工业数据驱动下薄板冷轧颤振的LSTM智能预报

doi:10.3901/JME.2020.11.121

摘要/Abstract

摘要： 冷轧机颤振问题由来已久，随着更薄规格、更高强度和更快速度的要求，该问题更加突出，诱发机理复杂多变且隐蔽，过度依赖经验判断难以及时有效地抑制颤振的发生，基于历史振动大数据实现冷轧颤振的智能预报是面向智能轧制的重要应用场景。考虑冷轧颤振工业数据的多源、多态与强耦合特征，通过多源数据的采集交互、时刻匹配、频率协同、数据降维等数据预处理建立颤振预报模型的样本空间；基于长短时记忆（Long short –term memory，LSTM）循环神经网络建立轧机颤振能量值的智能预报模型，利用轧件规格、轧辊状况、轧制工艺以及轧机振动状态的历史信息数据，对最为典型和振动频繁的第五机架振动能量值加以预测；分析了不同时间步长对预测效果的影响以获得最优的预测步长，模型预测结果的变化趋势与实际数据的变化趋势基本一致，训练数据集与测试数据集的均方误差较小；然后，将模型应用于未参与训练与测试的实际轧制过程数据，结果表明LSTM模型能够很好地实现颤振预测，且根据设定的报警阈值能够实现提前预报。研究结果表明基于多源历史振动数据的深度学习与挖掘能够实现对连轧颤振失稳的智能预报，不仅能够在实际生产中发挥作用，而且对实现冷轧过程的智能化具有积极的推动意义。

关键词: 轧机颤振, 工业大数据, 长短时记忆循环神经网络, 深度学习, 预报

Abstract: The chatter in the cold tandem rolling process is an old problem here, which is more prominent with the demand of thinner specifications, higher strength and higher rolling velocity. It is hard to rely too much on experiences to suppress chatter timely and effectively because of the complex and changeful induced mechanism, therefore the intelligent forecasting of chatter in cold rolling based on historical vibration big data is an important application scenario for intelligent rolling. Considering the multi-source, polymorphism and strong coupling characteristics of industrial data for cold rolling chatter, the sample space of chatter prediction model is built with data preprocess technique of multi-source data acquisition and interaction, moment matching, frequency coordination and data dimensionality reduction. The intelligent forecast model for the chatter is formulated based on LSTM recurrent neural network. Using the information of historical data of rolling product specifications, roll condition, rolling process parameters and vibration of mill, the vibration energy of fifth stand the most typical and frequently vibrating is predicted, and the influence of the number of time step on the prediction effect is analyzed to obtain the optimal number of prediction step. The trend of the results on the prediction of model is basically consistent with the trend of the actual data, the mean square errors of the training data set and the test data set are small. Then, the model is applied to the actual process data which are not involved in the training and testing process, and it is concluded that the LSTM model can effectively predict mill chatter in advance according to the alarm threshold. The results of research demonstrate that the deep learning and data mining based on multi-source historical vibration data can realize the intelligent forecast of chatter instability in the cold tandem rolling process, which not only play an important role in actual production, but have a promising future for promoting the intelligent of cold rolling process.

Key words: mill chatter, industrial big data, LSTM, deep learning, forecast

中图分类号:

TG133
TH113

刘阳, 郜志英, 周晓敏, 张清东. 工业数据驱动下薄板冷轧颤振的LSTM智能预报[J]. 机械工程学报, 2020, 56(11): 121-131.

LIU Yang, GAO Zhiying, ZHOU Xiaomin, ZHANG Qingdong. Industrial Data-driven Intelligent Forecast for Chatter of Cold Rolling of Thin Strip with LSTM Recurrent Neural Network[J]. Journal of Mechanical Engineering, 2020, 56(11): 121-131.

