LSTM Encoder-Decoder方法预测设备剩余使用寿命

赵志宏; 李晴; 李乐豪; 赵敬娇

doi:10.19818/j.cnki.1671-1637.2021.06.021

LSTM Encoder-Decoder方法预测设备剩余使用寿命

doi: 10.19818/j.cnki.1671-1637.2021.06.021

赵志宏^{1, 2,},
李晴¹,
李乐豪¹,
赵敬娇¹

1.
石家庄铁道大学信息科学与技术学院，河北石家庄 050043
2.
石家庄铁道大学省部共建交通工程结构力学行为与系统安全国家重点实验室，河北石家庄 050043

基金项目:

国家自然科学基金项目 11972236

国家自然科学基金项目 11790282

石家庄铁道大学研究生创新资助项目 YC2021077

详细信息

作者简介:
赵志宏(1972-)，男，河北石家庄人，石家庄铁道大学教授，工学博士，从事机械故障诊断与大数据分析研究

中图分类号: U270
计量
- 文章访问数: 1274
- HTML全文浏览量: 678
- PDF下载量: 113
- 被引次数: 0
出版历程
- 收稿日期: 2021-05-20
- 网络出版日期: 2022-02-11
- 刊出日期: 2021-12-01

Remaining useful life prediction for equipment based on LSTM encoder-decoder method

1.
School of Information Science and Technology, Shijiazhuang Tiedao University, Shijiazhuang 050043, Hebei, China
2.
State Key Laboratory of Mechanical Behavior and System Safety of Traffic Engineering Structures, Shijiazhuang Tiedao University, Shijiazhuang 050043, Hebei, China

Funds:

National Natural Science Foundation of China 11972236

National Natural Science Foundation of China 11790282

Graduate Innovation Support Project of Shijiazhuang Tiedao University YC2021077

More Information

Author Bio:
ZHAO Zhi-hong(1972-), male, professor, PhD, hb_zhaozhihong@126.com

摘要

摘要: 应用LSTM Encoder-Decoder提出了机械设备剩余使用寿命预测方法；对获取的传感器数据进行预处理，利用LSTM Encoder对数据序列进行编码，得到设备状态信息的中间表示，其中蕴含了设备状态的特征信息，利用LSTM Decoder对中间表示信息进行解码，利用解码后的信息预测剩余使用寿命；研究了LSTM Encoder-Decoder方法在公开的C-MAPSS数据集上的剩余使用寿命预测试验，与LSTM、D-LSTM等方法进行了对比试验；研究了不同滑动窗口大小对于剩余寿命预测结果的影响。研究结果表明：LSTM Encoder-Decoder方法的剩余使用寿命预测结果的评分函数值和均方根误差均优于LSTM、D-LSTM方法；在FD001子集上，LSTM Encoder-Decoder方法、LSTM方法和D-LSTM方法对应的均方根误差分别为11、12、16；当滑动窗口大小为30时，LSTM Encoder-Decoder方法在FD001~FD004子集对应的评分函数值分别为164、3 012、372、4 800，对应的均方根误差分别为11、20、14、22；当滑动窗口大小为40时，LSTM Encoder-Decoder方法在FD001~FD004子集对应的评分函数值分别为305、1 220、408、4 828，对应的均方根误差分别为14、16、15、19。可见，提出的LSTM Encoder-Decoder方法是一种有效的预测机械设备剩余使用寿命方法，并且滑动窗口大小对于剩余使用寿命预测结果存在一定的影响。
- 剩余使用寿命预测 /
- 编码器-解码器 /
- LSTM /
- 深度学习 /
- 特征提取
Abstract: A remaining useful life (RUL) prediction model of mechanical equipment was established based on the long short-term memory (LSTM) encoder-decoder method. The acquired sensor data were preprocessed. The data sequence was coded using the LSTM encoder method. An intermediate representation of the equipment status information was obtained. The characteristic information of the equipment status was obtained in the intermediate representation of the equipment status information. The intermediate representation information was decoded using the LSTM decoder method, and the RUL was predicted using the decoded information. RUL prediction experiments of the LSTM encoder-decoder method on open C-MAPSS data sets were performed. The LSTM encoder-decoder method was compared with the LSTM method, deep-LSTM (D-LSTM) method, and other methods. The effect of the sliding window size on RUL prediction results was evaluated. Research results show that scoring function values and root mean square error (RMSE) evaluation indexes of the RUL prediction results of the LSTM encoder-decoder method are more accurate than those of the LSTM method and D-LSTM method. In the FD001 subset, the RMSEs of the LSTM encoder-decoder method, LSTM method, and D-LSTM method are 11, 12, and 16, respectively. When the sliding window size is 30, the scoring function values corresponding to the FD001-FD004 subsets of the LSTM encoder-decoder method are 164, 3 012, 372, and 4 800, and the corresponding RMSEs are 11, 20, 14, and 22. When the sliding window size increases to 40, the respective scoring function values are 305, 1 220, 408, and 4 828, and the corresponding RMSEs are 14, 16, 15, and 19. Therefore, the proposed method based on the LSTM encoder-decoder effectively predicts the RUL of mechanical equipment, and the sliding window size significantly influences the RUL prediction results. 4 tabs, 6 figs, 32 refs.
- remaining useful life prediction /
- encoder-decoder /
- LSTM /
- deep learning /
- feature extraction

