基于贝叶斯的预应力混凝土桥梁可靠度评估

Assessment on Reliability of Prestressed Concrete Bridges Based on Bayesian

摘要:

在侵蚀环境下，桥梁受多种因素的耦合作用，抗力不断退化且存在时间变异性，其本质属于非平稳随机过程，若采用统一标准化的抗力退化模型将无法准确描述抗力退化过程中的不确定性。为探究在侵蚀环境下预应力钢筋混凝土桥梁的性能退化情况，对其安全性能进行准确评估，将已有的氯离子侵蚀模型、抗力退化计算模型、荷载效应时变模型视为先验知识，把桥梁检测数据作为观测数据，利用贝叶斯神经网络来减少先验知识中模型及其参数的不确定性，提高侵蚀环境下预应力钢筋混凝土桥梁可靠度评估的精度和可信度。研究结果表明，利用贝叶斯神经网络更新氯离子侵蚀模型、抗力退化计算模型、荷载效应时变模型后，可以更准确地预测预应力混凝土桥梁在侵蚀环境下的可靠度及剩余使用寿命。

Abstract:

In an erosive environment, the bridges are subject to the coupled effect of multiple factors, with their resistance constantly deteriorating and showing temporal variability, whose essences belong to a non-stationary stochastic process. If a unified and standardized resistance degradation model is adopted, it will be impossible to accurately describe the uncertainty during the resistance degradation process. To investigate the performance degradation of prestressed reinforced concrete bridges in erosive environment, the safety performance of bridges are accurately evaluated. The existing chloride ion erosion models, resistance degradation calculation models, and time-varying load effect models are considered as prior knowledge, and the bridge detection data are used as observation data. Bayesian neural networks are used to reduce the uncertainty of models and their parameters in prior knowledge, and to improve the accuracy and credibility of reliability evaluation of prestressed reinforced concrete bridges in erosive environments. The research results indicate that after updating the chloride ion erosion model, resistance degradation calculation model and load effect time-varying model by Bayesian neural networks, the reliability and remaining service life of prestressed concrete bridges in erosion environments can be predicted more accurately.