline in resilience.

Acquiring engineering data is frequently expensive, resulting in sparse data that may lead to a lack of knowledge for design and analysis. Thus, it is not always feasible to precisely determine the probability density functions (PDFs) of uncertain model parameters. Under such circumstances that involve simultaneous aleatory and epistemic uncertainties, repeated uncertainty propagation (UP) analysis is generally required. In this paper, a novel approach for hybrid UP is proposed by integrating B-spline chaos and augmented change of probability measure (aCOM) for meeting different goals. The B-spline chaos is adopted to represent the complicated computational model as a function of an arbitrary input random variable, while the aCOM is employed to reconstruct the PDF of the model output when the input PDF is changed due to epistemic uncertainty. In the case of small epistemic uncertainty, hybrid UP can be achieved directly by changing the assigned probabilities of existing sample results. While in the case of large epistemic uncertainty, additional samples from an augmenting PDF are generated. The proposed method is compatible with both cases. The numerical algorithm of the proposed method is presented and illustrated by four benchmark problems. Further, the accuracy and efficiency of the proposed method are substantiated by four numerical examples compared with analytical solutions or Monte Carlo simulations. An attempt to enhance the proposed method with the aid of active subspace methods to handle high-dimensional problems is also discussed in this work. The limitations and potential improvements of the proposed approach are outlined as well.