HTML全文

图 1 LSTM网络结构

Figure 1. LSTM network structure

下载: 全尺寸图片幻灯片

图 2 Encoder-Decoder框架

Figure 2. Encoder-Decoder structure

下载: 全尺寸图片幻灯片

图 3 基于LSTM Encoder-Decoder的RUL预测方法

Figure 3. RUL prediction method based on LSTM Encoder-Decoder

下载: 全尺寸图片幻灯片

图 4 FD001子集中21个传感器测量数据可视化结果

Figure 4. Data visualization results of 21 sensors in FD001 subset

下载: 全尺寸图片幻灯片

图 5 FD001~FD004子集测试集中剩余寿命实际值与预测值对比

Figure 5. Comparison between actual and predicted RULs in FD001-FD004 subsets' test sets

下载: 全尺寸图片幻灯片

图 6 每个测试数据集中不同发动机剩余寿命实际值与预测值对比

Figure 6. Comparison between actual and predicted RULs of different engines in each test data set

下载: 全尺寸图片幻灯片

表 1 C-MAPSS数据集

Table 1. C-MAPSS data sets

子集名称	FD001	FD002	FD003	FD004
训练发动机单元个数	100	260	100	249
测试发动机单元个数	100	259	100	248
操作条件	1	6	1	6
故障类型	1	1	2	2

下载: 导出CSV

表 2 每个子集26列具体含义

Table 2. Specific meanings of 26 columns in each subset

列数	1	2	3~5	6~26
具体含义	发动机单元号	当前工作周期数	操作设置	传感器值

下载: 导出CSV

表 3 八种方法与提出方法对比

Table 3. Comparison between eight methods and proposed method

数据集	评价指标	本文方法	D-LSTM^[27]	LSTM^[28]	MODBNE^[29]	TWBNN^[30]	FADCNN^[31]	GB^[29]	RF^[29]	SVM^[29]
FD001	评分函数	164	338		334		1 287	474	480	7 703
FD001	均方根误差	11	16	12	15	15	18	16	18	41
FD002	评分函数	3 012	4 450		5 585		13 570	87 280	70 457	316 483
FD002	均方根误差	20	24		25		30	29	30	53
FD003	评分函数	372	852		422		1 596	577	711	22 542
FD003	均方根误差	14	16		12		20	17	20	46
FD004	评分函数	4 800	5 550		6 558		7 886	17 818	46 568	141 122
FD004	均方根误差	22	28		29		29	29	31	60

