基于多变量卷积神经网络的发酵过程故障监测  

Fault monitoring of fermentation processes based on multivariable convolution neural network

作　　者：高学金 刘爽爽[1,2,3,4] 高慧慧 GAO Xue-jin;LIU Shuang-shuang;GAO Hui-hui(Faculty of Information Technology,Beijing University of Technology,Beijing 100124,China;Engineering Research Center of Digital Community,Ministry of Education,Beijing 100124,China;Beijing Laboratory for Urban Mass Transit,Beijing 100124,China;Beijing Laboratory of Computational Intelligence and Intelligent System,Beijing 100124,China)

机构地区：[1]北京工业大学信息学部,北京100124 [2]数字社区教育部工程研究中心,北京100124 [3]城市轨道交通北京实验室,北京100124 [4]计算智能与智能系统北京市重点实验室,北京100124

出　　处：《高校化学工程学报》2020年第6期1511-1519,共9页Journal of Chemical Engineering of Chinese Universities

基　　金：国家自然科学基金(61803005,61640312,61763037);北京市自然科学基金(4172007,4192011);山东省重点研发计划(2018CXGC0608);北京市教育委员会资助。

摘　　要：针对传统故障监测方法难以提取数据深层特征的问题,提出一种基于多变量深度卷积神经网络的故障监测方法,以提高监测精度。为捕获过程动态性,采用滑动窗技术对过程变量序列进行分割,利用希尔伯特-黄变换对分割后的序列进行分解,得到时频图,有效挖掘变量序列在幅值、频率、相位上的异常变化信息;以时频图为输入,基于深度卷积神经网络构建故障监测模型,提取故障信息深层特征,提高监测精度;利用青霉素发酵过程仿真数据和大肠杆菌生产数据进行实验验证,结果表明所提方法监测精度分别高达95%和93%以上。A fault detection method based on multivariate deep convolutional neural network was proposed to improve monitoring accuracy,as traditional fault detection methods have problems in deeply extracting data features.In order to capture process dynamics,the process variable sequences were segmented using sliding window technology.Hilbert-Huang transform was used to decompose the segmented sequence to obtain a time-frequency map,which can effectively mine abnormal change information of the variable sequence in amplitude,frequency and phase.A fault monitoring model was built based on the deep convolutional neural network using time-frequency maps as input,and the deep features of the fault information were extracted to improve monitoring accuracy.Simulations were carried out using simulation data of penicillin fermentation and E.coli production data,and the results show that the proposed method has detection accuracy of over 95%and 93%,respectively.

关 键 词：卷积神经网络 希尔伯特-黄变换 特征提取 发酵过程 故障监测 

分 类 号：TP206.3[自动化与计算机技术—检测技术与自动化装置]