大数据背景下的桥梁结构健康监测研究现状与展望

结构健康监测（SHM）技术在许多大型桥梁的运营养护管理中均有应用,但已有监测系统积累的海量数据并未被充分解读。为将大数据技术引入到桥梁SHM数据的处理分析中,首先总结大数据的概念和构成要素;然后分析SHM数据的工业大数据属性,梳理桥梁SHM大数据的研究方向;随后综述包括处理技术和分析方法在内的大数据技术在桥梁SHM中的应用现状,在由数据预处理、数据融合、特征工程、模式识别、可视化构成的大数据分析流程中提出SHM大数据研究的需求和应用场景;最后对大数据技术在桥梁SHM中的前景与驱动力进行展望和讨论。结果表明：SHM大数据研究应以结构状态评估为落脚点;大数据处理技术在SHM的系统框架搭建及数据分析能力扩展方面虽已得到较多应用,但其并非SHM大数据研究的重点;SHM数据融合对大数据分析方法有迫切需求,以实现桥梁SHM数据与外观检测等多源异构数据的多层面融合;深度学习、集成学习为结构状态敏感特征的提取提供了新的算法;有监督、无监督机器学习方法结合海量SHM数据将对结构状态评估下的模式识别问题形成更全面的认知;异常识别、相关分析、迁移学习等方法可为实桥SHM损伤识别提供支撑。研究结果可为SHM领域的大数据研究提供参考。

Development and Prospect of Bridge Structural Health Monitoring in the Context of Big Data

The structural health monitoring (SHM) technique has been adopted for use with many large span bridges. However, massive SHM data have not been well interpreted to support structural maintenance and management. In this study, to introduce big data techniques into bridge SHM for data processing and analysis, the concepts and components of big data were first summarized, and data properties and research directions of SHM were then analyzed. Big data techniques applied to bridge SHM were then reviewed according to computing techniques and data analysis methods, based on which the requirements and application scenarios for SHM were presented in a big data analysis pipeline consisting of data preprocessing, data fusion, feature engineering, pattern recognition, and data visualization. Finally, the prospects and motivations of big data techniques in bridge SHM were generalized. The results demonstrate that structural condition assessment should be the primary application scenario for employing big data techniques in SHM. Computing techniques are not the priorities of big data research in SHM even though they do have some applications. Data fusion that aims to fuse multisource and heterogeneous data from SHM system and bridge management system urgently requires support from big data analysis methods. Ensemble learning and deep learning provide new methods to extract features that are sensitive to structural conditions and enable new prospects for SHM research. Supervised and unsupervised learning methods combined with massive monitoring data enable a more comprehensive understanding of pattern recognition problems in SHM than with a small amount of data. Novelty detection, correlation analysis, and transfer learning can be verified in damage detection problems and further developed to support their applications in real bridges. This study is expected to provide a reference for big data research in bridge SHM.