2.765

2022影响因子

(CJCR)

  • 中文核心
  • EI
  • 中国科技核心
  • Scopus
  • CSCD
  • 英国科学文摘

留言板

尊敬的读者、作者、审稿人, 关于本刊的投稿、审稿、编辑和出版的任何问题, 您可以本页添加留言。我们将尽快给您答复。谢谢您的支持!

姓名
邮箱
手机号码
标题
留言内容
验证码

基于深度置信网络的多模态过程故障评估方法及应用

张凯 杨朋澄 彭开香 陈志文

张凯, 杨朋澄, 彭开香, 陈志文. 基于深度置信网络的多模态过程故障评估方法及应用. 自动化学报, 2024, 50(1): 89−102 doi: 10.16383/j.aas.c230156
引用本文: 张凯, 杨朋澄, 彭开香, 陈志文. 基于深度置信网络的多模态过程故障评估方法及应用. 自动化学报, 2024, 50(1): 89−102 doi: 10.16383/j.aas.c230156
Zhang Kai, Yang Peng-Cheng, Peng Kai-Xiang, Chen Zhi-Wen. A deep belief network-based fault evaluation method for multimode processes and its applications. Acta Automatica Sinica, 2024, 50(1): 89−102 doi: 10.16383/j.aas.c230156
Citation: Zhang Kai, Yang Peng-Cheng, Peng Kai-Xiang, Chen Zhi-Wen. A deep belief network-based fault evaluation method for multimode processes and its applications. Acta Automatica Sinica, 2024, 50(1): 89−102 doi: 10.16383/j.aas.c230156

基于深度置信网络的多模态过程故障评估方法及应用

doi: 10.16383/j.aas.c230156
基金项目: 国家自然科学基金(62073032, U21A20483, 62173349) 资助
详细信息
    作者简介:

    张凯:北京科技大学自动化学院副教授. 2016年获得德国杜伊斯堡–埃森大学电气工程和信息技术博士学位. 主要研究方向为故障诊断与先进控制, 智能运维与优化决策. 本文通信作者. E-mail: kaizhang@ustb.edu.cn

    杨朋澄:北京科技大学硕士研究生. 主要研究方向为数据驱动的故障诊断与容错控制. E-mail: yangpengc09@163.com

    彭开香:北京科技大学自动化学院教授. 2007年获得北京科技大学控制科学与工程博士学位. 主要研究方向为工业智能与智能制造, 故障诊断与容错控制. E-mail: kaixiang@ustb.edu.cn

    陈志文:中南大学自动化学院副教授. 2016年获得德国杜伊斯堡–埃森大学电气工程和信息技术博士学位. 主要研究方向为模型和数据驱动的故障诊断与健康监测, 数据分析. E-mail: zhiwen.chen@csu.edu.cn

A Deep Belief Network-based Fault Evaluation Method for Multimode Processes and Its Applications

Funds: Supported by National Natural Science Foundation of China (62073032, U21A20483, 62173349)
More Information
    Author Bio:

    ZHANG Kai Associate professor at the School of Automation and Electrical Engineering, University of Science and Technology Beijing. He received his Ph.D. degree in electrical engineering and information technology from the University of Duisburg-Essen, Germany, in 2016. His research interest covers fault diagnosis and advanced control, artificial intelligence for industrial maintenance and decision-making. Corresponding author of this paper

    YANG Peng-Cheng Master student at University of Science and Technology Beijing. Her research interest covers data-based fault diagnosis and fault-tolerant control

    PENG Kai-Xiang Professor at the School of Automation and Electrical Engineering, University of Science and Technology Beijing. He received his Ph.D. degree in control science and engineering from University of Science and Technology Beijing in 2007. His research interest covers industrial intelligence and intelligent manufacturing, fault diagnosis and fault-tolerant control

    CHEN Zhi-Wen Associate professor at the School of Automation, Central South University. He received his Ph.D. degree in electrical engineering and information technology from the University of Duisburg-Essen, Germany, in 2016. His research interest covers model-based and data-driven fault diagnosis and health monitoring, data analytics

