基于并行LSTM-CNN的化工过程故障检测

肖飞扬; 顾幸生

doi:10.14135/j.cnki.1006-3080.20220120001

基于并行LSTM-CNN的化工过程故障检测

doi: 10.14135/j.cnki.1006-3080.20220120001

肖飞扬,
顾幸生^,

华东理工大学能源化工过程智能制造教育部重点实验室, 上海 200237

基金项目: 国家自然科学基金项目（61973120）．

详细信息

作者简介:
肖飞扬（1997-），男，硕士生，研究方向为基于机器学习的故障诊断。E-mail：m17864291933@163.com

通讯作者:
顾幸生，E-mail：xsgu@ecust.edu.cn

中图分类号: CN31-1691/TQ
计量
- 文章访问数: 68
- HTML全文浏览量: 40
- PDF下载量: 10
- 被引次数: 0
出版历程
- 收稿日期: 2022-01-20
- 网络出版日期: 2022-04-24

Fault detection of chemical process based on parallel connection PLSTM-CNN

Xiao Feiyang,
Gu Xingsheng^,

Key Laboratory of Energy and Chemical Process Intelligent Manufacturing of Ministry of Education, East China University of Science and Technology, Shanghai 200237, China

摘要

摘要: 为了保证生产过程的安全稳定运行，避免因故障导致的损失，及时地检测出异常工况并准确地对异常工况进行诊断具有十分重要的研究意义。针对化工过程的复杂性，本文提出一种并行长短时记忆网络和卷积神经网络（Parallel long and short-term memory network and convolutional neural network，PLSTM-CNN）模型进行化工生产过程故障检测。该模型有效结合LSTM对时间序列数据全局特征提取能力和CNN模型擅于提取局部特征的能力，减少了特征信息的丢失，实现了较高的故障检测率。采用一维稠密卷积神经网络作为CNN的主体，结合LSTM网络对序列信息变化敏感的特点，在构建更深层网络的同时避免模型过拟合。采用了最大互信息（Maximum mutual information coefficient，MMIC）数据预处理方法，提高了数据的局部相关性，提高了PLSTM-CNN模型从不同初始条件下检测故障的效率。以田纳西州伊斯曼(Eastman Process of Tennessee，TE)过程为研究对象，PLSTM-CNN模型在故障平均检测率和漏报率等指标明显优于传统循环神经网络。
- 故障检测 /
- 一维稠密卷积神经网络 /
- 长短时记忆网络 /
- 互信息 /
- TE过程
Abstract: In order to ensure the safe and stable operation of the production process and avoid losses due to failures, timely detection of abnormal conditions and accurate diagnosis of abnormal conditions are of very important research significance. Aiming at the complexity of the chemical process, this paper proposes a parallel long and short-term memory network and convolutional neural network (PLSTM-CNN) model for fault detection in the chemical production process. This model effectively combines the LSTM's ability to extract global features from time series data and the CNN model's ability to extract local features, reducing the loss of feature information and achieving a higher fault detection rate. The one-dimensional dense convolutional neural network is used as the main body of CNN, combined with the LSTM network's sensitivity to sequence information changes, to avoid model overfitting while building a deeper network. The maximum mutual information coefficient (MMIC) data preprocessing method is adopted to improve the local correlation of the data and improve the efficiency of the PLSTM-CNN model in detecting faults under different initial conditions. Taking the Eastman Process of Tennessee (TE) process in Tennessee as the research object, the PLSTM-CNN model is significantly better than the traditional recurrent neural network in indicators such as the average failure detection rate and the false negative rate.
- Fault detection /
- One-dimensional dense convolutional neural network /
- Long and short-term memory network /
- Mutual information /
- TE process