下载: 导出CSV

表 4 滑动窗口大小对RUL的影响

Table 4. Effects of sliding windows sizes on RULs

数据集	评价指标	滑动窗口为10	滑动窗口为20	滑动窗口为30	滑动窗口为40	滑动窗口为50
FD001	评分函数	4 065	840	164	305	336
FD001	均方根误差	25	19	11	14	14
FD002	评分函数	3 693	4 124	3 012	1 220	1 107
FD002	均方根误差	19	20	20	16	15
FD003	评分函数	1 119	562	372	408	246
FD003	均方根误差	22	18	14	15	13
FD004	评分函数	8 697	9 559	4 800	4 828	3 116
FD004	均方根误差	24	23	22	19	19

下载: 导出CSV

参考文献(32)

[1]	KHAN S, YAIRI T. A review on the application of deep learning in system health management[J]. Mechanical Systems and Signal Processing, 2018, 107: 241-265. doi: 10.1016/j.ymssp.2017.11.024
[2]	年夫顺. 关于故障预测与健康管理技术的几点认识[J]. 仪器仪表学报, 2018, 39(8): 1-14. https://www.cnki.com.cn/Article/CJFDTOTAL-YQXB201808001.htm NIAN Fu-shun. Viewpoints about the prognostic and health management[J]. Chinese Journal of Scientific Instrument, 2018, 39(8): 1-14. (in Chinese) https://www.cnki.com.cn/Article/CJFDTOTAL-YQXB201808001.htm
[3]	ZHAO Ze-qi, LIANG Bin, WANG Xue-qian, et al. Remaining useful life prediction of aircraft engine based on degradation pattern learning[J]. Reliability Engineering and System Safety, 2017, 164: 74-83. doi: 10.1016/j.ress.2017.02.007
[4]	LIU Hui, LIU Zhen-yu, JIA Wei-qiang, et al. A novel deep learning-based encoder-decoder model for remaining useful life prediction[C]//IEEE. Proceedings of the 2019 International Joint Conference on Neural Networks. Washington DC: IEEE, 2019: 1-8.
[5]	LIM P, GOH C K, TAN K C. A time window neuralnetwork based framework for remaining useful life estimation[C]//IEEE. Proceedings of the 2016 International Joint Conference on Neural Networks. Washington DC: IEEE, 2016: 1746-1753.
[6]	LIAO Lin-xia, KÖTTIG F. Review of hybrid prognostics approaches for remaining useful life prediction of engineered systems, and an application to battery life prediction[J]. IEEE Transactions on Reliability, 2014, 63(1): 191-207. doi: 10.1109/TR.2014.2299152
[7]	朱朔, 白瑞林, 吉峰. 改进CHSMM的滚动轴承剩余寿命预测方法[J]. 机械传动, 2018, 42(10): 46-52, 95. https://www.cnki.com.cn/Article/CJFDTOTAL-JXCD201810010.htm ZHU Shuo, BAI Rui-lin, JI Feng. Rolling bearing remaining useful life prognosis method based on improved CHSMM[J]. Journal of Mechanical Transmission, 2018, 42(10): 46-52, 95. (in Chinese) https://www.cnki.com.cn/Article/CJFDTOTAL-JXCD201810010.htm
[8]	LI C J, LEE H. Gear fatigue crack prognosis using embedded model, gear dynamic model and fracture mechanics[J]. Mechanical Systems and Signal Processing, 2005, 19(4): 836-846. doi: 10.1016/j.ymssp.2004.06.007
[9]	FAN Jia-jie, YUNG K C, PECHT M. Physics-of-failure-basedprognostics and health management for high-power white light-emitting diode lighting[J]. IEEE Transactions on Device and Materials Reliability, 2011, 11(3): 407-416. doi: 10.1109/TDMR.2011.2157695
