【作者】 黄达；

【导师】 吴斌； 杨格； 孙国良；

【作者基本信息】 武汉理工大学 ， 土木工程， 2023， 硕士

【摘要】 振动台实时混合试验是一种新型抗震试验方法,可以解决目前振动台受尺寸限制而难以开展大型结构足尺试验的问题。其基本思想是利用子结构技术,即将整体结构划分为试验子结构和数值子结构,通过数值模拟计算和物理试验协调工作来得到整体结构在地震作用下的时程响应。然而,当实时混合试验所取的数值子结构较为复杂时,会出现计算时滞较大的问题,从而达不到实时的要求导致试验失败。为了满足计算的实时性,现有研究通常采用简化的数值模型,又会导致实时混合试验的精度降低。通信机柜群作为通信设备的承载体系,对于震后救援救灾以及恢复保障具有至关重要的作用,必须保障其在地震作用下的安全性。因此,有必要开展通信机柜群的振动台实时混合试验,研究其抗震性能。然而,通信机柜群的数值模型较为复杂,存在着如前所述计算时滞过大而无法满足实时性的问题。为了开展通信机柜群的振动台实时混合试验,本文首先提出基于数值子结构代理模型的方法,来解决实时混合试验中计算时滞过大的问题;然后对由多个通信机柜并排组成的通信机柜群开展了振动台实时混合试验仿真研究,为后续的物理试验提供参考和依据。本文主要进行了以下研究工作:(1)建立了通信机柜群整体结构的有限元模型,并开展有限元分析获得神经网络训练所需的数据集。基于通信机柜主要承重结构是框架结构的特点,主要采用梁单元建模。对结构进行多工况的有限元地震时程分析,得到边界上力与位移的时程数据作为下一步神经网络训练所需要的数据集。(2)在MATLAB中搭建神经网络模型,以有限元时程分析结果作为训练集建立了通信机柜群的代理模型。根据通信机柜群振动台实时混合试验的基本运动方程,提出了在弹性阶段以试验子结构的反力和地震波加速度为输入、在弹塑性阶段以前50步试验子结构的反力、前50步位移和当前步地震波加速度为输入的神经网络代理模型,通过训练确定了神经网络隐含层层数以及隐含层节点数,经过测试发现该神经网络代理模型在结构弹性阶段和弹塑性阶段都具有较好的预测效果。(3)以代理模型代替数值子结构的传统有限元模型,完成了通信机柜群的振动台实时混合试验仿真。在Simulink中搭建了通信机柜群振动台实时混合试验仿真系统,开展了不同地震峰值加速度作用工况下的代理模型振动台实时混合试验仿真。然后与整体结构时程分析的结果进行对比,,验证了所建立的代理模型可以克服计算时滞过大的问题,且具有较高的计算精度。更多还原

【Abstract】 The real-time hybrid test of the shaking table is a new type of seismic test method that can solve the problem that the current shaking table is limited by dimensions and it is difficult to carry out large structure foot-foot tests.Its basic idea is to use sub-structure technology to divide the overall structure into experimental structures and numerical structures.The overall structure response is obtained through the coordination of numerical simulation computing and physical test coordination.However,when the numerical sub-structure taken by the real-time hybrid test is more complicated,the problem of computing time delay will occur,which fails to meet the real-time requirements and cause the test to fail.In order to meet the real-time nature of calculation,the existing research usually uses a simplified numerical model,which will also cause the accuracy of real-time hybrid tests to reduce the accuracy.As a bearing system of communication equipment,the communication cabinet group plays a vital role in post-earthquake rescue and disaster relief and recovery support,and its safety under the action of earthquake must be guaranteed.Therefore,it is necessary to carry out real-time hybrid tests of shaking tables in communication cabinet groups to study their seismic performance.However,the numerical model of the communication cabinet cluster is more complex,and there is a problem that the calculation time delay is too large as mentioned above to meet the real-time performance.In order to carry out the real-time hybrid test of the shaking table of the communication cabinet group,this paper first proposes a method based on the numerical substructure surrogate model to solve the problem of excessive calculation time delay in the real-time hybrid test.Then,the real-time hybrid test simulation research of the shaker was carried out on the communication cabinet group composed of multiple communication cabinets side by side,which provided reference and basis for subsequent physical tests.In this paper,the following research work is carried out:(1)The finite element model of the overall structure of the communication cabinet parts is established,and the data sets required to obtain the finite element analysis obtain the data set of neural network training.Based on the main load-bearing structure of the communication machine cabinet is the characteristics of the frame structure,and the beam unit modeling is mainly used.Analysis of the finite element earthquake schedule of the structure of the structure,obtaining the timing data of the boundary force and displacement as the data set required for the next step of neural network training.(2)Construction of neural network models in Matlab,using the results of finite element time analysis as a training set to establish a surrogate model of communication cabinet group.According to the basic motion equation of the real-time hybrid test of the communication cabinet group,it was proposed that in the elastic phase,the reaction force of the experimental structure and the acceleration of the seismic waves were input,and the reaction force of the experimental structure of the test sub-test sub-step before the elastial stage,the first 50 step displacement And the current sub-seismic wave acceleration is an input neural network surrogate model.The number of hidden layers and the number of hidden layers nodes was determined by training.After testing,it is found that the neural network surrogate model has a good predictive effect in the structural elastic phase and the elastability stage.(3)By using surrogate model of numerical substructure,the real-time hybrid test simulation of communication cabinet group is completed.The real-time hybrid test simulation of shaking table system of communication cabinet group was built in Simulink.Then,the dynamic response of the whole structure was taken as the reference solution,and the real-time hybrid test results of shaking table based on the surrogate model were analyzed under multiple peak seismic acceleration conditions.It is verified that the proposed surrogate model has a high accuracy in the real time hybrid test of communication cabinet group when the calculation time delay is too large.更多还原