桥梁结构损伤识别的模式分类和聚类识别方法研究

发布时间：2018-07-20 10:53

【摘要】：桥梁这项建筑工程量大、造价高且在交通经济中重要，为确保其在运营期间符合生命安全标准，因此桥梁结构损伤的早期识别和评估是必要的。桥梁损伤位置和程度识别是这项研究的核心。当模态特征未知时，模式识别被广泛的用于此项研究，是一种用来帮助进行损伤识别的典型方法。模式识别方法是具备隐性识别相关和不相关变量的非线性关系的能力，具有自学习和容错能力。这些优势让它能够恰当的最小化在响应测度和有限元模型结构的负面影响。对于桥梁这样的大型结构，我们显然不能去了解每一个个体，也就是说每一个实测部位的状态，而是通过模式识别方法有效地实现对桥梁的损伤进行快速、精确和智能化识别，从而保障桥梁这样的重大工程结构的安全性、完整性、适用性与耐久性。目前还鲜有利用一系列模式识别算法进行系统的桥梁静力和动力损伤识别研究，因此全文作了该方面系统的研究工作如下： 1、用模式识别对桥梁进行损伤识别的前提是数据的真实性，但是这些数据是海量的，其有效性难以通过一些常规手段进行检验。传感器最优布置的建模方式和智能算法的选取是解决问题的关键，因此本论文中传感器的最优化布置问题将从这两个方面入手解决。一是建立了以模态振型为随机变量的传感器优化布置单目标和多目标整数规划期望值模型，二是利用DNA遗传算法求解此类问题的优点，并设计了求解算法，最后通过徐葛大桥为实例验证了算法的可行性和有效性。 2、利用SVM（支持向量机）这种模式识别方法做桥梁静力损伤识别，关键在于利用ANSYS进行数值模拟构造无损伤和损伤时的训练集与实际工程的相似性，以表现其抗干扰能力；用带有噪声的测试集去判定损伤位置和损伤程度的精确度以表现其对损伤的区分能力。针对这个问题，，本论文一是详细给出了不同噪声情况和不同工况加载模式下的挠度响应的高精度模式识别结果，二是利用专业数据挖掘软件WEKA作对比分析，证明了本方法的有效性。正确进行桥梁静力损伤识别后，另一个问题是加载模式的识别。本论文将轮廓系数应用到桥梁静力加载模式的识别当中，结果表明有很好的识别效果，并且具有一定的实际应用价值。利用SVM（支持向量机）这种模式识别方法做大型桥梁频域损伤识别，关键在于损伤节点和单元的选择和计算识别的精度与抗噪性。针对这两个问题，本文首先根据前文传感器优化布置点作为损伤识别对象，然后利用SVM模式识别方法进行损伤位置和程度识别的精确识别和噪声检验，最后利用专业数据挖掘软件WEKA作对比分析，证明了本方法具有一定的合理性和优势。在讨论车过桥的时域损伤的模式识别时，首要问题是怎样采用ANSYS软件进行模拟计算，获取相应的损伤指标数据。基于能量比的时域指标可以根据测点采样间隔获取车从上桥到下桥的速度响应，以损伤前后的能量比作为该测点的损伤指标。由此本文提出利用能量比指标的SVM损伤识别方法。 3、采取分步识别的桥梁损伤识别的模式识别方法主要分为两个步骤：损伤位置识别和损伤程度识别，其根本分别就是分类和回归问题。本文提出以SOM神经网络做损伤位置识别聚类分析，RBF神经网络做损伤程度识别回归分析。损伤位置识别是损伤识别的关键一步，在无先验知识的情况下只能进行聚类识别。虽然存在许多的聚类方法，但是没有一个通用的万能的聚类方法，能够适用于所有的聚类问题，聚类集成算法因此被提出并证明能解决更多的问题。本文利用基于Co-occurrence相似度的聚类集成（CSCE）和基于矩阵变换的聚类集成方法去识别桁架架构和徐葛大桥的损伤位置，达到完全识别。本文最后还利用专业数据挖掘软件WEKA作对比分析，证明了本方法的有效性。基于粗糙集的聚类方法能结合集合方法和概率方法计算样本的相似度，具有很好的聚类效果。本文利用粗聚类方法去识别桥梁的损伤位置，达到良好的识别效果，并与模糊聚类（FCM）作了比较。根据该方法还能够得到样本的属性约简结果和约简规则，为进一步研究样本特征提供了参考数据。
[Abstract]:Bridge construction has a large quantity, high cost and important in traffic economy. In order to ensure that the bridge conforms to the life safety standard during operation, it is necessary to identify and evaluate the damage of bridge structure early. The location and degree identification of bridge damage is the core of this research. When the modal characteristics are unknown, pattern recognition is widely used in this study. Item research is a typical method used to assist in the identification of damage. The pattern recognition method is the ability to recognize the nonlinear relation of the related and unrelated variables, and has the ability of self learning and fault tolerance. These advantages make it able to minimize the negative effects on the response measure and the structure of the finite element model. It is obvious that we can not understand each individual, that is, every state of the measured site, but through the pattern recognition method, we can effectively realize the damage of the bridge quickly, accurately and intelligently, so as to guarantee the safety, integrity, applicability and durability of the bridge such as the bridge. A series of pattern recognition algorithms have been used to identify the static and dynamic damage identification of the bridge, so the full text of this system is as follows:
1, the premise of identification of bridge damage by pattern recognition is the authenticity of the data, but the data are massive, its effectiveness is difficult to be tested by some conventional means. The key to solve the problem is the modeling method of the optimal layout of the sensor and the selection of intelligent algorithms. Therefore, the optimization layout of the sensor in this paper is a problem. The first is to solve these two aspects. One is to set up a single objective and multi-objective integer programming expectation model with the modal vibration type as the random variable. Two is the advantage of using the DNA genetic algorithm to solve the problem, and the solution algorithm is designed. Finally, the feasibility and the feasibility of the algorithm are verified by the Xu Ge bridge. Efficiency.
