运营模态分析是测试大跨桥梁振动特性的最主要途径，广泛应用于成桥测试、维修加固、健康监测和振动控制中。大跨桥梁所受环境荷载复杂多变，振动频率极低且模态密集，加之模型和测量误差的存在，识别的模态参数表现出很大的离散性和较差的可重复性。如何准确识别桥梁模态参数并度量其不确定性是研究和应用中的关键科学问题。贝叶斯算法为运营模态识别及其不确定性度量提供了系统性理论框架，其研究难点在于算法的实用化和自动化。鉴于此，本申请项目拟开展以下研究：（1）发展变分贝叶斯快速算法，增强算法的实用性；（2）应用稀疏贝叶斯学习原理剔除虚假模态，实现算法的自动化；（3）探究多步运营模态分析问题，拓展其应用范围。通过开发基于Web的大跨桥梁运营模态测试数据库、建立Benchmark模型、分享测试数据和软件代码，本项目将极大地促进大跨桥梁运营模态分析的发展和在实际工程中的应用，为桥梁设计、模型校验、养护维修提供技术支持。

Operational modal analysis (OMA) provides the main tool to learn the vibration properties of long-span bridges, and it plays a key role in bridge design validation, maintenance and repair, health monitoring and vibration control. There shows a large varibility and low repeatability in the identified modal parameters of long-span bridges, because of the high complexity and time-variability of external load, the low and dense modes and the ubiquitous model/measurement error. Reliable modal identification and uncertainty quantification has become a principle issue in the OMA of long-span bridges. To quantify the identification uncertainty, the Bayesian approach gives a systematic framework by calculating the posterior distribution of modal parameters. However, its immaturity in computation and automation has greatly impeded its application in practice. Based on the applicant’s research in time-domain Bayesian OMA, the following contents will be investigated in this project: (1) develop a fast variational Bayes algorithm to enhance its efficiency; (2) remove spurious modes using sparse Bayesian learning for an automated implementation; (3) extend the study to the case of multi-setup OMA of large-span bridges. Based on the outcome of this research, a web-based database for OMA of large-span bridges will be developed: Benchmark models, measured data and various algorithms will be openly accessible to researchers and practitioners, which will definitely enhance the development of OMA and its practical applications.