【作者】 李松辉；

【导师】 练继建；

【作者基本信息】 天津大学 ， 水利水电工程， 2008， 博士

【摘要】 基于结构模态参数识别理论的损伤诊断技术是现阶段水利水电工程研究中的热点问题之一。水工结构在实际运营中,由于设计、施工等先天缺陷或者使用载荷超出设计以及遭受强大的突加外在荷载(如地震作用等)的作用会使结构物出现不同程度的损伤,结构发生损伤以后将严重影响结构的承载力及耐久性,甚至会发生严重的工程事故,不仅造成重大的人员伤亡和经济损失,而且还会产生极坏的社会影响。因此,为了保证结构的安全性、完整性和耐久性,需采用有效的手段对结构的健康状态进行诊断。常规的一些结构损伤诊断方法,由于其本身的缺陷而不适用于水利水电工程。鉴于水工结构流激振动响应的获取较为容易,且不会对结构产生不良影响,利用结构在流激振动下的实测响应对结构进行模态参数识别,进而借助于机器学习理论强大的学习功能对结构的损伤进行诊断无疑是一种很好的方法。本文针对流激振动下基于机器学习和模态参数识别理论的水工结构损伤诊断方法进行了研究,主要得到以下创新性的研究成果:(1)提出了泄流振动下基于带通滤波的水工结构模态参数识别方法。即通过分析结构的特点和结构自振频率的范围,首先应用带通滤波的方法对信号进行预处理,提取出信号中感兴趣的部分,然后应用时域方法进行识别。工程实例表明:该种方法能够很好的对流激振动下水工结构的自振频率进行识别。(2)提出了水工结构模态参数的遗传识别方法。借助于遗传算法强大的全局寻优能力,将遗传算法应用到水工结构模态参数识别中来。由于该方法以结构的噪声为研究对象,所以其在提高结构自振频率识别精度的同时,在一定程度上也提高了阻尼比的识别精度。最后,通过工程实例表明了这种算法的有效性。(3)提出了基于机器学习理论和模态分析的水工结构损伤诊断方法。在结构模态参数识别的基础上,通过有限元计算各种损伤状态下的耦合动力特性,以构造支持向量机(SVM)学习的样本库,建立基于支持向量机(SVM)的结构损伤定位及损伤程度诊断方法。即通过支持向量机的二次优化问题求解,以保证小样本情况下机器学习得到的解是全局最优解,避免了人工神经网络等方法的网络结构难于确定、过学习、欠学习以及局部最小化等问题。以最小二乘线性系统作为损失函数,代替传统的支持向量机采用的二次规划方法。较成功地解决了泄流结构(特别是水下部位)损伤定位和损伤定量难的问题。(4)流激振动下基于水工结构模态参数识别的损伤诊断方法能够有效地节约资源,且识别结果可靠。本文首次提出将其应用到水利水电工程中来。以青铜峡坝体结构三大条贯穿性裂缝为研究对象,通过对现场采集信号的分析、模态参数识别和有限元计算,得出在裂缝深度为自基础向上22m左右,最后,就结构损伤发生后的安全性进行了分析。更多还原

【Abstract】 The structural damage diagnosis based on the modal parameter identification is one of issues nowadays in hydraulic structure. The hydraulic structure in the actual operations will be to some extent damaged caused by the design, construction, and the other defects or the load in excess of the design and the sudden powerful external load (such as earthquake, etc.). After happening in a certain degree, the structure damage will have a bad effect on the load capacity and durability greatly or even cause the serious accidents. Thus, it caused not only the significant casualties and economic losses but also a very bad social influence. Therefore, in order to ensure the structural safety, integrity and durability of the structure, some effective measures should be taken to fulfill the health diagnosis. Some conventional structural damage diagnostic methods, due to its own shortcomings, are not fit for the hydropower project. Whereas the response of the structure is easy to be gained under flow-induced vibration and has no negative impacts on the structure, it is no doubt that the modal parameter identification in the use of flow-induced vibration is a good method with the help of the machine learning theory. With the theory of the machine learning and the modal parameter identification and diagnostic methods under flow-induced vibration hydropower structure taken into research, the paper involves the following major research results of innovations:(1) It puts forward the modal parameter identification methods based on band-pass filter and discharge vibration, i.e., according to the analysis of the structural characteristics and the natural frequency, the signals are preprocessed by the band-pass filter. After extracted the interesting signals, the time domain method is used to identify the modal parameter. The results show that the methods can efficiently identify the structure natural frequency in the vibration on the discharge.(2) The genetic algorithm is applied in the modal parameters identification of the hydraulic structure, considering its global optimization. The genetic identification method used in hydraulic structure modal parameters is come up with. Because the measurement noise is taken into the research, not only is the identification accuracy of natural frequencies improved, but also the damping ratio’s identification accuracy does to a certain extent. Finally, the result shows that the efficiency of the structure’s natural frequency and damping ration is tested.(3) The damage identification method is proposed based on the machine learning theory and modal analysis of the guide wall structural. On the basis of the modal parameters identification, the coupling dynamic characteristics of the structure are calculated by the finite element method so as to construct the sample of the Support Vector Machine (SVM). The damage diagnoses method including the damage position and based on the establishment of Support Vector Machine (SVM) and the damage ration. That is to say, the Support Vector Machine can ensure the global optimal solution in the small sample of events through its quadratic optimization problem solved. It overcomes such the shortcomings caused by the artificial neural network methods as the difficulties in defining the network structure, the over fitting, under fitting and local minimum. Instead of the traditional support vector machines which use quadratic programming method, least-squares linear system is used as the loss function. Thus the problems such as damage positioning and damage degrees are solved successfully in the discharge structure (especially the underwater part).(4) Some of the main advantages of the damage diagnoses method, based on the discharge vibration and parameter identification, over other ones is that it is effectively helpful to save resources and get reliable recognition results. The paper first proposes damage diagnosis method based on the dynamic parameters identification and applies it to the hydraulic structure. Taking the three cross crack in Qing Tongxia as the research object and making the comprehensive analysis of signal collecting, modal identification and the finite element calculation, it concludes that the cross crack is 22m. Finally, the analysis of the safety is performed after the structure damages occur.更多还原