2, the key is to use the SVM (support vector machine) to identify the bridge static damage. The key is to use ANSYS to simulate the similarity between the training set and the actual project without damage and damage, in order to show its anti-interference ability, and to determine the accuracy of the damage position and degree of damage by using a test set with noise. In this paper, the high precision pattern recognition results of the different noise conditions and the deflection responses under different loading modes are given in this paper. Two is a comparative analysis using the professional data mining software WEKA, which proves the validity of this method. Then, another problem is the identification of loading mode. In this paper, the contour coefficient is applied to the identification of bridge static loading mode. The result shows that it has good recognition effect and has a certain practical application value. The key to damage identification of large bridge in frequency domain by using the SVM (support vector machine) pattern recognition method is the damage node. The selection of point and unit and the accuracy and noise resistance of calculation recognition. Aiming at these two problems, first of all, this paper is based on the optimization points of the previous sensor as the damage identification object, and then uses the SVM pattern recognition method to identify the damage location and degree recognition and the noise test. Finally, the comparison of the professional data mining software WEKA is used as a contrast. The analysis shows that this method has a certain rationality and advantages. When discussing the pattern recognition of time domain damage in the vehicle bridge, the first problem is how to use ANSYS software to simulate and obtain the corresponding damage index data. Based on the time domain index of the energy ratio, the speed of the vehicle from the upper bridge to the lower bridge can be obtained by the sampling interval of the measurement point. In response, the energy ratio before and after injury is taken as the damage index of the measuring point. In this paper, a SVM damage identification method using energy ratio index is proposed.
3, the pattern recognition method of bridge damage recognition by step identification is divided into two steps: damage location identification and damage degree recognition, which are classified and regression problems. This paper proposes a SOM neural network for damage location identification clustering analysis, RBF neural network network damage identification regression analysis. Recognition is a key step in the identification of damage, which can only be identified without prior knowledge. Although there are many clustering methods, there is no universal universal clustering method, which can be applied to all clustering problems. Therefore, the clustering integration algorithm has been proposed and proved to be able to solve more problems. This paper is based on Co. -occurrence similarity clustering integration (CSCE) and cluster integration method based on matrix transformation are used to identify the damage location of the truss structure and Xu Ge bridge to complete recognition. In the end, the validity of this method is proved by the comparative analysis of the professional data mining software WEKA. The clustering method based on Rough Sets can combine the collective cube. The method and probability method are used to calculate the similarity of the sample. This paper uses the rough clustering method to identify the damage location of the bridge, and achieves a good recognition effect, and compares it with the fuzzy clustering (FCM). According to this method, the results of the reduction and reduction of the sample are also obtained, which can be used for the further study of the sample characteristics. For reference data.
【学位授予单位】：武汉理工大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：U446

【参考文献】