Nonlinear Degradation Modeling and Residual Life Prediction for Rollers Based on Kernel-based Wiener Process

doi:10.16183/j.cnki.jsjtu.2022.004

Abstract

Abstract:

In the process of steel rolling, due to wear and other reasons, the working performance of the roll under long and complex working conditions has a gradual decline. Considering the characteristics of complex working conditions and strong random interference of the roll working environment, this paper proposed a kernel-based Wiener process (KWP) degradation model to characterize the strong randomness of the roll degradation trend by using the Wiener process, and to capture the nonlinear degradation path of the roll by using the kerna function. This paper derives the analytical expression of parameter estimation in the Bayesian framework, and constructs the health index of the roll working rotation, then predicts the remaining useful life (RUL) of the roll. In combination with the field data of 1580 hot rolling production line of an iron and steel company, the goodness of fit of the model built is 0.989, and the residual life prediction error is less than 4.7%. Compared with the common machine learning algorithm, it has achieved better results, which is helpful to improve the operating efficiency and safety of equipment and achieve maintenance as needed.

Key words: roller, performance degradation, nonlinear modeling, kernel function, Wiener process, remaining useful life (RUL) prediction

CLC Number:

TH17

WANG Hanyu, CHEN Zhen, ZHOU Di, CHEN Zhaoxiang, PAN Ershun. Nonlinear Degradation Modeling and Residual Life Prediction for Rollers Based on Kernel-based Wiener Process[J]. Journal of Shanghai Jiao Tong University, 2023, 57(8): 1037-1045.

Figures/Tables 10

Tab.1

Kernel functions and formulas

核函数	表达式	参数
线性核函数	$t t i$	无特殊参数
高斯核函数	$e x p - t - t i \| 2 2 v 2$	v为高斯核函数的宽度
多项式核函数	$(t t i + r) d, ? d = 1,2, …, n$	$r 为偏置系数; d$ 为多项式阶数
Laplace核函数	$e x p - t - t i \| o$	o>0
Sigmoid核函数	$t a n h (ρ t t i + θ)$	$ρ > 0, θ < 0$

