基于一维卷积神经网络的塔式起重机结构损伤诊断方法研究  

Research on structural damage diagnosis method of tower crane based on one-dimensional convolution neural network

作　　者：宋世军[1] 刘昂 安增辉 杨蕊 吴月华 Song Shijun;Liu Ang;An Zenghui;Yang Rui;Wu Yuehua

机构地区：[1]山东建筑大学机电工程学院,济南250101 [2]山东富友科技有限公司,济南250101

出　　处：《起重运输机械》2022年第18期44-50,共7页Hoisting and Conveying Machinery

基　　金：国家自然科学基金(52005300)。

摘　　要：文中针对塔式起重机结构损伤诊断依赖较强的专业知识、诊断过程复杂且效率低的问题,基于深度学习理论提出面向塔式起重机结构数据自动分类的一维卷积神经网络模型。首先针对塔式起重机结构损伤数据匮乏的问题,通过分析塔式起重机顶端位移数据的结构特点,研究了塔式起重机结构损伤数据增强方法;其次基于1D CNN模型泛化理论,建立塔式起重机结构损伤诊断模型;最后通过研究超参数对诊断结果的影响规律,以及与目前方法的对比,验证了所提出方法有效性与鲁棒性,并给出了参数选择方案。结果表明:该方法在推荐超参数下诊断准确率达到97.3%,说明其能够有效、较为准确地实现塔式起重机结构损伤的诊断,为基于深度学习的塔式起重机智能损伤诊断方法提供了方法指导与理论依据。Considering that the structural damage diagnosis of tower crane depends on strong professional knowledge,the diagnosis process is complicated and inefficient,a one-dimensional convolution neural network model for automatic classification of tower crane structural data is proposed based on deep learning theory.Firstly,for the lack of structural damage data of tower cranes,by analyzing the structural characteristics of the top displacement data of tower cranes,the enhancement method of structural damage data of tower cranes was studied.Secondly,based on the generalization theory of 1D CNN model,the structural damage diagnosis model of tower crane was established.Finally,the effectiveness and robustness of this method were proved by studying the influence of super parameters on diagnosis results and comparing with the current methods,and the scheme of parameter selection was given.Results show that the diagnostic accuracy rate of this method is 97.3%under the recommended super-parameters,which indicates that it can effectively and accurately diagnose the structural damage of tower cranes,and provides the method guidance and theoretical basis for the intelligent damage diagnosis method of tower cranes based on deep learning.

关 键 词：塔式起重机 一维卷积神经网络 结构损伤 超参数 深度学习 

分 类 号：TH213.3[机械工程—机械制造及自动化] TH242