【作者】 王坤；

【导师】 叶继红；

【作者基本信息】 中国矿业大学 ， 结构工程， 2021， 硕士

【摘要】 轻钢龙骨复合剪力墙结构具有环保节能、施工快捷等优势,发展多层轻钢结构契合我国装配式建筑的需要。在国内推广应用时,多层轻钢结构的强震倒塌研究是一个关键问题。利用人工智能技术,依靠已有的大量试验或数值模拟结果分析和预测新结构的抗震性能,具有较高的科技价值。细胞自动机(Cellular Automata,CA)技术已被用于多层轻钢结构在强烈地震作用下的倒塌预测研究,但该方法对于竖向不规则多层轻钢结构的预测误差偏大。在此基础上,本文以竖向不规则多层轻钢结构为研究对象,对原有CA法进行改进,并引入相关向量机(Relevance Vector Machine,RVM)技术建立结构的倒塌预测模型,丰富利用人工智能技术进行多层轻钢结构倒塌研究的内容。为验证本文提出的预测方法具有普适性,选取了3种层间位移角限值(2.0%、3.5%及4.0%)判定结构倒塌。考虑到竖向不规则多层轻钢结构试验数据较少,基于本课题组已完成的轻钢复合剪力墙抗侧性能试验及墙体简化模型,本文将多层轻钢结构二维平面简化模型的数值分析结果作为倒塌真实值,以验证预测方法的可靠性。根据竖向不规则多层轻钢结构存在刚度突变、在强震作用下进入弹塑性阶段的特点,本文对原有CA法的细胞空间、细胞状态值及匹配准则进行修正,提出了改进CA模型。其中,将细胞空间的邻居细胞增加为2,并相应修正匹配准则的计算公式;给出两种细胞状态值定义方式——结构的等效振型值与Pushover分析后结构层间位移的归一化结果。采用两种细胞状态值所对应的改进CA法及原有CA法,对5层待预测结构进行倒塌预测,并将预测结果与时程分析结果对比。结果表明:在3种倒塌判别准则下,归一化等效振型值定义细胞状态值对应的改进CA法与原有CA法的预测误差大多处于±40%以内,最大误差绝对值超过78%,误差过大。相比于上述采用结构弹性阶段特征值定义细胞状态值对应的CA法,本文提出的归一化Pushover层间位移(弹塑性阶段特征值)定义细胞状态值对应的改进CA法的预测误差较小,大多处于±20%以内,且误差绝对值均未超过32%,该方法明显提高了竖向不规则多层轻钢结构的倒塌预测精度。考虑到结构刚度突变会影响竖向不规则多层轻钢结构倒塌时的层间位移,故本文采用相关向量机建立了结构层间刚度与倒塌层间位移之间的映射关系。选用高斯核函数和快速序列稀疏贝叶斯算法作为RVM模型的核心部分。将结构的层间刚度与刚度竖向变化值作为模型的输入量,结构倒塌时层间位移的时程分析结果作为模型的输出量,训练RVM模型。对4层竖向不规则轻钢结构进行强震倒塌预测,并将预测结果与时程分析结果比较。结果表明:在3种倒塌判别准则下,RVM法的预测误差绝对值均未超过34%,且大部分处于20%以内,预测效果较好。针对因样本数量少,导致CA法难以通过匹配准则得到更为准确预测结果的问题,本文采用相关向量机替换改进CA法的匹配准则,实现CA与RVM的联合预测。将细胞状态值(结构归一化Pushover分析后的层间位移)作为联合模型的输入数据,结构倒塌时层间位移的时程分析结果作为模型的输出数据。对6层竖向不规则轻钢结构,分别采用联合法、改进CA法、RVM法及原有CA法进行强震倒塌预测,并将预测结果与时程分析结果对比。结果表明:在3种倒塌判别准则下,原有CA法的预测误差大部分在±40%以内,但最大误差绝对值超过70%;改进CA法、RVM法的预测误差主要分布在±20%以内,最大误差绝对值均不超过40%,在可接受范围内。相比于上述方法,本文提出的CA与RVM联合法的预测精度最高,其预测误差主要分布于±10%以内,最大误差绝对值不超过15%,可准确得到待预测结构在不同地震波作用下每一层的倒塌层间位移。预测结果表明,本文提出的三种预测方法对于竖向不规则多层轻钢龙骨复合剪力墙结构进行强震倒塌预测均具有可行性与适用性。具体应用时,可根据输入数据(层间刚度或Pushover分析后的层间位移)、训练过程(匹配准则或相关向量机)及预测精度等不同需求选择预测方法。更多还原