Tab.1

Fig.1

Fig.2

Fig.3

Tab.2

Tab.3

Fig.4

Tab.4

Tab.5

Fig.5

References 24

[1]	郑国栋, 陈其慎, 邢佳韵, 等. 典型国家钢铁产业发展路径与启示[J]. 中国国土资源经济, 2021, 34(8): 51-56.
	ZHENG Guodong, CHEN Qishen, XING Jiayun, et al. A prediction model for wear of working rolls in hot rolling mills[J]. Natural Resource Economics of China, 2021, 34(8): 51-56.
[2]	PRAKASH G. A Bayesian approach to degradation modeling and reliability assessment of rolling element bearing[J]. Communications in Statistics-Theory and Methods, 2021, 50(23): 5453-5474. doi: 10.1080/03610926.2020.1734826 URL
[3]	刘子英, 孙彦广, 宋向荣, 等. 热连轧轧机工作辊磨损预报模型[J]. 热加工工艺, 2015, 44(1): 131-133.
	LIU Ziying, SUN Yanguang, SONG Xiangrong, et al. Prediction model of work roll wear for hot tandem rolling mill[J]. Hot Working Technology, 2015, 44(1): 131-133.
[4]	CAI J X, CHENG X, ZHAO B J, et al. Study on the corrosion mechanism of the oxide scale on hot rolled steel in an atmospheric environment[J]. Anti-Corrosion Methods and Materials, 2019, 66(5): 613-620. doi: 10.1108/ACMM-03-2019-2102 URL
[5]	LIU ZY, GUAN YP, WANG FQ. Model development of work roll wear in hot strip mill[J]. Materials Science and Engineering, 2017, 207(1): 012022.
[6]	杨阳, 苏小平, 赵春磊. 热轧辊表面疲劳寿命研究[J]. 热加工工艺, 2016, 45(1): 151-154.
	YANG Yang, SU Xiaoping, ZHAO Chunlei. Research on fatigue life of hot roll surface[J]. Hot Working Technology, 2016, 45(1): 151-154.
[7]	李长生, 张晓明, 刘相华, 等. 轧制过程轧辊磨损数学模型试验研究[J]. 机械工程学报, 2002, 38(7): 28-30.
	LI Changsheng, ZHANG Xiaoming, LIU Xianghua, et al. Experimental investigation of mathematic model on rollswear in rolling[J]. Chinese Journal of Mechanical Engineering, 2002, 38(7): 28-30.
[8]	宋光义. 热轧带钢平整机工作辊磨损与工艺参数优化研究[D]. 北京: 北京科技大学, 2019.
	SONG Guangyi. Research on work roll wear and process parameter optimization for hot strip skin-pass mill[D]. Beijing: University of Science and Technology Beijing, 2019.
[9]	潘尔顺, 陈震. 高可靠性产品退化建模研究综述[J]. 工业工程与管理, 2015, 20(6): 1-6.
	PAN Ershun, CHEN Zhen. Review of degradation model for high reliability products[J]. Industrial Engineering and Management, 2015, 20(6): 1-6.
[10]	吴进, 邱春林, 齐克敏, 等. 热轧精轧轧辊摩擦和磨损研究[J]. 钢铁研究, 2006, 34(6): 31-34.
	WU Jin, QIU Chunlin, QI Kemin, et al. Investigation on friction and wear of finishing roll of hot rolling[J]. Research on Iron and Steel, 2006, 34(6): 31-34.
[11]	JIAO R H, PENG K X, DONG J. Remaining useful life prediction for a roller in a hot strip mill based on deep recurrent neural networks[J]. IEEE/CAA Journal of Automatica Sinica, 2021, 8(7): 1345-1354. doi: 10.1109/JAS.2021.1004051 URL
[12]	陶红玉, 周炳海. 基于随机退化的串行生产系统机会维护模型[J]. 上海交通大学学报, 2013, 47(12): 1911-1917.
	TAO Hongyu, ZHOU Binghai. Opportunistic maintenance model for series production systems based on stochastic degradations[J]. Journal of Shanghai Jiao Tong University, 2013, 47(12): 1911-1917.
[13]	GIORGIO M, POSTIGLIONE F, PULCINI G. Bayesian estimation and prediction for the transformed Wiener degradation process[J]. Applied Stochastic Models in Business and Industry, 2020, 36(4): 660-678. doi: 10.1002/asmb.v36.4 URL
[14]	司小胜, 胡昌华, 李娟, 等. 具有不确定测量的非线性随机退化系统剩余寿命预测[J]. 上海交通大学学报, 2015, 49(6): 855-860.
	SI Xiaosheng, HU Changhua, LI Juan, et al. Remaining useful life prediction of nonlinear stochastic degrading systems subject to uncertain measurements[J]. Journal of Shanghai Jiao Tong University, 2015, 49(6): 855-860.
[15]	GAO H D, CUI L R, KONG D J. Reliability analysis for a Wiener degradation process model under changing failure thresholds[J]. Reliability Engineering & System Safety, 2018, 171: 1-8. doi: 10.1016/j.ress.2017.11.006 URL
[16]	郭昊, 张田, 李亚平, 等. 基于逆高斯过程的竞争失效建模研究[J]. 工业工程与管理, 2017, 22(1): 89-94.
	GUO Hao, ZHANG Tian, LI Yaping, et al. Research on competing failure modeling based on the inverse Gaussian process[J]. Industrial Engineering and Management, 2017, 22(1): 89-94.
[17]	BANNA O, MISHURA Y, SHKLYAR S. Approximation of a wiener process by integrals with respect to the fractional Brownian motion of power functions of a given exponent[J]. Theory of Probability and Mathematical Statistics, 2015, 90: 13-22. doi: 10.1090/tpms/2015-90-00 URL
[18]	LI J X, WANG Z H, ZHANG Y B, et al. A nonlinear wiener process degradation model with autoregressive errors[J]. Reliability Engineering & System Safety, 2018, 173: 48-57. doi: 10.1016/j.ress.2017.11.003 URL
[19]	TANG J X, ZHENG G H, HE D, et al. Rolling bearing remaining useful life prediction via weight tracking relevance vector machine[J]. Measurement Science and Technology, 2021, 32(2): 12-20.
[20]	林杰, 叶鸿庆, 郑美妹, 等. 基于状态的预防性替换和备件订购联合优化[J]. 工业工程与管理, 2021, 26(6): 1-8.
	LIN Jie, YE Hongqing, ZHENG Meimei, et al. Joint condition-based preventive replacement and spare parts provisioning policy[J]. Industrial Engineering and Management, 2021, 26(6): 1-8.
[21]	张洋. 基于相关向量机的锂离子电池在线剩余寿命预测方法研究[D]. 长沙: 国防科学技术大学, 2016.
	ZHANG Yang. Online remaining useful life prediction of lithium-ion batteries based on relevance vector machine[D]. Changsha: National University of Defense Technology, 2016.
[22]	MACKAY D J C. Bayesian interpolation[J]. Neural Computation, 1992, 4(3): 415-447. doi: 10.1162/neco.1992.4.3.415 URL
[23]	ZHANG C L, HE Y G, YUAN L F, et al. Capacity prognostics of lithium-ion batteries using EMD denoising and multiple kernel RVM[J]. IEEE Access, 2017, 5: 12061-12070. doi: 10.1109/ACCESS.2017.2716353 URL
[24]	周炳海, 侍雨, 张于贤. 考虑质量的退化生产系统可用性中心维护决策[J]. 东北大学学报(自然科学版), 2021, 42(6): 814-820. doi: 10.12068/j.issn.1005-3026.2021.06.009
	ZHOU Binghai, SHI Yu, ZHANG Yuxian. Availability-centered maintenance policies for degrading manufacturing systems considering product quality[J]. Journal of Northeastern University (Natural Science), 2021, 42(6): 814-820.

轧辊编号	上机时间	下机时间	生产量/t	轧制距离/km	轧辊直径/mm
1580I1	2017-12-27T15:04	2017-12-27T17:36	1016	4.863	825.92
1580I2	2017-12-27T20:27	2017-12-27T22:42	739	4.803	825.92
1580I2	2017-12-28T1:04	2017-12-28T4:34	1427	6.588	825.83
1580I2	2017-12-28T12:26	2017-12-28T17:48	2570	14.504	825.83
1580I2	2017-12-29T10:32	2017-12-29T17:25	2360	14.013	825. 83

核函数类型	RMSE	CD	MAE
高斯核函数	1109.8	0.9893	865.1
多项式核函数	7713.1	0.4781	5916.9
线性核函数	10717.1	0.0000	7589.3
Laplace核函数	2046.5	0.8664	1475.4
Sigmoid核函数	10717.0	0.0000	7589.3

模型	RMSE	CD	MAE
基于高斯核函数的Wiener过程模型	1109.7	0.9893	865.1
指数Wiener过程模型	8059.8	0.4726	6369.4
基于高斯核函数的相关向量机模型	1992.7	0.8719	1591.2

时间序列	RUL预测值	RUL实际值	误差/%
No.80	53790.5	56537.7	4.9
No.100	52630.2	52021.6	1.2
No.120	40764.1	37671.3	8.2