基于机器视觉的高速铁路基础设施服役状态智能监测理论及方法研究

The intelligent monitoring and assessment of the serviceability of high speed railroad infrastructure is indispensable for rail operation and safety. The machine vision based monitoring and assessment have emerged as one of the effective methods. This research study is aimed at tackling the key challenging problems associated with the use of machine vision based methods, including the development of referencing moving platform of high precision, accurate measurement of deformation parameters of engineering structures, efficient recognition of images of structure surface defects, and the engineering application of the monitoring information from machine vision. First, the relations between machine vision based key performance indicators of engineering structures (static and kinematic deformation, surface defects, etc.) and degradation mechanisms of corresponding in-service performances were analyzed. The theory and methods for developing the referencing position and movement of the moving platform are studied with which the high-precision, machine-vision-based measurement of both static and kinematic geometry parameters of engineering structures can be achieved. The deep learning neural network algorithms are investigated for their use in solving multi-task and multi-flag recognition problems encountered in surface defect monitoring with complex and irregular background noises. The goal is to achieve efficient and accurate image analysis and recognition of surface defects. Finally, both big data analysis and deep learning models are employed to develop in-service performance evaluation and prediction methods by combining machine vision based monitoring information and those from other monitoring sources. The findings of this study can provide technical guidance and assurance for the operational safety and cost-effective maintenance of high speed railroad infrastructure.

高铁基础设施服役状态智能监测对铁路安全有重要意义。视觉监测是高效的基础设施服役状态监测手段，本项目针对其在高速铁路工程应用中必须解决的运动平台高精度基准建立、工程结构变形几何量高精度测量、结构表面损伤图像高效判读、视觉监测信息工程应用等关键难题展开研究。首先分析基于机器视觉的工程结构关键性能指标（静动态变形、表观病害等）与服役性能劣化机理的关系。然后，研究运动平台位置姿态基准建立理论与方法，实现基于运动平台的工程结构静、动态几何量精确视觉测量；研究深度神经网络算法，解决表观病害监测中复杂不规则背景干扰下弱图像信号的多任务多标签识别问题，实现表观病害图像精准识别和分析及预测。最后，运用大数据分析、深度学习模型，研究融合机器视觉等多源监测信息的结构服役性能评估与预测理论与方法。为高速铁路基础设施安全运营与科学维养提供技术保障。

高铁基础设施服役状态智能监测对铁路安全有重要意义，机器视觉是高效的基础设施服役状态监测手段。项目基于机器视觉开展高铁基础设施静动态变形（路基沉降、隧道衬砌变形、桥梁静动态挠度）和表观伤损（裂缝）智能监测理论与方法的系统研究，在动平台高精度摄影测量、精准图像识别和基础设施服役性能智能评估方面取得创新性研究成果。.（1）探明了混凝土梁裂缝宽度、静力挠度和自振频率等为症状指标的演化规律，提出了针对环境激励下土木工程结构模态参数的识别方法，创建了基于抗力-车辆荷载效应双随机过程的在役钢筋混凝土桥梁构件时变可靠度分析与寿命预测方法。.（2）提出了一种无需人工靶点的非接触式多点结构动位移测量方法；发明了基于动平台的高速铁路桥梁静态挠度监测方法，开发了具有位移测量、频谱分析、屏幕自适应、实时数据显示、动态捕捉以及系统重置功能的无目标视觉位移测量系统；建立了基于图像三维重建的隧道变形监测技术。.（3）建立了局部和非局部图像恢复模型，提出了广义特征问题的D2SPP求解方案，实现自然场景的良好视觉感知；创建了基于数字图像技术的高铁隧道衬砌裂缝检测、铁路路基病害识别感知和铁路轨道结构伤损精细化检测和特征分析技术。.项目研究发表SCI 检索论文46篇，EI 检索论文22篇，CSCD检索论文8篇，会议论文3篇；出版学术专著1部；授权国家发明专利17项，实用新型专利19 项，软件著作权5项，另有7项发明专利处于实审阶段；培养博士研究生2人，硕士研究生14人。研究成果已实现了部分转化应用，先后承担了贵南高铁、焦柳铁路等10多项交通基础设施变形、病害检（监测）与维护技术服务工作，相关成果已直接应用于焦柳线黄土岗隧道、施溶溪大桥的病害检测评估，为其加固处置提供了理论与实践指导。

内容获取失败，请点击重试