发布时间 : 星期日 文章基于神经网络算法的大型刚构拱桥有限元模型修正 - 图文更新完毕开始阅读
武汉理工大学硕士学位论文
摘要
襄阳汉江五桥是典型的大跨度连续刚构拱桥，拱肋与箱梁结合部位构造复杂。大桥是襄阳市内环线的控制性工程，利用成桥荷载试验结果进行有限元模型修正对长期健康监测具有重要价值。本文以襄阳汉江五桥为工程背景，首先研究分析了其整体有限元模型建立中几个关键因素的影响，并对拱脚的构造复杂部位进行了局部有限元分析。在此基础上利用成桥荷载试验结果和人工神经网络方法对初始有限元模型进行修正，重点对拱桥拱脚部位的刚域合理尺寸及结构的重要物理参数杨氏模量的取值进行了分析研究，最终获得了合理可靠的有限元模型。以下为本文具体开展的工作：
（1）在论文前半部分，主要针对襄阳汉江五桥整体有限元模型的合理成桥状态进行了研究。讨论了拱肋采用梁单元或板单元、成桥吊杆力优化、桩土效应等的影响。并对拱脚有限元模型的局部应力进行了细部的分析，对比ANAYS与MIDAS软件计算结果的差异，对可修正因素做出探讨，为后期的模型修正奠定基础。
（2）经过对襄阳汉江五桥拱脚区域节点刚度的分析研究，结果表明在连续刚构组合体系拱桥中，梁、拱交汇处存在着明显的刚域效应。如果未将其纳入基于荷载试验的有限元模型修正考虑范围，将会使结构的荷载效验系数产生较大的误差。现有的建筑规范对于刚域的规定尚不明确，对于连续刚构拱桥的刚域合理长度的确定需要做具体的分析研究。论文利用人工神经网络强大的自适应性学习能力，在有限元模型设置不同刚域尺寸的情况下，训练得到一组较佳的刚域尺寸，按此刚域尺寸修正的有限元模型的计算结果与静载试验实测数据比较吻合。
（3）考虑刚域效应后多数传感器测点的挠度效验系数提高至0.7~0.8内，还未得到与工程实际结构高度吻合的有限元模型。作为结构的重要物理参数，杨氏模量的真实值会与规范设计值有一定的偏差。由于施工原因、材料老化、外部环境等影响因素，混凝土、拱肋与预应力束的杨氏模量均需要修正。基于灵敏度考虑，将混凝土、钢拱肋及预应力束的弹性模量纳入模型修正的考虑范围，与刚域尺寸同时作为可调结构参数。将弹性模量作为模型修正的主考虑因素，
1
武汉理工大学硕士学位论文
将刚域长度作为副考虑因素，采用均匀设计的方法进行参数组合，利用人工神经网络对数据进行训练，在其中挑选出一组最优组合，基于此组最优组合参数的襄阳汉江五桥有限元模型计算的挠度与成桥静载试验实测值更加接近，计算的前五阶固有频率与成桥脉动试验的实测值吻合较好，振型能够完全对应，表明获得了最符合襄阳汉江五桥实际结构性能的有限元模型，可作为运营期长期健康监测系统的基准有限元模型。
关键词：连续刚构拱桥，荷载试验，有限元模型修正，刚域，局部有限元分析
2
武汉理工大学硕士学位论文
Abstract
Xiangyang Hanjiang Bridge Five is a typical long-span continuous rigid frame bridge, the structure of the arch rib and box beam combining parts is complex. As the controlling project of Xiangyang inner ring, it is of great value to use the load test results to make an updating of the finite element model in the long-term health monitoring. This paper systematically introduces the influences of several key factors in establishing the whole finite element model, and makes a local finite element analysis on the complex parts of the arch foot. On this basis, using the load test results and the artificial neural network method a correction of the initial finite element model is made. Then we mainly analyze and discuss the reasonable rigid zone and structural physical parameters at the stewback of arch bridge. The influence of Young?s modulus on the deflection in static loading test is analyzed. Using the research results above in the fluctuation test combined with the boundary condition factors, the datum of static load test of the artificial neural network are reasonably corrected, finally a more reasonable and reliable finite element model has been got. The following is the specific work carried out in this article:
(1) In the first half of the thesis, the reasonable finished state of the whole finite element model of Xiangyang Hanjiang Bridge Five is studied, the impact of the usage of beam elements or plate elements in stewback, pile-soil effect, cable force optimization are discussed, and local stress of ANSYS finite element model is analyzed in detail. By comparing the differences between the calculating results of ANSYS and MIDAS software, some modifiable factors are discussed. The work sets up a good foundation for the later model correction.
(2)The analysis and study on arch foot region?s joint stiffness of Xiangyang Hanjiang Bridge Five show that an influence of nodal rigid zone exists at the confluence of beam and arch in continuous rigid frame arch bridges. If it has not been included in the data correction consideration of static and dynamic load test, the structure?s load effect coefficient will produce a large error. Because of the regulations in the existing ?Building Codes? for rigid zone is not clear, to determine the appropriate length of rigid region in continuous rigid frame arch bridges needs
3
武汉理工大学硕士学位论文
specific analysis. This paper uses artificial neural network?s powerful adaptive learning ability to train a better domain size under the condition of setting up a group of different rigid domain sizes. Under the domain size the calculating results of the corrected finite element model fit the measured datum of the static loading test.
(3) After considering the effect of rigid zone the effect coefficient of most sensors increase to 0.7~0.8. The result has not been highly consistent with the structure of the actual engineering. As an important physical parameter of the structure, the young's modulus of real value and specification design has certain deviation. Due to the construction, material aging, external environment factors, Young's modulus of concrete, arch rib and pre-stressed tendon all have to be modified. Based on the sensitivity, the elastic modulus of concrete, steel arch rib and pre-stressed tendon have been taken into consideration as adjustable parameters of model?s modification with rigid domain size. We take the Young?s modulus as the main consideration factor of model?s modification and rigid domain size as deputy factor. They are combined again using the method of uniform design. Using the artificial neural network the datum have been trained. Finally a group of optimal portfolios have been selected. Based on the group of optimal portfolios, the deflection of the finite element model of Xiangyang Hanjiang River Five is closer to the measured values. The calculated results of the first five order?s natural frequency are in good agreement with the measured values of the bridge?s pulsation test, and
vibration modes can correspond completely. It indicates that we have got the finite
element model which is most consistent with the actual structure of Xiangyang Hanjiang River Five. It can be used as the baseline finite element model of long-term health monitoring system in the bridge?s operation period.
Keywords: continuous rigid-frame arch bridge; load test; finite element model updating; rigid zone; local finite element analysis
4