神经网络搭载Inception模块的框架结构集成故障诊断  

Integrated Fault Diagnosis of Frame Structure Based on Neural Network with Inception Model

作　　者：蔡超志 池耀磊 郭璐彬 CAI Chao-zhi;CHI Yao-lei;GUO Lu-bin(Shool of Mechanical and Equipment Engineering Hebei University of Engineering,Hebei Handan 056038,China)

机构地区：[1]河北工程大学机械与装备工程学院,河北邯郸056038

出　　处：《机械设计与制造》2024年第6期170-176,共7页Machinery Design & Manufacture

基　　金：河北省自然科学基金(E2020402060)。

摘　　要：针对于框架结构的使用环境恶劣,同时常常伴随着大量的噪声,在使用普通的一维卷积神经网络对框架结构进行故障诊断时,存在无法做出有效故障诊断的问题。本研究在一种抗噪声能力较强的卷积神经网络中加入Inception模块,提出了一种识别率和抗噪声能力更高的卷积神经网络—BICNN(Convolution Neural Network based on Inception),并用BICNN卷积神经网络基于数据驱动的方式,对楼体框架模型进行了集成故障诊断研究。集成诊断结果表明BICNN具有更高的识别率和较强的抗噪声能力,而且在训练步数较少的情况下振荡次数少收敛情况良好。因此采取本研究所提出的方法,对框架结构进行故障诊断时具有高诊断率和稳定性,为维护框架结构的稳定运行具有重大安全意义。In view of the bad working environment of frame structure,which is often accompanied by a lot of noise,when using ordi-nary one-dimensional convolution neural network for fault diagnosis of frame structure,it is unable to make effective fault diagno-sis.In this study,the inception module is added to a convolutional neural network with strong anti-noise ability,the recognition rate and anti-noise are obtained.Therefore,a more powerful convolution neural network named BICNN(Convolution Neural Net‐work based on Inception)is raised.Based on the data-driven method,BICNN convolution neural network is used to study the inte-grated fault diagnosis of a building frame model.The integrated diagnosis results show that BICNN has higher recognition rate,stronger anti-noise ability,less oscillation times and good convergence under the condition of less training epoch.The method pro-posed in this study has high diagnosis rate and stability in fault diagnosis of frame structure,which is of great safety significance to maintain the stable operation of frame structure.

关 键 词：框架结构 故障诊断 卷积神经网络 Inception模块 抗噪声能力 正确率 

分 类 号：TH16[机械工程—机械制造及自动化] TP277[自动化与计算机技术—检测技术与自动化装置]