摘要
间歇过程数据上呈现出较强的非线性、高维性以及耦合性等特点,故障识别难度较大,而弱故障又具有幅值低、易被噪声干扰的特点,当间歇过程中出现弱故障时,更加大了检测人员对故障识别的困难。为解决该问题,提出一种基于时空融合特征的间歇过程弱故障识别方法,该方法设计了一种并行时空特征提取网络,对数据进行特征提取,通过对特征进行识别来判断故障类别。并行时空特征提取网络由卷积神经网络和长短期记忆网络并行构成,同步计算,网络的输出端由一定尺寸的卷积核对各网络提取的特征进行深度融合,最后通过一个全连接层将特征输入分类器,进行故障识别。利用青霉素发酵仿真实验数据进行实验,验证了方法的有效性。
The data of batch process shows strong nonlinear,high dimensional and coupling characteristics.It is difficult to identify the fault,while the small fault has the characteristics of low amplitude and easy to be interfered by noise.When the small fault occurs in the batch process,it is more difficult to identify the fault.In order to solve this problem,a weak fault identification method based on time-space fusion features was proposed.This method designed a parallel time-space feature extraction network to extract data features and judge fault categories by identifying features.The parallel spatiotemporal feature extraction network was composed of convolutional neural networks(CNN)and long short-term memory(LSTM)in parallel.The output of the network consisted of convolution of a certain size to check the extracted features of each network for deep fusion.Finally,the features were input into the classifier through a full connection layer for fault recognition.The simulation experiment data of penicillin fermentation was used to verify the effectiveness of the method.
作者
张敏
李贤均
王瑞琦
张则强
ZHANG Min;LI Xianjun;WANG Ruiqi;ZHANG Zeqiang(School of Mechanical Engineering,Southwest Jiaotong University,Chengdu,Sichuan 61003,China;Technology and Equipment of Rail Transit Operation and Maintenance Key Laboratory of Sichuan Province,Chengdu,Sichuan 610031,China)
出处
《工业工程与管理》
北大核心
2022年第3期64-73,共10页
Industrial Engineering and Management
基金
国家重点研发计划资助项目(2020YFB1712200)
四川省科技计划资助项目(2020JDTD0012)
中国博士后科学基金资助项目(2020M673279)
中铁工程服务资助项目(2019H010103)。
关键词
间歇过程
智能故障诊断
特征提取
卷积神经网络
长短期记忆网络
batch process
intelligent fault diagnosis
feature extraction
convolutional neural network
long short-term memory network