【Abstract】 Cold-formed steel framed structures with composite shear walls have the advantages of environmental protection,energy saving,and quick construction.The development of mid-rise cold-formed steel structures meets the needs of prefabricated buildings in China.When it is promoted and applied in practice,the study of strong earthquake collapse of mid-rise cold-formed steel structures is a key issue.Using artificial intelligence technology to analyze and predict the seismic performance of new structures based on the results of a large number of existing tests or numerical simulations has a high scientific and technological value.Cellular Automata(CA)technology has been used to predict the collapse of mid-rise cold-formed steel structures under strong earthquakes,but this method has a large prediction error for vertically irregular mid-rise cold-formed steel structures.On this basis,vertical irregular mid-rise cold-formed steel structures are taken in this paper as the research object,and the collapse prediction model of the structure is established by improving the original CA method and introducing the Relevance Vector Machine(RVM)technology.This study makes full use of artificial intelligence technology to investigate the collapse behavior of mid-rise cold-formed steel structures.In order to verify the universal applicability of the prediction method proposed in this paper,three types of story drift angle limits(2.0%,3.5% and 4.0%)are selected to determine the structure collapse.Considering that the test data of vertically irregular mid-rise cold-formed steel structures is less and based on the lateral performance tests of cold-formed steel composite shear wall and the simplified model of the wall completed by this research group,the numerical analysis results of the two-dimensional simplified model of mid-rise cold-formed steel structures are taken as the real collapse values to verify the reliability of the prediction method.According to the characteristics of vertically irregular mid-rise cold-formed steel structures that have sudden changes in stiffness and enter the elastoplastic stage under strong earthquakes,this paper revises the cell space,cell state values and matching criteria of the original CA method,and proposes an improved CA model.Among them,the neighbor cells in the cell space are increased to 2,and the calculation formula of the matching criterion is revised accordingly;two kinds of cell state values are defined: the normalized results of the equivalent mode of the structure and the normalized results of the story drift after Pushover analysis.The improved CA method corresponding to the two cell state values and the original CA method were used to predict the collapse of the 5-layer structure to be predicted,and the predicted results were compared with the results of time-history analysis.The results show that under the three collapse criteria,the prediction errors of the improved CA method corresponding to the normalized equivalent mode values defined cell state values and the original CA method are mostly within ±40%,and the maximum absolute error is more than 78%,which is too large.Compared to the CA method using elastic stage characteristic value as cell status value,the improved CA method put forward in this paper using the normalized Pushover story drift of the elastic-plastic stage eigenvalues as cell state values has a prediction error of ±20%,where absolute errors are no more than 32%.This method obviously improves the prediction accuracy of collapse of vertically irregular mid-rise cold-formed steel structures.Considering that the sudden change of structural rigidity will affect the story drift when the vertically irregular mid-rise cold-formed steel structure collapses,this paper uses relevance vector machine to establish the mapping relationship between the structural inter-story stiffness and the collapsed story drift.Gaussian kernel function and fast sequence sparse Bayesian algorithm are selected as the core part of the RVM model.The inter-story stiffness and the vertical stiffness variation of the structure are taken as the input of the model,and the time history analysis results of the displacement between the floors when the structure collapses are taken as the output of the model to train the RVM model.The collapse prediction of strong earthquakes is carried out for the 4-story vertically irregular cold-formed steel structure,and the prediction results are compared with the time history analysis results.The results show a good prediction accuracy under the three collapse criteria since the absolute value of the prediction error of the RVM method does not exceed 34%,and most of them are within 20%.Aiming at the problem that it is difficult to obtain more accurate prediction results from CA method through matching criteria due to the small number of samples,this paper uses relevance vector machine to replace the matching criteria of improved CA method,and thus to realize the joint prediction of CA and RVM.The cell state value(the story drift after structure normalization Pushover analysis)is used as the input data of the joint model,and the time history analysis results of the story drift when the structure collapses are used as the output data of the model.For the 6-story vertically irregular cold-formed steel structure,the combined method,the improved CA method,the RVM method and the original CA method were used to predict the collapse of strong earthquakes,and the prediction results were compared with the time history analysis results.The results show that under the three collapse criteria,most of the prediction errors of the original CA method are within ±40%,but the absolute value of the maximum error is more than 70%.The prediction errors of the improved CA method and RVM method are mainly distributed within ±20%,and the absolute value of the maximum error is no more than 40%,which is within the acceptable range.Compared with the above methods,the combined method of CA and RVM proposed in this paper has the highest prediction accuracy,with the prediction error mainly distributed within ±10% and the maximum absolute error less than 15%.This method can accurately obtain the story drift of the structure to be predicted under the action of different seismic waves.The prediction results show that the three prediction methods proposed in this paper are feasible and applicable for the collapse prediction of vertically irregular mid-rise cold-formed steel framed structures with composite shear walls under strong earthquakes.In the practical application,the prediction method can be selected according to the input data(inter-layer stiffness or story drift after Pushover analysis),training process(matching criterion or relevance vector machine)and prediction accuracy.更多还原