  • 摘要: 传统的多模态过程故障等级评估方法对模态之间的共性特征考虑较少, 导致当被评估模态故障信息不充分时, 评估的准确性较低. 针对此问题, 首先, 提出一种共性–个性深度置信网络 (Common and specific deep belief network, CS-DBN), 该网络充分利用深度置信网络 (Deep belief network, DBN) 的深度分层特征提取能力, 通过度量多模态数据间分布的相似性和差异性, 进一步得到能够反映多模态过程共有信息的共性特征以及反映每个模态独有信息的个性特征; 其次, 基于CS-DBN, 利用多模态过程的已知故障等级数据生成多模态共性–个性特征集, 通过加权逻辑回归构建故障等级评估模型; 最后, 将所提方法应用于带钢热连轧生产过程的故障等级评估中. 应用结果表明, 随着多模态故障等级数据的增加, 所提方法的评估准确率逐渐增加, 当故障信息充足时, 评估准确率可达98.75%; 故障信息不足时, 与传统方法相比, 评估准确率提升近10%.
  • 图  1  融合共性–个性特征的故障评估方法示意图

    Fig.  1  Schematic diagram of fault evaluation method integrating common and specific features

    图  2  热连轧过程单变量共性−个性特征分解示意图

    Fig.  2  Schematic diagram of single variable common and specific feature decomposition in hot rolling process

    图  3  轧制力故障共性特征等级评估示意图

    Fig.  3  Schematic diagram for evaluating the common feature grades of rolling force faults

    图  4  弯辊力故障共性特征等级评估示意图

    Fig.  4  Schematic diagram for evaluating the common feature grades of bending force faults

    图  5  DBN与CS-DBN网络结构示意图

    Fig.  5  Schematic diagram of DBN and CS-DBN network structure

    图  6  基于CS-DBN的故障等级评估流程图

    Fig.  6  Flow chart of fault grade evaluation based on CS-DBN

    图  7  热连轧机组及精轧机组布局图

    Fig.  7  Schematic diagram of hot continuous rolling unit and finishing rolling unit

    图  8  热连轧故障注入系统

    Fig.  8  Fault injection system for hot continuous rolling

    图  9  共性特征维度$ n_{c}$与重构误差

    Fig.  9  Common feature dimension $ n_{c}$ and reconstruction error

    图  10  CS-DBN训练过程迭代曲线

    Fig.  10  Iterative curve of CS-DBN training process

    图  11  前3个模态全部故障信息已知时的评估结果

    Fig.  11  Evaluation results with full knowledge of faults in the first three modes

    图  12  共性特征权重因子分析

    Fig.  12  Weighting factor analysis of common features

    表  1  各类共性–个性特征提取方法特点总结

    Table  1  Summary of characteristics of various common and specific feature extraction methods

    方法线性/非线性原理数据等长正交性
    CS-PCA[30]线性特征方差最大
    张量分解[31]线性重构误差最小
    CS-DBN非线性重构误差最小,
    MK-MMD距离最小
    下载: 导出CSV

    表  2  热连轧过程多模态数据描述

    Table  2  Multimode data description of hot continuous rolling process

    钢种规格 (厚度)工作模态选取变量
    Q235B碳素结构钢2.30 mm1$F_1\sim F_7$辊缝 (mm)
    $F_1\sim F_7$轧制力 (KN)
    $F_2\sim F_7$弯辊力 (KN)
    2.70 mm2
    3.00 mm3
    3.95 mm4
    下载: 导出CSV

    表  3  热连轧过程故障等级划分及标签添加

    Table  3  Fault grade division and label addition in the hot continuous rolling process

    数据类型受影响变量出口厚度差 (mm)等级等级标签
    正常±0.01正常1
    ${F_5}$弯辊力传感器故障${F_5}$和${F_6}$弯辊力±0.02轻微故障2
    ${F_4}$辊缝故障${F_4}$和${F_5}$辊缝及轧制力±0.04一般故障3
    ${F_2}$与${F_3}$间冷却水阀执行器故障${F_3}$至${F_7}$辊缝及轧制力±0.08严重故障4
    下载: 导出CSV

    表  4  CS-DBN模型参数

    Table  4  CS-DBN model parameters

    $\varepsilon$$N_b$epochdr$n_c$$n_s$${ \alpha _{re}}$${ \alpha _{c}}$${ \alpha _{s}}$
    0.0001806000.5570.30.20.05
    下载: 导出CSV

    表  5  各模态全部故障信息已知下的评估结果 (%)

