基于全连接神经网络与传递率函数相结合的钢结构损伤检测方法被引量：7

Damage detection method of steel structure based on fully connected neural network model and transmissibility function

下载PDF

导出

摘要针对现有的传递率函数法对钢结构损伤检测不全面、不精确的问题,提出了将全连接神经网络模型与传递率函数相结合的钢结构损伤检测方法。利用Ansys软件对钢结构框架模型进行瞬态分析,得到垂直于检测面的节点加速度,通过Matlab软件对各节点加速度进行傅里叶变换、频谱相除得到传递率函数,再将传递率函数进行差分运算得到传递率函数变化量。将传递率函数变化量作为全连接神经网络的输入参数,采用反向传播对误差进行修正,从而得到钢结构各检测位置的损伤指标值。通过钢结构框架损伤检测实验,对基于全连接神经网络与传递率函数的钢结构损伤检测算法进行实验验证。结果表明,新方法与传统的传递率函数法相比,所有位置的钢结构损伤识别率平均提高了29.14%,能更准确与全面地识别钢结构各位置的损伤情况,可对钢结构损伤位置进行精确定位。 Towards the problem that the existing transmissibility function method is not comprehensive and detection accuracy is not high for steel structure damage detection,a damage detection method of steel structure which combines the full connection neural network model and transmissibility function is studied.The transient analysis of the steel frame structure model is carried out by using Ansys software,and the node acceleration perpendicular to the detection surface is obtained.The transmissibility function is obtained by Fourier transform and spectrum division of each node acceleration with Matlab software,and then the transmissibility function is obtained by differential operation.The change of the transmissibility function is used as the input parameter of the fully connected neural network,and the error is corrected by back propagation to obtain the damage index value of each detection position of the steel structure.Through the experiment of steel structure frame damage detection,the algorithm of steel structure damage detection based on full connection neural network and transfer rate function is verified.The results show that compared with the traditional transmissibility function method,the damage identification rate of steel structure in all positions is increased by 29.14%on average,which can identify the damage situation of steel structure in all positions more accurately and comprehensively,and accurately locate the damage position of steel structure.

作者艾青林林小贝徐巧宁 Ai Qinglin;Lin Xiaobei;Xu Qiaoning(Key Laboratory of E&M(Zhejiang University of Technology),Ministry of Education&Zhejiang Province,Hangzhou 310023)

机构地区浙江工业大学特种装备制造与先进加工技术教育部/浙江省重点实验室

出处《高技术通讯》 CAS 2021年第8期824-835,共12页 Chinese High Technology Letters

基金国家自然科学基金(52075488,51705456,51275470) 浙江省自然科学基金(LY20E050023)资助项目。

关键词钢结构损伤检测全连接神经网络反向传播传递率函数损伤识别率 steel structure damage detection fully connected neural network back propagation transmissibility function damage identification rate