基于函数型数据分析的监测数据分布相关性挖掘与健康诊断方法

Structural health monitoring (SHM) has arisen as an important basis for managing and maintaining bridges. Long-term online monitoring produces huge amounts of data; data processing, information interpretation and knowledge discovery have become new challenges of SHM. Aiming to the problem of big data processing in SHM and breaking through the traditional scalar data mining category, this proposal is to study correlation mining methods for probability distributions of SHM data as well as its application in bridge health diagnoses. Firstly, basic correlation mining methods for probability distributions of SHM data are studied based on functional data analysis, including principal correlation mode mining of distribution mappings, time-space evolution mining of distributions and their mappings, overall structural breaking as well as abnormity mining of distribution mappings. Then, the monitoring data of strain and displacement collected by real bridge SHM systems are selected for study, the proposed methods are utilized in mining the distributional correlations of such data, revealing the spatial interactions of multipoint monitoring data as well as corresponding mechanical mechanisms. Finally, structural condition diagnosis methods (including diagnostic indices) are studied on the basis of the revealed distributional correlations of monitoring data. This study will develop new methods for analyzing and mining SHM data as well as new structural health diagnosis methods, deeply reveal the mechanical behaviors contained in monitoring data for bridge structures, enrich the theory of data science of structural health monitoring, which has important academic and application prospect.

结构健康监测已成为桥梁管理与维护的重要基础。长期在线监测产生了海量数据，数据处理、信息解释和知识发现是结构健康监测领域的新挑战。面向结构健康监测大数据处理问题，突破传统标量数据挖掘范畴，研究监测数据概率分布相关性挖掘方法及其在桥梁健康诊断中的应用。首先，基于函数型数据分析研究概率分布相关性挖掘的基础数据挖掘方法，包括概率分布主相关映射模式挖掘方法、概率分布及其映射关系时-空演化规律挖掘方法、概率分布映射关系整体突变及异常映射关系挖掘方法；其次，研究实际桥梁结构应变、位移等监测数据概率分布的关联规律，揭示桥梁结构空间多点监测数据相互作用及其力学机理；最后，研究基于桥梁结构空间多点监测数据分布相关性变化的结构状态诊断方法和指标。本研究将发展新的结构健康监测数据分析与挖掘及健康诊断方法，深刻揭示数据蕴含的结构力学行为，丰富和发展结构健康监测数据科学理论，具有重要的学术价值与应用前景。

结构健康监测已成为桥梁管理与维护的重要基础。长期在线监测产生了海量数据，数据处理、信息解释和知识发现是结构健康监测领域的新挑战。针对结构健康监测复杂数据处理与分析问题，本项目突破传统欧式空间数据挖掘范畴，研究了监测数据概率分布相关性挖掘方法及其在桥梁健康诊断中的应用。首先，从统计学角度基于函数型数据分析理论研究了以概率分布为数据对象的分布型数据统计分析方法，包括概率分布主相关模式分析方法、分布回归分析方法、分布时-空相关演化规律挖掘方法、分布异常与突变检测方法等。然后，结合实际大跨度桥梁的监测数据，进一步发展了基于分布型数据分析的结构健康状态诊断方法。本研究发展了新的分布型数据统计分析方法与结构健康诊断方法，揭示了结构健康监测数据中蕴含的复杂分布相关信息，具有重要科学与实际应用价值。本项目相关成果发表高水平SCI论文3篇，包括国际工业统计顶级期刊Technometrics 1篇、国际结构健康监测顶级期刊论文2篇。在本项目资助下，培养毕业硕士2名、在读博士1名、在读硕士1名。