    Table  5  Evaluation results for cases that all fault information in different modes is known (%)

    评估指标FDASVMSAEDBNCS-DBN
    Accuracy82.5095.2795.8793.3898.75
    Precision89.6995.3496.0894.2698.96
    MacroF185.9495.3195.9893.8298.85
    下载: 导出CSV

    表  6  各模态部分故障信息已知下的评估准确率结果 (%)

    Table  6  Evaluation accuracy results for cases that part of fault information in different modes is known (%)

    评估指标案例A: 每个训练模态中包含最多两种等级故障数据下的评估准确率平均值
    A-1A-2A-3A-4A-5A-6A-7A-8
    FDA57.5050.0060.5065.0065.0050.0050.0050.0057.25
    SVM49.7050.0050.0050.0050.2550.0044.8550.0049.35
    SAE50.7050.4850.8265.4254.0250.2050.2050.0052.73
    DBN53.6562.4553.1050.3573.1052.8557.2550.2056.62
    CS-DBN68.2364.6564.4564.4074.6067.8265.4561.3066.36
    评估指标案例B: 每个训练模态中均有两种等级故障数据下的评估准确率平均值
    B-1B-2B-3B-4B-5B-6B-7B-8
    FDA50.0045.0025.0065.2550.0052.7562.5054.2550.59
    SVM49.3850.3350.0148.0541.3540.7562.5850.2549.09
    SAE57.0854.2557.6050.0064.6548.7263.4870.2058.25
    DBN63.6260.4568.8345.5350.0072.4559.5071.3561.47
    CS-DBN65.0085.5074.0065.5573.2068.1569.5574.4071.92
    评估指标案例C: 每个训练模态中至少有两种等级故障数据下的评估准确率平均值
    C-1C-2C-3C-4C-5C-6C-7C-8
    FDA50.2550.0051.3850.0050.0050.0050.0050.0050.20
    SVM58.4350.5050.1550.2550.0058.0350.0052.2552.45
    SAE67.5060.4872.9560.8760.6264.5561.4870.3264.85
    DBN71.8370.5071.6571.5071.2860.1574.1572.7070.47
    CS-DBN73.6371.5574.3373.0570.5374.3576.9570.6073.12
    评估指标案例D: 训练模态中至少有两个模态有三种等级故障数据下的评估准确率平均值
    D-1D-2D-3D-4D-5D-6D-7D-8
    FDA51.1550.0054.3850.2552.4352.4550.0052.3551.63
    SVM53.0552.3550.6550.5551.8551.7052.2550.0051.55
    SAE74.5266.5568.7071.4568.6070.5264.4368.6269.17
    DBN72.6569.6070.8373.6574.4074.0569.9866.5071.46
    CS-DBN87.6576.4376.2088.2583.3080.1076.0376.3580.79
    下载: 导出CSV
  • [1] Zou X Y, Zhao C H. Concurrent assessment of process operating performance with joint static and dynamic analysis. IEEE Transactions on Industrial Informatics, 2020, 16(4): 2776−2786 doi: 10.1109/TII.2019.2934757
    [2] Zhu H Q, Wang Q L, Zhang F X, Yang C H, Li Y G, Zhou C. Fuzzy comprehensive evaluation strategy for operating state of electrocoagulation purification process based on sliding window. Process Safety and Environmental Protection, 2022, 165: 217−229 doi: 10.1016/j.psep.2022.06.063
    [3] Yu J X, Wu S B, Yu Y, Chen H C, Fan H Z, Liu J H, et al. Process system failure evaluation method based on a Noisy-OR gate intuitionistic fuzzy Bayesian network in an uncertain environment. Process Safety and Environmental Protection, 2021, 150: 281−297 doi: 10.1016/j.psep.2021.04.024
    [4] Zou X Y, Zhao C H, Gao F R. Linearity decomposition-based cointegration analysis for nonlinear and nonstationary process performance assessment. Industrial & Engineering Chemistry Research, 2020, 59(7): 3052−3063
    [5] Chang L L, Dong W, Yang J B, Sun X Y, Xu X B, Xu X J, et al. Hybrid belief rule base for regional railway safety assessment with data and knowledge under uncertainty. Information Sciences, 2020, 518: 376−395 doi: 10.1016/j.ins.2019.12.035
    [6] Zhou X Y, Liu Z J, Wang F W, Wu Z L, Cui R D. Towards applicability evaluation of hazard analysis methods for autonomous ships. Ocean Engineering, 2020, 214: Article No. 107773 doi: 10.1016/j.oceaneng.2020.107773
    [7] Fu F Z, Wang D Y, Li W B, Zhao D, Wu Z G. Overall fault diagnosability evaluation for dynamic systems: A quantitative-qualitative approach. Automatica, 2022, 146: Article No. 110591 doi: 10.1016/j.automatica.2022.110591
    [8] 褚菲, 赵旭, 代伟, 马小平, 王福利. 数据驱动的最优运行状态鲁棒评价方法及应用. 自动化学报, 2020, 46(3): 439−450