[10]	姜媛媛, 曾文文, 沈静静, 等. 基于凸优化-寿命参数退化机理模型的锂离子电池剩余使用寿命预测[J]. 电力系统及其自动化学报, 2019, 31(3): 23-28. https://www.cnki.com.cn/Article/CJFDTOTAL-DLZD201903004.htm JIANG Yuan-yuan, ZENG Wen-wen, SHEN Jing-jing, et al. Prediction of remaining useful life of lithium-ion battery based on convex optimization-life parameter degradation mechanism model[J]. Proceedings of the CSU-EPSA, 2019, 31(3): 23-28. (in Chinese) https://www.cnki.com.cn/Article/CJFDTOTAL-DLZD201903004.htm
[11]	张继冬, 邹益胜, 邓佳林, 等. 基于全卷积层神经网络的轴承剩余寿命预测[J]. 中国机械工程, 2019, 30(18): 2231-2235. https://www.cnki.com.cn/Article/CJFDTOTAL-ZGJX201918014.htm ZHANG Ji-dong, ZOU Yi-sheng, DENG Jia-lin, et al. Bearing remaining life prediction based on full convolutional layer neural networks[J]. China Mechanical Engineering, 2019, 30(18): 2231-2235. (in Chinese) https://www.cnki.com.cn/Article/CJFDTOTAL-ZGJX201918014.htm
[12]	陈自强. 基于LSTM网络的设备健康状况评估与剩余寿命预测方法的研究[D]. 合肥: 中国科学技术大学, 2019. CHEN Zi-qiang. Research on equipment health assessment and remaining useful life prediction method based on LSTM[D]. Hefei: University of Science and Technology of China, 2019. (in Chinese)
[13]	KONG Zheng-min, CUI Yan-de, XIA Zhou, et al. Convolution and long short-term memory hybrid deep neural networks for remaining useful life prognostics[J]. Applied Sciences, 2019, DOI: 10.3390/app9194156.
[14]	葛阳, 郭兰中, 牛曙光, 等. 基于t-SNE和LSTM的旋转机械剩余寿命预测[J]. 振动与冲击, 2020, 39(7): 223-231, 273. https://www.cnki.com.cn/Article/CJFDTOTAL-ZDCJ202007032.htm GE Yang, GUO Lan-zhong, NIU Shu-guang, et al. Prediction of remaining useful life based on t-SNE and LSTM for rotating machinery[J]. Journal of Vibration and Shock, 2020, 39(7): 223-231, 273. (in Chinese) https://www.cnki.com.cn/Article/CJFDTOTAL-ZDCJ202007032.htm
[15]	康守强, 周月, 王玉静, 等. 基于改进SAE和双向LSTM的滚动轴承RUL预测方法[J]. 自动化学报, 2020, DOI:10.16383/j.aas.c190796. KANG Shou-qiang, ZHOU Yue, WANG Yu-jing, et al. RUL prediction method of a rolling bearing based on improved SAE and Bi-LSTM[J]. Acta Automatica Sinica, 2020, DOI:10.16383/j.aas.c190796.(in Chinese)
[16]	SCHMIDHUBER J. Deep learning in neural networks: an overview[J]. Neural Networks, 2015, 61: 85-117.
[17]	CHO K, VAN MERRIENBOER B, GULCEHRE C, et al. Learning phrase representations using RNN encoder-decoder for statistical machine translation[J]. Computer Science, 2014, DOI: 10.3115/v1/D14-1179.
[18]	ZHU Yong-hua, ZHANG Wei-lin, CHEN Yi-ha, et al. A novel approach to workload prediction using attention-based LSTM encoder-decoder network in cloud environment[J]. EURASIP Journal on Wireless Communications and Networking, 2019(1): 1-18.
[19]	BENGIO Y, SIMARD P, FRASCONI P. Learning long-term dependencies with gradient descent is difficult[J]. IEEE Transactions on Neural Networks, 1994, 5(2): 157-166. doi: 10.1109/72.279181
[20]	HOCHREITER S, SCHMIDHUBER J. Long short-term memory[J]. Neural Computation, 1997, 9(8): 1735-1780. doi: 10.1162/neco.1997.9.8.1735
