随着我国桥梁老龄化和桥梁健康问题的日益重视，亟需建立科学有效的桥梁安全评估方法。目前的桥梁评估结果受监测设备、环境噪声、车辆等不确定性影响较大，数值分析、健康监测、人工检测三方面数据没有得到有效的结合，因此，应用智能分析与数据融合理论建立科学有效的桥梁安全等级评估体系，以期实现对多源评价指标进行融合得到更确定、可靠的评估结果，对于桥梁的运营管理与养护维修具有重要的意义。 首先，论文从环境侵蚀、材料老化、外部荷载三方面，分析了桥梁常见病害及其产生机理，同时研究了在混凝土碳化、冻融、钢筋锈蚀、拉索腐蚀、温度、风、车辆等因素影响下，各力学指标的退化规律，得到不同影响因素的变化区间，并将其施加到数值模型中，作为形成先验评估数据库的基础。 其次，基于数据融合的桥梁损伤识别方法中，应用贝叶斯融合对桥梁的损伤位置进行识别，提出了粗糙集结合贝叶斯的损伤程度识别方法，证明了经粗糙集属性约简后的剩余特征满足独立性要求，从而解除了朴素贝叶斯的应用限制。对一座混凝土连续梁桥进行了现场环境激励模态试验，通过模态分析与参数识别获取了桥梁的频率与振型值，应用损伤识别方法判断桥梁的损伤位置和损伤程度，并与现场裂缝的调查结果相比较，分析了该识别方法的适用条件。 第三，将桥梁安全评估系统分为先验评估数据库与数据融合处理两个模块，在多源影响因素作用下得到多工况数值计算结果，作为桥梁的先验评估数据库。将现场监测数据与人工检测结果传输到数据融合处理模块中，引入信息熵、模糊隶属度函数、模糊层次分析法确定指标的变权重，利用证据理论对多指标融合得到最终的桥梁安全等级。 最后，将论文提出的桥梁安全评估方法应用于两座桥梁的工程实例中。在混凝土连续梁桥安全评估中，将先验数据库与数据融合处理模块建立自适应融合，分析判断得到桥梁的安全等级。在斜拉桥安全评估中，将现场的应变、挠度、索力、车流量信息等监测数据传输到梁、索、塔、桥面板评估子系统中，对桥梁的安全状况进行评价，为桥梁的运营管理与维修决策提供参考。 关键词：桥梁工程；安全评估；数值分析；现场监测；数据融合

 As the increasing attention of China's bridge aging and bridge health issues, it is urgent to establish a scientific and effective bridge safety evaluation system. At present, the evaluation results are greatly influenced by uncertain factors, such as monitoring equipment, environmental noise and vehicle load. The data of numerical analysis, health monitoring and visual inspection are not effectively integrated. Therefore, the application of artificial intelligence technology and data fusion theory to establish a scientific and effective bridge safety assessment system, in order to achieve the more definitive and reliable results of multi-source evaluation indexes fusion, is of great significance for bridge management and maintenance. Firstly, common bridge diseases and their generation mechanism are analyzed from three aspects: environmental corrosion, material aging and external load. The degradation rules of mechanical indexes under the influences of multi-source factors such as concrete carbonation, freeze-thaw cycles, steel corrosion, temperature, wind and vehicle load are studied, then the variation interval of each influence factor is obtained and applied to the numerical model, which is regarded as the basis of prior database. Secondly, in the damage identification method based on bridge dynamic fingerprints, Bayesian fusion is applied to identify the damage locations. A novelty detection method of rough set theory combined with Naive Bayes classifier is proposed for damage extent detection. It is proved that the remaining features after attribute reduction by rough set can satisfy the independence requirement, thus removing the rigorous limitations of Naive-Bayes. Ambient excitation modal test is applied to a concrete continuous bridge. Frequencies and mode shapes of the bridge are obtained by modal analysis and parameter identification. The damage location and extent of the bridge is judged by the damage identification method. Compared with the investigation results of the girder crack, the applicable conditions of the damage detection method are analyzed. Thirdly, the bridge safety assessment system consists of prior database and data processing module. The calculation results of the various loadcases under the influences of multi-source factors are used as the prior database. The monitoring data and visual inspection results are transmitted to the data processing module. Information entropy, fuzzy membership function and fuzzy analytical hierarchy method is introduced to determine the variable weight. Dempster-Shafer theory is utilized to fuse various evaluation indexes for obtaining the final safety grade. Finally, the bridge safety assessment system proposed in this paper is applied to two bridges. In the concrete continuous bridge safety assessment system, the prior database and the data processing module are adaptively fused. Then the safety grade of the bridge is finially analyzed and judged. In the cable-stayed bridge safety assessment system, the stress, defection, cable force and traffic flow are transmitted to the beam, cable, tower and bridge deck sub-system. Then the safety condition of the bridge is evaluated, which provides a reference for the management and maintenance of the bridge. Key words: bridge engineering; safety assessment; numerical analysis; health monitoring; data fusion