HTML全文

图 1 卷积层

Figure 1. Convolution layer

下载: 全尺寸图片幻灯片

图 2 稠密网络结构图

Figure 2. Dense network structure diagram

下载: 全尺寸图片幻灯片

图 3 LSTM结构

Figure 3. LSTM structure

下载: 全尺寸图片幻灯片

图 4 基于PLSTM-CNN故障诊断方法

Figure 4. Fault diagnosis method based on PLSTM-CNN

下载: 全尺寸图片幻灯片

图 5 MI网格计算方法

Figure 5. MI grid calculation method

下载: 全尺寸图片幻灯片

图 7 变量排序后的格式

Figure 7. Format of the sorted variables

下载: 全尺寸图片幻灯片

图 6 正常数据与故障数据互信息值

Figure 6. Mutual information value of normal data and fault data

下载: 全尺寸图片幻灯片

图 8 PLSTM-CNN网络模型

Figure 8. PLSTM-CNN network model

下载: 全尺寸图片幻灯片

图 9 基于t-SNE的故障诊断可视化

Figure 9. Fault diagnosis visualization using t-SNE

下载: 全尺寸图片幻灯片

图 10 串行LSTM-CNN网络结构

Figure 10. Serial LSTM-CNN network structure

下载: 全尺寸图片幻灯片

图 11 平均故障检测率

Figure 11. Mean failure detection rate

下载: 全尺寸图片幻灯片

表 1 故障检测结果比较

Table 1. The comparison of fault detection results

故障类型	FDR			FPR
故障类型	2D-CNN	LSTM	PLSTM-CNN	2D-CNN	LSTM	PLSTM-CNN
Normal	0.91	0.96	0.96	0.09	0.04	0
Fault1	1.0	1.0	1.0	0	0	0
Fault2	1.0	1.0	1.0	0	0	0
Fault3	0.48	0.75	0.92	0.44	0.15	0.08
Fault4	1.0	1.0	1.0	0	0	0
Fault5	1.0	1.0	1.0	0	0	0
Fault6	1.0	1.0	1.0	0	0	0
Fault7	1.0	1.0	1.0	0	0	0
Fault8	0.80	1.0	1.0	0.2	0	0
Fault9	0.34	0.36	0.85	0.58	0.23	0.18
Fault10	0.93	0.94	0.92	0.06	0.02	0.01
Fault11	0.96	0.93	0.97	0.04	0.03	0.02
Fault12	0.95	0.98	0.98	0.05	0.02	0.01
Fault13	0.82	0.95	0.96	0.18	0.04	0.04
Fault14	0.84	0.92	0.96	0.16	0.05	0.04
Fault15	0.04	0.12	0.37	0.8	0.64	0.51
Fault16	0.09	0.18	0.33	0.75	0.62	0.48
Fault17	0.96	0.97	1.0	0.04	0.02	0.03
Fault18	1.0	1.0	0.98	0	0	0
Fault19	1.0	1.0	1.0	0	0	0
Fault20	1.0	1.0	1.0	0	0	0
Average	0.83	0.86	0.914	0.16	0.088	0.066

下载: 导出CSV

表 2 训练、推理时间比较

Table 2. Comparison of training and inference time

Model	Training time for one epochs(s)	Reasoning time for one epochs(ms)
1D-CNN	2.54	10
2D-CNN	65	200
LSTM	3	12
串联 LSTM-CNN	3.8	20
并联 LSTM-CNN	4.2	25

下载: 导出CSV

表 3 小样本平均故障检测率

Table 3. Average failure detection rate of small samples

Model	FDR	FPR
1D-CNN	83.4%	0.09
2D-CNN	78.4%	0.07
LSTM	84.1%	0.085
串行LSTM-CNN	84.9%	0.12
并行LSTM-CNN	90.2%	0.054

下载: 导出CSV

参考文献(25)