参考文献

[1] TLUSTY J,CRITCHLEY S,PATON D. Chatter in cold rolling[J]. Annals of the CIRP,1982,31(1):195-199.
[2] PATON D L,CRITCHLEY S. Tandem mill vibration:Its cause and control[J]. Iron and Steel Making,1985,12(3):37-43.
[3] YUN I S,WILSON W R D,EHMANN K F. Chatter in the strip rolling process[J]. Journal of Manufacturing Science and Engineering,1998,120(5):330-348.
[4] MEEHAN P A. Vibration instability in rolling mills:Modelling and experimental results[J]. Journal of Vibration and Acoustics,2002,124(2):221-228.
[5] HU Peihua,ZHAO Huyue,EHMANN K F. Third-octave-mode chatter in rolling chatter model[J]. Journal of Engineering Manufacture,2006,220(4):1267.
[6] 郜志英,白露露,李强. 薄板冷连轧自激振动的临界轧制速度研究[J]. 机械工程学报,2017,53(12):118-132. GAO Zhiying,BAI Lulu,LI Qiang. Research on critical rolling speed of self-excited vibration in the tandem rolling process of thin strip[J]. Journal of Mechanical Engineering,2017,53(11):118-132.
[7] 孙志辉,吕文泉. 基于形态非抽样小波和S变换的轧机振动信号分析[J]. 北京科技大学学报,2013,35(3):366-370. SUN Zhihui,LÜ Wenquan. Single analysis of rolling mill vibration based on morphological undecimated wavelets and s-transform[J]. Journal of University of Science and Technology Beijing,2013,35(3):366-370.
[8] 凌启辉,闫晓强,张义方. 基于S变换的热连轧机耦合振动特征提取[J].振动测试与诊断,2016,36(1):115-119,201-202. LING Qihui,YAN Xiaoqiang,ZHANG Yifang. Vibration feature extraction of hot continuous rolling based on s-transform[J]. Journal of Vibration,Measurement & Diagnosis,2016,36(1):115-119,201-202.
[9] 闫晓强. 热连轧机机电液耦合振动研究[J]. 机械工程学报,2011,47(17):61-65. YAN Xiaoqiang. Machinery-electric-hydraulic couplingvi-bration control of hot continuous rolling mills[J]. Journal of Mechanical Engineering,2011,47(17):61-65.
[10] YANG Jingming,ZHANG Qian,CHE Haijun,et al. Multi-objective optimization for tandem cold rolling schedule[J]. Journal of Iron and Steel Research International,2010,17(11):0-39.
[11] 郜志英,臧勇,曾令强. 轧机颤振建模及理论研究进展[J]. 机械工程学报,2015,51(16):87-105. GAO Zhiying,ZANG Yong,ZENG Lingqiang. Review of modeling and theoretical studies on chatter in the rolling mills[J]. Journal of Mechanical Engineering,2015,51(16):87-105.
[12] 雷亚国,贾峰,孔德同,等. 大数据下机械智能故障诊断的机遇与挑战[J]. 机械工程学报,2018,54(5):84-104. LEI Yaguo,JIA Feng,KONG Detong,et al. Oppor-tunities and challenges of machinery intelligent fault diagnosis in big data era[J]. Journal of Mechanical Engineering,2018,54(5):84-104.
[13] 殷瑞钰. 关于智能化钢厂的讨论:从物理系统一侧出发讨论钢厂智能化[J]. 钢铁,2017,52(6):1-12. YIN Ruiyu. A discussion on "smart" steel plant-view from physical system side[J]. Iron & Steel,2017,52(6):1-12.
[14] 郭朝晖. 钢铁行业推进智能制造的路径思考[N]. 中国冶金报,2018-11-28. GUO Zhaohui. Thinking on the path of steel industry to promote intelligent manufacturing[N]. New of Chinese Metallurgy,2018-11-28.
[15] DURBIN J. The fitting of time-series models[J]. Review of the International Statistical Institute,1960,28(3):233-244.
[16] AKAIKE H. Fitting autoregressive models for prediction[J]. Annals of the Institute of Statistical Mathematic,1969,21(1):243-247.
[17] DUONG Q P. On the choice of the order of autoregressive models:a ranking and selection approach[J]. Journal of Time Series Analysis,1984,5(3):145-157.
[18] DURBIN J. Efficient estimation of parameters in moving-average models[J]. Biometrika,1959,46(3/4):306-316.
[19] ROBINSON P. The estimation of a nonlinear moving average model[J]. Stochastic Processes and Their Applications,1977,5(1):81-90.
[20] TANAKA K. Testing for a moving-average unit root[J]. Econometric Theory,1990,6(4):433-444.
[21] PETER J B,RICHARD A D. Introduction to time series and forecasting[M]. New York:Springer Press,2002.
[22] DE GOOIJER J G,ABRAHAM B,GOULD A,er al. Methods of determining the order of an autoregressive-moving-average process:A survey[J]. International Statistical Review,1985,53(3):301-329.
[23] HANNAN E J. The estimation of the order of an ARMA process[J]. Annals of Applied Statistics,1980,8(5):1071-1081.
[24] GRAY H L,KELLEY G D,MCINTIRE D D. A new approach to ARMA modeling[J]. Communications in Statistics,1978,7(1):1-77.
[25] WILLIAMS R J,PENG J. An efficient gradient-based algorithm for on-line training of recurrent network trajectories[J]. Neural Computation,1990,2(4):490-501.
[26] YOSHUA B,PATRICE S,PAOLO F. Learning long-term dependencies with gradient descent is difficult[J]. Neural Networks,1994,5(2):157-166.
[27] SEPP H,JURGEN S. Long short-term memory[J]. Neural Computation,1997,9(8):1735-1780.
[28] GERS F A,SCHMIDHUBER J. Recurrent nets that time and count[J]. Proceedings of the IEEE-INNS-ENNS International Joint Conference on Neural Networks,2000,3:189-194.
[29] YANG Beibei,YIN Kunlong,SUZANNE L,et al. Time series analysis and long short-term memory neural network to predict landslide displacement[J]. Landslides,2019:1-18.
[30] MA Xiaolei,TAO Zhimin,WANG Yinhai,et al. Long short-term memory neural network for traffic speed prediction using remote microwave sensor data[J]. Transportation Research Part C:Emerging Technologies,2015,54:187-197.
[31] SAK H,SENIOR A,BEAUFAYS,et al. Long short-term memory based recurrent neural network architectures for large vocabulary speech recognition[J]. Computer Science,2014.