【作者】 刘建民；

【导师】 李清富； 韩菊红；

【作者基本信息】 郑州大学 ， 水工结构工程， 2003， 硕士

【摘要】 桥梁损伤评估与预测是现有桥梁养护、维修和技术改造的必不可少的重要环节，也是国内外学术界和工程界关注的热点问题之一，特别是随着老旧桥梁工程数量的不断增加，这一问题更加引人注目。论文结合桥梁工程的特点，较为深入和系统地研究了公路混凝土桥梁损伤评估中的若干技术问题，主要包括： (1)结合工程调研成果，系统地分析了我国混凝土桥梁结构损伤的原因，为桥梁结构的规划、设计、施工和管理提出了有针对性的建议，同时，也为桥梁结构损伤评估体系的建立和决策分析奠定了一定的基础。 (2)根据桥梁结构的构造特点，将桥梁系统分为多级子系统，从而建立了桥梁损伤评估的层次结构模型。在此基础上，讨论并给出了完整的损伤评价因素评分体系，为混凝土桥梁损伤的神经网络评估奠定了基础。 (3)在桥梁损伤评估层次结构模型的基础上，提出了基于神经网络的桥梁损伤评估方法，以评价指标的评分值作为输入，以损伤评估值作为输出，建立了桥梁损伤多级评估模型，从而得出桥梁损伤的评估结果，并根据给定的标准，确定桥梁的损伤等级。该法通过神经网络对已有的桥梁损伤评估实例的学习，可使神经网络较好地表达评估结果与评价因素之间的关系，从而代替评估群体进行评估。这不仅可以减少评估的工作量，节约人力物力，还可以较好地利用和积累专家经验知识，降低评估过程中人为因素的影响，保证评估结果的客观性。同时，建立的评估系统具有自学习功能，可以在没有专家或缺少专家的条件下，模拟有经验的评估机理，完成对旧桥损伤的评估。实例计算表明，各网络模型花费很少的时间完成对样本的训练后，便可利用训练好的隐含层权值与阈值对实际桥梁进行评估。 (4)考虑到工程资料少、管理水平不一的实际困难，建议了基于专家知识的加权平均法、模糊综合评判法、模糊广义量化法和集对论方法等四种损伤评估方法，作为桥梁损伤神经网络评估方法的有益补充，这些方法均能从一定程度上消除由于专家个人偏好、知识层次不同等造成的差异，且一般不涉及复杂的数学或力学计算，只需根据经验进行一些简单推理和比较就能得到结果。实例计算证明，所得结论较符合工程实际。 (5)寿命作为桥梁结构的一个重要特征，在损伤评估中十分必要。为此，论文给出了桥梁结构剩余寿命预测的四种模型，即回归预测模型、灰色预测模型、改进灰色模型和神经网络预测模型。决策者可根据自己的知识水平、所掌握的信息种类和数量等主客观条件来选用不同的模型进行预测分析和决策制定。 (6)维修是桥梁使用管理过程中的重要环节，建立合理的维修制度对保证结构的正常运行、提高结构的使用效益具有重要的意义。作为初步分析，该项研究应用马尔柯夫过程模型建立了描述公路桥梁结构性能变化的状态方程。在此基础上，讨论了维修对桥梁结构性能及其使用效益的影响，以结构在使用期内的总期望效益最大为目标，建立了确定结构最优维修策略的数学模型，该模型可为桥梁工程的科学化管理提供一定的参考。更多还原

【Abstract】 The assessment and prediction of damage in bridges is a necessary part in bridge conservation maintenance and technical reconstruction of today, which is also one of the most heated problems that are being discussed by scientists in engineering area and scholars at home and abroad, and along with the increment of old bridges, the problem attracts more attention. Combining with the character of bridges, the thesis gives a systematic and thorough study on some technical problems of assessment of damage in concrete bridges, which includes:(1) Integrating with the research results in engineering, a systematic analysis for the cause of damaged concrete bridges in our country was given in the thesis, which will consequently lead to some pertinent proposals towards plan, design , construction and administration of bridges as well as lay a foundation for the establishment and decision-making of the assessment system of damaged bridges.(2) According to its character, bridge system can be divided into several multilevel-systems, and thus a hierarchy assessment model is established. On the basis of these research works, a comprehensive grading system for assessment factors is discussed and developed, being prepared for the neural network assessment of damaged concrete bridges.(3) Based on the hierarchy assessment model, a neural network method to assess the damage in bridges is put forward. The scores of assessment factors being made the input value and the score of damage assessment the output value, a multilevel assessment model for bridges came into being, which in turn could come to the final result of the bridge as a whole and get the damage rating according to the given standard. The method, making use of neural network to learn from exiting assessment examples, could undergo assessment instead of expert group with the relationship between assessment result and assessment factors best expressed in the network. As a result, it could not only reduce assessment workload and save resources, but also accumulate experts’ experience and get rid of human factors in the course of assessment to make the result sound objective. Simultaneously, the established system is characterized by learning of itself, which can complete assessment through simulating experienced assessment mechanism when there is no expert or has not got enough experts. To conclude from the given examples, each model, after spending little time on training themselves from sample data, could assess the damage degree for existing bridges using trained weighted values and thresholds.(4) Considering such actual difficulties as lacking of engineering data and discrepancy in managing level, four methods for damage assessment are proposed in the thesis based on expert knowledge, which are weighted average method , fuzzy comprehensive evaluation method, generalized quantification method and set pair analysis method to make useful supplements to the neural network method. Disadvantageous effects, caused by dissimilarity of subjective factors such as everyAbstractexpert’s knowledge , level, discernment and emphasis on problem, are eliminated to some extent in each method. Furthermore, there are not any complicate mathematic or mechanical calculations, and conclusion can be easily got only by some simple reasoning and comparing from experience. Given examples indicate that it’s conclusion accord with the engineering practice quite well.(5) Being a very important factor of concrete bridges, remaining life become much more crucial in process of damage assessment. Accordingly, the thesis proposes four models for predicting the remaining life of concrete bridges, which are regression prediction model, grey prediction model, improved grey model and neural network model. Decision-makers can choose different models in terms of subjective conditions, including knowledge level as well as the quantity and kinds of information mastered, to carry thorough prediction analysis and decision making.(6) Maintenance is a significant part更多还原