Probabilistic assessment of bridge safety by integrating measurements with computational simulations
Cited 0 times inCited 0 times in
- Probabilistic assessment of bridge safety by integrating measurements with computational simulations
- Lee, Jaebeom
- Lee, Young-Joo
- probabilistic assessment; bridge safety; finite element reliability analysis; structural condition monitoring; Bayesian inference
- Issue Date
- Graduate School of UNIST
- The condition monitoring and safety assessment of bridges have been emphasized to prevent bridge collapses, which can cause tremendous economic losses and fatalities. Indeed, bridges may suffer from structural deterioration, and accordingly, regular inspection including visual inspection has been adopted as the standard bridge condition monitoring method in several countries. In some cases, the nondestructive evaluation has been incorporated to evaluate local damages, such as steel corrosion and crack, and structural health monitoring has been sometimes introduced to monitor bridges in real time. Although such inspections provide useful information about the condition and safety of bridges, bridge safety assessment driven by measurement data has several limitations: it generally requires sufficient and reliable monitoring data, which are often difficult to obtain due to measurement errors, and the inspection data themselves may have an ambiguous link to the bridge safety. Therefore, sophisticated structural analysis models were proposed in various studies to assess the complex failure mechanism and safety of bridges. However, such a model-based safety assessment also has a limitation in that modeling with high fidelity is often difficult and expensive due to model errors.
Several studies have combined the data-driven and model-based approaches to overcome the aforementioned limitations. However, these studies generally did not consider uncertain factors in both of measurement data and structural analysis models. Indeed, several bridges have been reported to suffer from structural failures, in spite of significant efforts toward understanding and identifying the structural condition and safety. Thus, various uncertainty factors with respect to aleatory and epistemic uncertainties have been emphasized in assessing the structural safety of bridges in several standards and studies. In other words, probabilistic safety assessment of bridges by combining measurement data and structural analysis models needs to be intensively studied, which is the motivation behind this study.
This study suggests a probabilistic assessment method for the condition monitoring and safety assessment of bridges by integrating measurements with sophisticated structural analysis (i.e., computer simulation) models. First, a new probabilistic framework of multi-scale modeling considering the measurement data is introduced to build accurate finite element (FE) models with practical computational cost. This framework enables to model the lower-scale of structural members (e.g., geometrically complex steel strands) as a relatively simple geometrical elements in the upper-scale model without a significant reduction of accuracy; hence, it can decrease the computational cost of the structural analysis. Once the model is built using the proposed framework, it is utilized in two approaches for probabilistic bridge safety assessment and condition monitoring.
First, a new finite element reliability analysis (FERA) platform is proposed. In general, a structural reliability analysis systematically requires multiple times of structural analysis, which would be impractical when incorporating a computationally expensive analysis model, such as a three-dimensional nonlinear FE model. Therefore, structural analysis models are sometimes simplified in some studies to reduce the cost, resulting in a reduction of accuracy. Accordingly, this study proposes an efficient FERA platform, which is a reliability analysis tool utilizing an FE model with high accuracy. As this platform enables to decrease the required number of structural analyses during a reliability analysis, the ultimate or serviceability performance of bridges against future loading (unprecedented situations such as natural disasters or sever deterioration) can be efficiently estimated using this platform.
Second, a new probabilistic method of structural condition monitoring (SCM) utilizing the Bayesian inference method by integrating the sensor measurements with the preconstructed FE model is developed. Traditionally, the structural condition has been monitored with thresholds defined statistically as a constant value, where the environmental and loading effects may not be accounted for and filtered out. To consider those uncertainties, Gaussian process regression (GPR), which is one of the machine learning-based methods, is introduced. In GPR, data are assumed to follow a multivariate Gaussian distribution, and the covariance matrix is optimized to maximize the marginal likelihood for given measurements. As a result, the probabilistic regression model (i.e., the predictive mean and prediction interval) can be derived with the optimized covariance matrix which considers the environmental and loading variabilities.
Probabilistically unusual data (i.e., anomalies) can be detected using this probabilistic monitoring model of structural condition, and subsequent inspection may be considered when these anomalies occur continuously. When the structural condition (e.g., corrosion, crack size, and loading) is reported to be different from the previously recorded conditions, the structural parameters of the FE model can be calibrated. Correspondingly, the FERA results and SCM model are also updated with the advanced FE model, and the bridge safety assessment results for the target bridge are expected to be accurate in this cycle.
- Department of Urban and Environmental Engineering (Disaster Management Engineering)
- Go to Link;
- Appears in Collections:
- Files in This Item:
- There are no files associated with this item.
can give you direct access to the published full text of this article. (UNISTARs only)
Show full item record
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.