    Chu Fei, Zhao Xu, Dai Wei, Ma Xiao-Ping, Wang Fu-Li. Data-driven robust evaluation method for optimal operating status and its application. Acta Automatica Sinica, 2020, 46(3): 439−450
    [9] 常玉清, 孙雪婷, 钟林生, 王福利, 刘英娇. 基于改进随机森林算法的工业过程运行状态评价. 自动化学报, 2021, 47(9): 2406−2415

    Chang Yu-Qing, Sun Xue-Ting, Zhong Lin-Sheng, Wang Fu-Li, Liu Ying-Jiao. Industrial operation performance evaluation of industrial processes based on modified random forest. Acta Automatica Sinica, 2021, 47(9): 2406−2415
    [10] Peng K X, Guo Y X. Fault detection and quantitative assessment method for process industry based on feature fusion. Measurement, 2022, 197: Article No. 111267 doi: 10.1016/j.measurement.2022.111267
    [11] Tao Y, Shi H B, Song B, Tan S. Operating performance assessment and non−optimal cause identification for chemical process. The Canadian Journal of Chemical Engineering, 2019, 97(S1): 1475−1487 doi: 10.1002/cjce.23401
    [12] YeongGwang O, Kasin R, Moise B, Daeil K, Namhun K. Adaptive SVM-based real-time quality assessment for primer-sealer dispensing process of sunroof assembly line. Reliability Engineering & System Safety, 2019, 184: 202−212
    [13] Wang P, Zhang Z Y, Huang Q, Lee W J. Wind farm dynamic equivalent modeling method for power system probabilistic stability assessment. In: Proceedings of the IEEE Industry Applications Society Annual Meeting. Baltimore, USA: IEEE, 2019. 1−7
    [14] Zhao B, Zhang X M, Zhan Z H, Wu Q Q. Deep multi-scale separable convolutional network with triple attention mechanism: A novel multi-task domain adaptation method for intelligent fault diagnosis. Expert Systems With Applications, 2021, 182: Article No. 115087 doi: 10.1016/j.eswa.2021.115087
    [15] 褚菲, 傅逸灵, 赵旭, 王佩, 尚超, 王福利. 基于ISDAE模型的复杂工业过程运行状态评价方法及应用. 自动化学报, 2021, 47(4): 849−863

    Chu Fei, Fu Yi-Ling, Zhao Xu, Wang Pei, Shang Chao, Wang Fu-Li. Operating performance assessment method and application for complex industrial process based on ISDAE model. Acta Automatica Sinica, 2021, 47(4): 849−863
    [16] Peng R R, Zhang X Z, Shi P M. Bearing fault diagnosis of hot-rolling mill utilizing intelligent optimized self-adaptive deep belief network with limited samples. Sensors, 2022, 22(20): 7815−7836 doi: 10.3390/s22207815
    [17] 王功明, 乔俊飞, 关丽娜, 贾庆山. 深度信念网络研究现状与展望. 自动化学报, 2021, 47(1): 35−49

    Wang Gong-Ming, Qiao Jun-Fei, Guan Li-Na, Jia Qing-Shan. Review and prospect on deep belief network. Acta Automatica Sinica, 2021, 47(1): 35−49
    [18] Ying Y H, Li Z, Yang M L, Du W L. Multimode operating performance visualization and nonoptimal cause identification. Processes, 2020, 8(1): Article No. 123
    [19] Ma M, Sun C, Zhang C, Chen X F. Subspace-based MVE for performance degradation assessment of aero-engine bearings with multimodal features. Mechanical Systems and Signal Processing, 2019, 124: 298−312 doi: 10.1016/j.ymssp.2018.12.008
    [20] 褚菲, 许杨, 尚超, 王福利, 高福荣, 马小平. 基于静–动态特性协同感知的复杂工业过程运行状态评价. 自动化学报, 2023, 49(8): 1621-1634

