【作者】 冯丹；

【导师】 郑君君；

【作者基本信息】 武汉大学 ， 管理科学与工程， 2023， 博士

【摘要】 随着以人工智能为主的新一代信息技术的蓬勃发展,大数据、高性能计算和深度学习等现代技术科学正逐渐被应用到桥梁工程领域。而相应地,将桥梁全生命周期与现代化智能理论、方法及技术深度融合,实现桥梁建造与管理的智能化、高精度化逐渐成为学界和实践界致力目标。近十几年来,我国大跨度桥梁的建设已经进入了辉煌的发展时期。该类工程通常具有体量大、线路长、地质条件与施工环境复杂、质量要求高及安全管理控制严等特点,是衡量一个国家桥梁技术水平的重要体现。而目前,“智能桥梁”在国内发展仍处于起步阶段,未能满足实际工程的需求。基于此,充分利用智能优化理论、方法及新兴技术,进一步提升大跨度桥梁建造的信息化及智能化,以确保工程高效、安全及经济施工具有重要意义。本文针对大跨度桥梁工程项目的施工安全风险管理,以“风险识别-风险分析-风险应对-风险监测”为研究主链,结合机器学习、自然语言处理及本体技术等新型方法技术,对大跨度桥梁施工过程中潜在的事故风险进行系统的挖掘、剖析、评估与应对。本文依托桥梁施工风险管理全过程架构,借鉴性地引入人工智能方法及技术,来拓展大跨度桥梁施工安全风险智能管控研究,以期为有效破解桥梁施工安全“管控难”等问题提供些许参考与思路。论文主要开展了以下几个方面的研究:(1)研究基于深度学习与自然语言处理,构建了一种自动信息提取的小样本训练框架。该训练框架能有效克服手动标注成本高等缺陷,实现桥梁施工安全风险信息的自动提取,为下一步的桥梁施工安全风险评估提供技术支撑。首先,基于深度神经网络、集成字符语义编码系统,建立信息自动提取模型;进而,充分利用标注良好的小样本,并通过交叉和组合操作扩充数据集,训练出可靠、鲁棒的深度神经网络;最后,以事故新闻报道数据集为例进行研究,验证所建立的自动信息提取模型的性能,并对所提出的小样本训练框架进行了评价。(2)在系统梳理大跨度桥梁施工过程中安全风险源的基础上,对各影响因素进行多层级划分,以便明晰各因素之间的层级关系。随之,以桥梁坍塌失稳为例,构建了基于贝叶斯网络技术的桥梁施工安全事故致因机理分析模型,对桥梁安全事故进行致因分析。剖析安全行为层次各因素的主要诱发路径,明晰大跨度桥梁施工安全事故的高频因素和故障链条,增强忧患意识,以提高事故防范能力,避免此类事故的发生。(3)着眼于已有桥梁风险评估多数研究以专家的主观判断为主,难以满足桥梁实际操作的现象,提出了一种经典的智能优化算法—粒子群优化BP神经网络方法,来定量地评估大跨度桥梁施工风险。相关结果证实了训练后的模型在大跨度桥梁的施工安全风险评估上具有有效性,为后面的内容奠定基础。(4)基于BIM技术与本体技术对桥梁施工安全风险实现智能化、可视化协同管理,能有效提高桥梁施工安全风险应对的信息化水平与安全状态预控。首先,构建了蕴含丰富知识的桥梁施工安全风险管理体系;其次,将本体引入到桥梁施工安全风险管理,实现桥梁施工安全风险相关知识的共享、复用与积累;最后,基于桥梁施工安全风险管理知识与二次开发后的Revit软件,实现施工安全风险智能识别与安全状态预控。(5)梳理桥梁施工过程中安全监测流程,分析了一般风险监测的方法。在此基础上,建立了一种以随机场模型在多个尺度上的结构动力学的概率表征框架,并提出了一种基于元素平均的随机场参数评估方法。该框架为后续案例研究中的桥梁安全监测奠定理论基础。最后,本文以塞内加尔方久尼大桥为例进行大跨度桥梁施工安全风险智能管控的分析与研究,验证了论文的研究成果。结果显示:塞内加尔方久尼大桥的施工安全状态基本契合论文所构建的大跨度桥梁施工安全智能化管控的知识框架,案例分析较好支持了该研究成果。概述之,上述研究借鉴性地引入多种智能手段与方法,丰富了大跨度桥梁施工安全风险智能管控的研究成果库,是桥梁智能建造等研究热点的合理深化与有益补充。此外,本研究成果为大跨度桥梁施工安全风险管控注入新活力,对实现桥梁智能建造“增质提效”具有重要的实践意义。更多还原

