基于长期监测数据和贝叶斯推理的桥梁结构可靠性预后方法研究

Utilizing the massive data accumulated by bridge structural health monitoring system for real-time evaluation and prediction of bridge reliability can provide a scientific guidance on the safe operation, maintenance and rehabilitation, and management decision of bridges. The effectiveness of bridge reliability assessment results largely depends on the accuracy of the load effect and resistance model. Bayesian inference technique is adopted to integrate multi-source information of bridges, including long-term monitoring data, inspection data, proof loads, etc., in order to well characterize the load effect and resistance model. The main research tasks of this project are: (1) adaptive Gaussian process surrogate model and Bayesian inference-based stochastic model updating technique are proposed to update the finite element model (FEM), and then the probability model of the dead load effect can be established by performing finite element analysis on the updated (accurate) FEM; (2) generalized Gaussian mixture model and Bayesian inference are developed to estimate the probability model of the strain distribution over time, thereby establishing the probability model of the time-varying live load effect; and (3) Bayesian method is used to effectively integrate the objective information (monitoring data, inspection data, proof loads, etc.) and the subjective information (prior time-varying resistance model) to set up the accurate probability model of the time-varying structural resistance. The research achievements can be used for assessment of bridge health condition and characterization of bridge reliability evolution, thus having great theoretical significance as well as engineering application value.

桥梁结构健康监测系统积累的海量数据可用来实时评估和预测桥梁结构可靠性，可为桥梁结构的安全运营、维修加固与管理决策提供科学依据。桥梁结构可靠度评估结果的有效性取决于荷载和抗力模型是否准确，本项目基于贝叶斯推理方法融合桥梁结构的长期监测数据、检测资料、验证荷载等多源信息来建立荷载和抗力的时变概率模型，实现实时及预后评估桥梁结构可靠度的目的。研究内容包括：(1)基于自适应的高斯过程替代模型和贝叶斯推理的随机模型修正技术来获取一个精确的有限元模型，然后通过有限元分析手段建立恒载效应的概率模型；(2)基于广义高斯混合模型和贝叶斯推理手段来获得应变分布随时间变化的概率模型，用于建立活载效应的时变概率模型；(3)基于检测资料、验证荷载、监测数据等已有信息和先验的时变抗力模型，贝叶斯推理手段用来建立结构抗力的真实的时变概率模型。项目成果可用于桥梁结构的健康状态评估和可靠性演化研究，具有较大的理论意义和工程实用价值。

桥梁是公路和铁路交通网的关键枢纽，在交通运输中发挥着重要的作用。有效利用长期监测数据准确评估桥梁结构的安全可靠性意义重大，为桥梁结构的健康监测、安全评估与维修加固提供科学依据和理论支撑。项目针对桥梁结构服役安全保障需求，开展了考虑不确定性的物理-数据联合驱动的桥梁结构安全状态评估方法研究。取得的创新性研究成果包括：（1）构建了整合高、低精度训练样本的自适应广义协同高斯过程模型，提出了结构不确定性分析解析方法，有效揭示了不确定性影响结构性能的机理；（2）发展了基于高斯过程代理模型的有限元模型修正高效方法，为桥梁结构状态评估提供了准确的物理模型；（3）提出了基监测数据压缩采样新方法及监测数据异常诊断方法，实现了数据高效压缩与数据异常诊断，保证了监测数据的完整性和可靠性；（4）建立了基于监测数据的风荷载建模方法及结构响应预测方法，为实时掌握不同荷载作用下桥梁结构的服役行为提供了理论支撑；（5）发展了基于概率主成分分析的结构异常状态识别方法，有效诊断出复杂服役环境下结构的异常状态。上述研究成果应用于江阴大桥、苏通大桥、九堡大桥等多座大型桥梁，为桥梁结构健康状态诊断提供了理论基础与科学支撑。