【作者】 张波；

【导师】 郭惠勇；

【作者基本信息】 重庆大学 ， 土木工程， 2024， 硕士

【摘要】 目前,建筑结构在运维期间易受到长期荷载效应、材料老化等因素的影响,导致结构服役性能降低。因此,有必要建立全周期的结构健康评价体系对结构进行健康监测。结构健康评价作为结构健康监测领域重要的组成部分,已发展出众多的结构健康评价方法。然而随着建筑结构形式的日益复杂,传统方法已难以满足现有的监测需求。针对传统健康评价方法难以客观解决复杂工程问题,提出一种基于莱维引力搜索优化极限学习机(Lévy Gravitational Search-Extreme Learning Machine,LGS-ELM)模型的结构健康评价方法,并对钢梁、混凝土T梁构件和钢框架结构进行试验与仿真研究,实现从构件到整体结构的健康评价,本文主要研究内容如下:(1)本文针对传统神经网络计算效率低的缺点,提出了一种简单高效的LGS-ELM模型。通过在引力搜索优化(Gravitational Search Algorithm,GSA)中嵌入Lévy更新策略,增强GSA的全局搜索能力,提出莱维引力搜索算法(LGS)。为提高ELM模型的稳定性,本文采用LGS对模型的输入权值和隐藏层偏置进行调优,从而建立了LGS-ELM模型,奠定了本文的模型基础。(2)本文提出了一种基于LGS-ELM模型的结构健康评价方法,并从评价指标体系的建立、结构健康等级划分、模型网络参数的设置以及评价流程等方面进行详细阐述。对钢梁进行有限元建模,获取不同损伤工况下各指标数据,建立起损伤指标与钢梁健康状态的映射关系,进而实现未知工况下钢梁健康状态评价。为说明该方法在材料非均匀性较强的钢筋混凝土结构上的评价性能,本文以钢筋混凝土T型梁损伤试验为例进行方法验证,结果表明本文所提方法具有优异的评价性能、抗噪性能以及普适性。(3)本文针对实际结构监测指标多,模型易出现维数灾难,进而导致计算效率和准确率降低的问题,结合证据网络(Evidence Networks,EN)正向推理特性,提出了一种基于LGS-ELM-EN模型的结构健康评价方法。所提方法将整体结构划分为若干子结构,分配LGS-ELM子模块对不同子结构进行评价,通过输出概率化改造获得初始基本概率分配,利用证据网络进行综合决策判定整体结构健康状态。通过对两层单榀钢框架低周往复试验进行研究,将钢框架健康状态划分为五个等级,分析其在损伤过程中局部和整体性态指标与结构整体健康状态的映射关系,采用本文方法对钢框架健康状态进行评价,结果表明所提方法具有良好的评价性能,并且在更为复杂的结构健康评价中具有巨大的评价潜力。更多还原

【Abstract】 Recently,the building structures are susceptible to long-term loading effects,material ageing and other factors during operation and maintenance,resulting in reduced structural service performance.Therefore,it is necessary to establish a full-cycle structural health evaluation system for structural health monitoring.As an important part of the structural health monitoring field,numerous structural health evaluation methods have been developed.However,with the increasing complexity of building structural forms,the traditional methods have been difficult to satisfy the existing monitoring needs.This thesis proposes a structural health evaluation method based on the Lévy Gravitational Search-Extreme Learning Machine(LGS-ELM)model in response to the difficulty of traditional health evaluation methods to objectively solve complex engineering problems,and conducts experimental and simulation studies on steel beam,concrete T-beam members and steel frame structures.It realizes health evaluation from components to the integral structure.The main research contents of this thesis are as follows:(1)This thesis proposes a simple and efficient LGS-ELM model for the disadvantage of low computational efficiency of traditional neural networks.The Lévy Gravitational Search(LGS)is proposed by embedding the Lévy update strategy in the Gravitational Search Algorithm(GSA)to enhance the global search capability of the GSA.To improve the stability of the ELM model,this thesis uses LGS to optimize the input weight and hidden layer bias of the model,establishing the LGS-ELM model and laying the model foundation of this thesis.(2)This thesis proposes a structural health evaluation method based on the LGS-ELM model.It also elaborates on the establishment of the evaluation index system,the classification of structural health class,the setting of model network parameters and the evaluation process.The finite element modelling of steel beams is carried out to obtain the data of each index under different damage conditions,and establish the mapping relationship between the damage indexes and the health state of steel beams.It realizes the evaluation of the health state of steel beams under unknown conditions.To illustrate the evaluation performance of the method on reinforced concrete structures with high material non-uniformity,this thesis conducts a reinforced concrete T-beam damage test as an example for method validation.The results show that the proposed method has excellent evaluation performance,noise resistance,and universality.(3)This thesis proposes a structural health evaluation method based on the LGS-ELM-EN model by combining the forward inference property of Evidence Networks(EN).It is dedicated to solving the problem of reduced computational efficiency and accuracy with many actual structural monitoring indicators,and the model is prone to dimensional catastrophe.The proposed method divides the integral structure into substructures and assigns LGS-ELM sub-modules to evaluate different substructures.The initial basic probability assignments are obtained by probabilistic transformation of the model outputs,and the integral structural health status is determined by comprehensive decision making with evidence networks.This thesis investigates the two-story single-bay steel frame in low-cycle reciprocating tests,classifies the health state of the steel frame into five grades,and analyses the mapping relationship between the local and global state indexes and the overall health state of the structure during the damage process.The results show that the proposed method has excellent performance in evaluating the health state of steel frames and has great potential in evaluating the health state of more complex structures.更多还原