基于动态响应的简支梁桥移动荷载识别研究

发布时间：2018-02-13 09:55

本文关键词： 移动荷载识别 BP神经网络动力响应模型试验　出处：《内蒙古科技大学》2014年硕士论文　论文类型：学位论文

【摘要】：桥梁的移动荷载识别是桥梁结构健康监测的重要环节，获得精确可靠的荷载数据可以对桥梁设计中选用的荷载进行校核，对荷载谱进行分析也可为结构疲劳分析提供更接近实际的依据。而目前桥梁移动荷载识别技术还不够成熟，而且利用车桥系统模型识别移动荷载是一个反卷积求解问题，，其数学反演过程往往是不适定的，导致了这种方法对噪声很敏感。本文研究了将BP神经网络用于桥梁移动荷载识别的理论和方法，对一跨度为30m的简支梁桥进行了移动荷载识别的数值仿真，分析了桥梁挠度和应变对移动荷载的敏感性，讨论了网络的不同转移函数组合和算法对识别结果的影响，研究了不同荷载工况下的识别结果和噪声的影响，并通过试验验证了该方法的合理性。研究结果表明：用人工神经网络方法识别桥梁移动荷载是可行的；桥梁应变响应比挠度响应对移动荷载更敏感；网络不同组合的转移函数对荷载识别结果影响不大，网络的均方误差最大的为3.7288，最小的为2.8518，相关系数均大于0.97，而训练方法对结果有很大影响，网络的均方误差在2.491到1677.6382不等，相关系数也从0.1354到0.97717不等；网络对荷载位置的识别结果很好，顺利识别出了荷载的上下桥状态和在桥上的位置，最大误差为0.54m；网络对轴距识别的精度好坏变化性较大，总体规律是轴距越大，车速越慢识别效果越好，速度从25m/s降到5m/s时网络的正确识别率增加了26.43%；网络对荷载进行识别时，在车辆的上下桥段识别误差比车辆完全在桥上时的识别误差大，不同的轴距和速度对荷载的识别影响也很大，车速和轴距越大网络的识别精度越差，反之越好；轴距对速度识别的精度影响不大，速度识别的精度与速度本身的大小有关，速度越大识别的精度越低；该方法具有很好的抗噪能力，在噪声水平20%的情况下，网络的正确识别率仍大于60%。试验结果表明：模型梁一到四阶频率相对误差分别为5.3%、11.6%、13.5%、15.7，模型梁的阻尼很小，一阶模态阻尼为0.618%；最大位置识别误差为0.464m；速度识别相对误差在5%以内；识别出的动荷载时程曲线在静载线上下波动。
[Abstract]:Bridge moving load identification is an important link in bridge structure health monitoring. Accurate and reliable load data can be used to check the load selected in bridge design. The analysis of load spectrum can also provide a more practical basis for structural fatigue analysis, but the identification technology of bridge moving load is not mature enough, and the identification of moving load using vehicle-bridge system model is a deconvolution problem. The mathematical inversion process is often ill-posed, which leads to the sensitivity of this method to noise. In this paper, the theory and method of applying BP neural network to the identification of moving loads of bridges are studied. The moving load identification of a simply supported beam bridge with a span of 30 m is simulated, and the sensitivity of deflection and strain of the bridge to moving loads is analyzed. The effects of different transfer function combinations and algorithms on the recognition results are discussed. The effects of identification results and noise under different load conditions are studied. The rationality of the method is verified by experiments. The results show that the method of artificial neural network is feasible to identify the moving load of bridge, the strain response of bridge is more sensitive to the moving load than the deflection response, and the transfer function of different combinations of the network has little effect on the load identification results. The mean square error of the network is 3.7288, the least is 2.8518, the correlation coefficient is more than 0.97, and the training method has a great influence on the result. The mean square error of the network ranges from 2.491 to 1677.6382, and the correlation coefficient ranges from 0.1354 to 0.97717. The result of network recognition of load position is very good, and the load upper and lower bridge status and position on the bridge are recognized smoothly, the maximum error is 0.54 m, the accuracy of network identification of wheelbase is more variable, the overall rule is that the greater the wheelbase, the bigger the network is. The slower the speed, the better, when the speed drops from 25m / s to 5m / s, the correct recognition rate of the network increases by 26.43. When the network recognizes the load, the recognition error between the upper and lower segments of the vehicle is greater than that of the vehicle when the vehicle is completely on the bridge. Different wheelbase and speed have great influence on the identification of load, the greater the speed and the greater the wheelbase, the worse the recognition accuracy is, the better the vice versa; the less the effect of wheelbase on the accuracy of velocity identification, the more the accuracy of velocity recognition is related to the speed itself. The higher the speed is, the lower the accuracy is, and the method has a good anti-noise capability, and the correct recognition rate of the network is still greater than 60% when the noise level is 20%. The experimental results show that the relative errors of the first to fourth order frequencies of the model beams are 5.311.6 and 13.53.The damping of the model beams is very small, the first order modal damping is 0.618, the maximum position identification error is 0.464m, the relative error of velocity identification is less than 5%. The identified dynamic load history curve fluctuates up and down the static load line.
【学位授予单位】：内蒙古科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：U441.2;U448.217

【参考文献】