【Abstract】 With the vigorous development of artificial intelligence,modern technologies,such as big data,high-performance computing and deep learning,are gradually applied to the bridge engineering.Accordingly,it has become the common goal of academia and practice to deeply integrate the bridge life cycle with modern theory and technology,and then achieve the intelligence and high precision of bridge construction and management.Over the past decade,the construction of long-span bridges in China has entered a glorious period.This type of engineering usually has the characteristics of large quantities,long line,complicated geological conditions and construction environment,high quality requirements and strict safety management control,making it an important standard to measure the technical level of a country’s bridges.However,the development of―intelligent bridge‖in China now is still in its infancy,which can not meet the needs of practical projects.In response,it is of great significance to make full use of artificial intelligence and other new information technology to further improve the intelligence of long-span bridge construction and ensure efficient,safe and economical construction of the project.Aiming at the construction safety management and risk management of long-span bridge engineering projects,this research takes―risk identification-risk analysis-risk response-risk monitoring‖as the main research chain,and combines machine learning,natural language processing,ontology technology and other new generation information technologies to systematically analyze and evaluate the potential safety risks in the construction of large-span bridges.And furtherly,based on the whole process system of bridge construction risk management,artificial intelligence methods and technologies were introduced,and the research on intelligent risk management and control of long-span bridge construction was expanded,which provides reference and ideas for effectively solving the problems of"difficult management and control"of bridge construction safety.The research of this subject mainly includes the following aspects:(1)Based on deep learning and natural language processing,a small sample training framework for automatic information extraction is constructed.This training framework can effectively overcome the high cost of manual marking,realize the automatic extraction of bridge construction risk information,and provide technical support for the next step of bridge construction safety risk assessment.First,the automatic information extraction model was established based on deep neural network and integrated character semantic coding system.Second,the well-labeled small samples are use to expand the data set through crossover and combination operations.And a reliable and robust deep neural network was trained.Third,taking the data set of accident news reports as an example,the performance of the established automatic information extraction model is verified,and the proposed training framework with small samples is evaluated.(2)On the basis of systematically sorting out the safety risk sources in the process of long-span bridge construction,the influencing factors are divided into multiple levels in order to clarify their relationship.Taking bridge collapse and instability as an example,Bayesian network is used to analyze the causes of bridge safety accidents.This can enhance the sense of urgency,improve the ability of accident prevention and avoid such accidents.(3)Given that most of the existing researches on bridge risk assessment are based on experts’subjective judgment,which is difficult to meet the actual operation of bridges,a classical intelligent optimization algorithm-particle swarm optimization BP neural network method is adopted to quantitatively evaluate the construction risk of long-span bridges.Results confirm that the trained model is effective in the construction safety risk assessment of long-span bridges,which lays the foundation for the following contents.(4)BIM technology and ontology technology were adopted to realize self-intelligent and visual collaborative management of bridge construction safety risks,which can effectively improve the information level and safety state pre-control of bridge construction safety risks.First,the bridge construction safety risk management system with rich knowledge is built.Second,Ontology is introduced into bridge construction safety risk management to realize the sharing,reuse and accumulation of knowledge.Third,based on the knowledge of bridge construction safety risk management,and through the secondary development of Revit software,the intelligent identification of construction safety risk and pre-control of safety status are realized.(5)The process of safety monitoring in bridge construction is sorted out,and the general risk monitoring methods are analyzed.A probabilistic representation framework of structural dynamics based on random field model at multiple scales is established,and a random field parameter evaluation method based on element average is proposed.The results lay a theoretical foundation for bridge safety monitoring in subsequent case studies.Taking Fangjiuni Bridge in Senegal as an example,this research analyzes the intelligent management and control of safety risks in large-scale bridge construction.The results of the case study well support the knowledge framework of intelligent management and control of long-span bridge construction safety built before.In short,the findings introduce a variety of intelligent means and methods,which enriches the research pool of intelligent management and control of safety risks in long-span bridge construction,and is a reasonable deepening and beneficial supplement to the research hotspots.Besides,the findings inject new vitality into the safety risk management and control of long-span bridge construction,and are also of practical significance to realize the"quality improvement and efficiency improvement"of bridge intelligent construction.更多还原