    Chu Fei, Xu Yang, Shang Chao, Wang Fu-Li, Gao Fu-Rong, Ma Xiao-Ping. Evaluation of complex industrial process operating state based on static and dynamic cooperative perception. Acta Automatica Sinica, 2023, 49(8): 1621-1634
    [21] Shao W M, Ge Z Q, Yao L, Song Z H. Bayesian nonlinear Gaussian mixture regression and its application to virtual sensing for multimode industrial processes. IEEE Transactions on Automation Science and Engineering, 2020, 17(2): 871−885 doi: 10.1109/TASE.2019.2950716
    [22] Lu C Q, Wang S P. Performance degradation prediction based on a Gaussian mixture model and optimized support vector regression for an aviation piston pump. Sensors. 2020, 20(14): 3854−3874 doi: 10.3390/s20143854
    [23] Tang L, Hui Y, Yang H, Zhao Y H, Tian C G. Medical image fusion quality assessment based on conditional generative adversarial network. Frontiers in Neuroscience, 2022, 16(8): 986153−986162
    [24] Launay H, Ryckelynck D, Lacourt L, Besson J, Mondon A, Willot F. Deep multimodal autoencoder for crack criticality assessment. International Journal for Numerical Methods in Engineering, 2022, 123(6): 1456−1480 doi: 10.1002/nme.6905
    [25] 李宝琴, 吴俊勇, 李栌苏, 史法顺, 赵鹏杰, 王燚. 基于主动迁移学习的电力系统暂态稳定自适应评估. 电力系统自动化, 2023, 47(4): 121−132

    Li Bao-Qin, Wu Jun-Yong, Li Lu-Su, Shi Fa-Shun, Zhao Peng-Jie, Wang Yi. Adaptive assessment of power system transient stability based on active transfer learning. Automation of Electric Power Systems, 2023, 47(4): 121−132
    [26] Zhang Y W, Li S. Modeling and monitoring between-mode transition of multimodes processes. IEEE Transactions on Industrial Informatics, 2013, 9(4): 2248−2255 doi: 10.1109/TII.2012.2220977
    [27] Gretton A, Sriperumbudur B, Sejdinovic D, Strathmann H, Balakrishnan S, Pontil M, et al. Optimal kernel choice for large-scale two-sample tests. In: Proceedings of the 25th International Conference on Neural Information Processing Systems (NIPS). Lake Tahoe, USA: ACM, 2012. 1205−1213
    [28] Hinton G E. Training products of experts by minimizing contrastive divergence. Neural Computation, 2002, 14(8): 1771−1800 doi: 10.1162/089976602760128018
    [29] Choo S, Lee H. Learning framework of multimodal Gaussian-Bernoulli RBM handling real-value input data. Neurocomputing, 2018, 275: 1813−1822 doi: 10.1016/j.neucom.2017.10.018
    [30] Zhang K, Peng K X, Zhao S S, Chen Z W. A novel common and specific features extraction-based process monitoring approach with application to a hot rolling mill process. Control Engineering Practice, 2020, 104: Article No. 104628 doi: 10.1016/j.conengprac.2020.104628
    [31] Zhang K, Peng K X, Zhao S S, Wang F. A novel feature extraction-based process monitoring Method for multimode processes with common features and its applications to a rolling process. IEEE Transactions on Industrial Informatics, 2021, 17(9): 6466−6475 doi: 10.1109/TII.2020.3012024
    [32] Jessen H C. Applied logistic regression analysis. Journal of the Royal Statistical Society: Series D (The Statistician), 1996, 45(4): 534−535
  • 加载中
图(12) / 表(6)
计量
  • 文章访问数:  195
  • HTML全文浏览量:  128
  • PDF下载量:  99
  • 被引次数: 0
出版历程
  • 收稿日期:  2023-03-27
  • 录用日期:  2023-07-22
  • 网络出版日期:  2023-12-13
  • 刊出日期:  2024-01-29

目录

    /

    返回文章
    返回