[1]	刘强, 柴天佑, 秦泗钊, 等. 基于数据和知识的工业过程监视及故障诊断综述[J]. 控制与决策, 2010, 25(6): 801-807+813.
[2]	Dong Y, Qin S J. A novel dynamic PCA algorithm for dynamic data modeling and process monitoring[J]. Journal of Process Control, 2018, 67: 1-11. doi: 10.1016/j.jprocont.2017.05.002
[3]	Dong J, Zhang K, Huang Y, et al. Adaptive total PLS based quality-relevant process monitoring with application to the Tennessee Eastman process[J]. Neurocomputing, 2015, 154: 77-85. doi: 10.1016/j.neucom.2014.12.017
[4]	Ajami A, Daneshvar M. Data driven approach for fault detection and diagnosis of turbine in thermal power plant using Independent Component Analysis (ICA)[J]. International Journal of Electrical Power & Energy Systems, 2012, 43(1): 728-735.
[5]	Chiang L H, Russell E L, Braatz R D. Fault diagnosis in chemical processes using Fisher discriminant analysis, discriminant partial least squares, and principal component analysis[J]. Chemometrics and intelligent laboratory systems, 2000, 50(2): 243-252. doi: 10.1016/S0169-7439(99)00061-1
[6]	Venkatasubramanian V, Chan K. A neural network methodology for process fault diagnosis[J]. AIChE Journal, 1989, 35(12): 1993-2002. doi: 10.1002/aic.690351210
[7]	胡寿松, 王源. 基于支持向量机的非线性系统故障诊断[J]. 控制与决策, 2001, 16(5): 617-620. doi: 10.3321/j.issn:1001-0920.2001.05.025
[8]	Choi S W, Park J H, Lee I B. Process monitoring using a Gaussian mixture model via principal component analysis and discriminant analysis[J]. Computers & chemical engineering, 2004, 28(8): 1377-1387.
[9]	He Q P, Wang J. Fault detection using the k-nearest neighbor rule for semiconductor manufacturing processes[J]. IEEE transactions on semiconductor manufacturing, 2007, 20(4): 345-354. doi: 10.1109/TSM.2007.907607
[10]	何宁. 基于 ICA-PCA 方法的流程工业过程监控与故障诊断研究[D]. 浙江大学, 2004.
[11]	刘永斌, 何清波, 孔凡让, 等. 基于 PCA 和 SVM 的内燃机故障诊断[J]. 振动. 测试与诊断, 2012, 32(2): 250-255.
[12]	Xie D, Bai L. A hierarchical deep neural network for fault diagnosis on Tennessee-Eastman process[C]//2015 IEEE 14th International Conference on Machine Learning and Applications (ICMLA). IEEE, 2015: 745-748.
[13]	Zhang Z, Zhao J. A deep belief network based fault diagnosis model for complex chemical processes[J]. Computers & chemical engineering, 2017, 107: 395-407.
[14]	Wang Y, Pan Z, Yuan X, et al. A novel deep learning based fault diagnosis approach for chemical process with extended deep belief network[J]. ISA transactions, 2020, 96: 457-467. doi: 10.1016/j.isatra.2019.07.001
[15]	Wu H, Zhao J. Deep convolutional neural network model based chemical process fault diagnosis[J]. Computers & chemical engineering, 2018, 115: 185-197.
[16]	肖雄, 王健翔, 张勇军, 等. 一种用于轴承故障诊断的二维卷积神经网络优化方法[J]. 中国电机工程学报, 2019, 39(15): 4558-4568.
[17]	曲建岭, 余路, 袁涛, 等. 基于一维卷积神经网络的滚动轴承自适应故障诊断算法[J]. 仪器仪表学报, 2018, 39(7): 134-143.
[18]	S. Kiranyaz, A. Gastli, L. Ben-Brahim, N. Alemadi, M. Gabbouj, Real-time fault detection and identification for MMC using 1D convolutional neural networks, IEEE Trans. Ind. Electron, 2018, 8760-8771
[19]	Huang G, Liu Z, Van Der Maaten L, et al. Densely connected convolutional networks[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2017: 4700-4708.
[20]	Zaremba W, Sutskever I, Vinyals O. Recurrent neural network regularization[J]. arXiv preprint arXiv: 1409.2329, 2014.
[21]	Staudemeyer R C, Morris E R. Understanding LSTM--a tutorial into Long Short-Term Memory Recurrent Neural Networks[J]. arXiv preprint arXiv: 1909.09586, 2019.
[22]	Bathelt A, Ricker N L, Jelali M. Revision of the Tennessee Eastman process model[J]. IFAC-PapersOnLine, 2015, 48(8): 309-314. doi: 10.1016/j.ifacol.2015.08.199
[23]	Kraskov A, Stögbauer H, Grassberger P. Estimating mutual information[J]. Physical review E, 2004, 69(6): 066138. doi: 10.1103/PhysRevE.69.066138
[24]	Reshef D N, Reshef Y A, Finucane H K, et al. Detecting novel associations in large data sets[J]. science, 2011, 334(6062): 1518-1524. doi: 10.1126/science.1205438
[25]	Van der Maaten L, Hinton G. Visualizing data using t-SNE[J]. Journal of machine learning research, 2008, 9(11): 2579-2605.