基于信息更新与融合的在役桥梁动态可靠性预测方法研究

Bridge health monitoring (BHM) systems produce a huge amount of monitored data during the long-term service periods, how to dynamically predict load effects, dynamically update resistance model and further predict structural time-variant reliability with these data, is one of the key and urgent scientific problems in the present field of Structural Health Monitoring (SHM). Considering the monitored data's coupling, randomness, et al, in this project, first, the coupled monitored data is decoupled into high-frequency and low-frequency monitored data through the combination of moving average method and least square method. Then, with Bayesian updating and prediction theory, aiming at the characteristics of high and low frequency load effects, such as randomness, trend, periodicity, et al, the Improved Gaussian Mixture Particle Filter (IGMPF) fusion prediction algorithms, about high-frequency and low-frequency load effects’ extreme sequences obtained with interval selection method, are respectively studied, which make each load effect’s extreme value predicted dynamically, solve the problems of short-term prediction and low precision about the traditional prediction methods. With the finite element reliability method (FERM), the truncated distribution method and Bayesian method, based on the structural design information, monitored data and inspection information, the multi-layer nested time-varying updated model for the resistance of the existing bridge is also studied, which more comprehensively reflects the effects of information updating on resistance. Finally, based on the predicted load effects' extreme data and the updated resistance information, the dynamic reliability prediction method of the existing bridge is thoroughly studied, which provides the theoretical foundation for the reliability prediction of the actual BHM systems.

桥梁健康监测(BHM)系统在长期运营中积累了大量数据，怎样利用这些数据预测荷载效应和修正抗力模型，进而预测结构可靠性，是当前结构健康监测(SHM)领域迫切需要解决的关键问题之一。考虑到监测数据的耦合性、随机性等特点，本项目首先采用移动平均法和最小二乘法对信号进行解耦，将高、低频信号分开；然后，采用Bayesian修正与预测理论，针对高、低频荷载效应的随机性、趋势性和周期性等特点，研究由区间选择法得到的各荷载效应极值序列的改进高斯混合粒子滤波器(IGMPF)融合预测算法，实现各荷载效应极值的动态预测，解决传统预测方法的短期性和精度不高的问题；采用有限元可靠度方法、截尾分布法和Bayesian方法，研究设计、监测和检测信息相融合的桥梁抗力多层嵌套时变修正模型，更全面地反映信息更新对抗力的影响；最后，深入研究在役桥梁动态可靠性预测方法，为实际BHM系统的可靠性预测提供理论基础。

桥梁健康监测系统在长期运营过程中积累了大量监测数据，如何合理利用这些数据进行结构动态可靠性预测是结构健康监测领域亟需解决的关键科学问题。本项目围绕此问题，研究得到了由区间选择法得到的桥梁极值应力以及极值挠度等极值动力响应的高精度预测方法，包括贝叶斯动态预测方法、改进粒子滤波预测方法以及改进高斯混合粒子滤波预测方法等数据同化方法，实现了桥梁极值荷载效应的动态预测，解决了传统动态预测方法精度不高的问题；建立了桥梁容许应力以及容许挠度等抗力的多层嵌套动态修正方法，揭示了监测极值荷载效应对桥梁抗力时变修正的影响规律；基于修正的容许荷载效应信息和动态预测的极值荷载效应，利用一次二阶矩方法和Gaussian Copula模型，给出了失效模式独立和相关的桥梁动态可靠性预测方法。本项目的研究成果将为桥梁安全预后和预防性养护维修决策提供理论基础和应用方法。项目资助共计发表22篇论文，其中10篇SCI期刊论文，6篇EI期刊论文，2篇核心期刊论文和4篇会议论文。培养硕士生5名，其中2名已取得硕士学位，3名在读。项目投入经费18万元，支出13.7796万元，各项支出基本与预算相符。剩余经费4.2204万元，将用于本项目研究的后续支出。