[21]	CHEN Yuan-hang, PENG Gao-liang, ZHU Zhi-yu, et al. A novel deep learning method based on attention mechanism for bearing remaining useful life prediction[J]. Applied Soft Computing Journal, 2019, DOI:https://doi.org/ 10.1016/j.asoc.2019.105919.
[22]	李敏, 李红娇, 陈杰. 差分隐私保护下的Adam优化算法研究[J]. 计算机应用与软件, 2020, 37(6): 253-258, 296. doi: 10.3969/j.issn.1000-386x.2020.06.044 LI Min, LI Hong-jiao, CHEN Jie. Adam optimization algorithm based on differential privacy protection[J]. Computer Applications and Software, 2020, 37(6): 253-258, 296. (in Chinese) doi: 10.3969/j.issn.1000-386x.2020.06.044
[23]	GOU Peng-qi, YU Jian-jun. A nonlinear ANN equalizer with mini-batch gradient descent in 40Gbaud PAM-8 IM/DD system[J]. Optical Fiber Technology, 2018, 46: 113-117. doi: 10.1016/j.yofte.2018.09.015
[24]	李杰, 贾渊杰, 张志新, 等. 基于融合神经网络的航空发动机剩余寿命预测[J]. 推进技术, 2021, 42(8): 1725-1734. https://www.cnki.com.cn/Article/CJFDTOTAL-TJJS202108007.htm LI Jie, JIA Yuan-jie, ZHANG Zhi-xin, et al. Remaining useful life prediction of aeroengine based on fusion neural network[J]. Journal of Propulsion Technology, 2021, 42(8): 1725-1734. (in Chinese) https://www.cnki.com.cn/Article/CJFDTOTAL-TJJS202108007.htm
[25]	ZHAO Zhi-bin, WU Jing-yao, LI Tian-fu, et al. Challenges and opportunities of AI-enabled monitoring, diagnosis and prognosis: a review[J]. Chinese Journal of Mechanical Engineering, 2021, 34(3): 16-44.
[26]	SAXENA A, GOEBEL K, SIMON D, et al. Damage propagation modeling for aircraft engine run-to-failure simulation[C]//IEEE. 2008 International Conference on Prognostics and Health Management. Washington DC: IEEE, 2008: 1-9.
[27]	ZHENG Shuai, RISTOVSKI K, FARAHAT A, et al. Long short-term memory network for remaining useful life estimation[C]//IEEE. 2017 IEEE International Conference on Prognostics and Health Management (ICPHM). Washington DC: IEEE, 2017: 88-95.
[28]	HSU C S, JIANG J R. Remaining useful life estimation using long short-term memory deep learning[C]//IEEE. International Conference on Applied System Invention. Washington DC: IEEE, 2018: 58-61.
[29]	ZHANG C, LIM P, QIN A K, et al. Multiobjective deep belief networks ensemble for remaining useful life estimation in prognostics[J]. IEEE Transactions on Neural Networks and Learning Systems, 2017, 28(10): 2306-2318. doi: 10.1109/TNNLS.2016.2582798
[30]	LIM P, GOH C K, TAN K C, A time window neural network based framework for remaining useful life estimation[C]//IEEE. International Joint Conference on Neural Networks. Washington DC: IEEE, 2016: 1746-1753.
[31]	BABU G S, ZHAO Pei-lin, LI Xiao-li. Deep convolutional neural network based regression approach for estimation of remaining useful life[C]//Springer. International Conference on Database Systems for Advanced Applications. Berlin: Springer, 2016: 214-228.
[32]	AL-DULAIMI A, ZABIHI S, ASIF A, et al. A multimodal and hybrid deep neural network model for remaining useful life estimation[J]. Computers in Industry, 2019, 108: 186-196. doi: 10.1016/j.compind.2019.02.004