基于退化感知时序建模的装备维保时机预测方法

doi:10.11996/JG.j.2095-302X.2025061233

图学学报 ›› 2025, Vol. 46 ›› Issue (6): 1233-1246.DOI: 10.11996/JG.j.2095-302X.2025061233

• 制造产品核心工业软件 • 上一篇下一篇

基于退化感知时序建模的装备维保时机预测方法

薄文¹^,²(), 琚晨², 刘维青³, 张焱⁴, 胡晶晶¹, 程婧晗², 张常有²()

¹ 北京理工大学计算机学院，北京 100081
² 中国科学院软件研究所，北京 100190
³ 中铁十九局集团有限公司，北京 100176
⁴ 新疆大学软件学院，新疆乌鲁木齐 830046

收稿日期:2025-09-10 接受日期:2025-11-04 出版日期:2025-12-30 发布日期:2025-12-27
通讯作者:张常有(1970-)，男，研究员，博士。主要研究方向为工业仿真软件和并行计算等。E-mail：changyou@iscas.ac.cn
第一作者:薄文(1995-)，男，博士研究生。主要研究方向为系统仿真与工业智能。E-mail：bowen@iscas.ac.cn
基金资助:
国家重点研发计划(2023YFB3611303);中华人民共和国水利部重大项目(SKS-2022104)

Degradation-driven temporal modeling method for equipment maintenance interval prediction

BO Wen¹^,²(), JU Chen², LIU Weiqing³, ZHANG Yan⁴, HU Jingjing¹, CHENG Jinghan², ZHANG Changyou²()

¹ School of Computer Science & Technology, Beijing Institute of Technology, Beijing 100081, China
² Institute of Software, Chinese Academy of Sciences, Beijing 100190, China
³ China Railway 19 Bureau Group Co., Ltd., Beijing 100176, China
⁴ School of Software, Xinjiang University, ürümqi Xinjiang 830046, China

Received:2025-09-10 Accepted:2025-11-04 Published:2025-12-30 Online:2025-12-27
First author：BO Wen (1995-)，PhD candidate. His main research interests cover system simulation and industrial intelligence. E-mail：bowen@iscas.ac.cn
Supported by:
National Key Research and Development Program of China(2023YFB3611303);Major Project of the Ministry of Water Resources of the People’s Republic of China(SKS-2022104)

摘要/Abstract

摘要：

维保时机强调装备停机的主动性，是在性能退化达到预设值前，结合工程节奏合理安排停机检修。该任务的精准预测对装备可靠运行至关重要，但仍面临多源数据融合、退化特征量化难及长依赖学习等挑战。因此提出一种基于退化感知时序建模的装备维保时机预测方法，以动态表征装备连续运行过程中的性能退化，并自适应捕获多传感器数据间的深层依赖关系。首先，提出性能退化指标(PDI)，通过时序数据驱动的性能量化器，实现动态的装备性能衰减感知；然后，构建基于多头注意力机制与序列到序列的维保时机预测模型，以自适应学习多源特征的相关性；最后，融合退化感知参数以强化特征权重分配，提升模型对装备长期运行趋势的预测能力。实验结果表明，融合PDI后模型最佳性能提升近13.5%，在隧道掘进机(TBM)工程数据集上较标准长短期记忆网络(LSTM)的均方根误差(RMSE)提升约25%，相比其他模型提升近15%以上，实现了较高的预测精度。在C-MAPSS数据集上与循环神经网络(RNN)和图神经网络(GNN)及注意力机制等主流时序预测方法进行了对比验证，结果表明该方法在维保时机预测任务中表现最优，并详细分析不同传感器数量对模型性能的影响。此外，该方法具备良好的可扩展性，可进一步融合装备运行环境信息感知，为装备的智能运维决策与操控闭环提供技术支撑。

关键词: 维保时机预测, 多头注意力机制, 序列到序列, 深度学习, 智能运维

Abstract:

Maintenance-interval prediction focuses on the proactive scheduling of equipment downtime, arranging maintenance before performance degradation reaches a predefined threshold while aligning with engineering operations. Accurate prediction of such intervals is vital for reliable equipment operation but remains challenging due to difficulties in multi-source data fusion, quantitative degradation characterization, and long-term dependency learning. This study presented a degradation-driven temporal modeling method that dynamically represented performance deterioration during continuous operation and adaptively captured complex dependencies among multi-sensor data. A performance-degradation indicator (PDI) quantified equipment performance decline using time-series measurements. To capture correlations among multi-source features, a sequence-to-sequence prediction model with multi-head attention was constructed and degradation-aware parameters were integrated, which optimized feature weighting and improved long-term trend prediction. The experimental results indicated that the optimal performance of the model improved by nearly 13.5% after integrating PDI. On the TBM (tunnel boring machine) engineering dataset, an RMSE (root mean square error) improvement of approximately 25% was achieved compared to the standard LSTM (long short-term memory), and outperformed other models by nearly 15%, yielding high prediction accuracy. Further evaluation on the C-MAPSS dataset against RNN (recurrent neural network), GNN (graph neural network), and attention-based methods confirmed the approach’s effectiveness, offering a detailed analysis of how varying the number of sensors affected model performance. The method also exhibited strong scalability and could be extended to incorporate environmental-condition awareness, providing technical support for intelligent maintenance decision-making and closed-loop operational control.

Key words: maintenance interval prediction, multi-head attention mechanism, sequence-to-sequence model, deep learning, intelligent maintenance systems

中图分类号:

TH17

薄文, 琚晨, 刘维青, 张焱, 胡晶晶, 程婧晗, 张常有. 基于退化感知时序建模的装备维保时机预测方法[J]. 图学学报, 2025, 46(6): 1233-1246.

BO Wen, JU Chen, LIU Weiqing, ZHANG Yan, HU Jingjing, CHENG Jinghan, ZHANG Changyou. Degradation-driven temporal modeling method for equipment maintenance interval prediction[J]. Journal of Graphics, 2025, 46(6): 1233-1246.

图/表 19

图1 基于退化感知时序建模的装备维保时机预测方法框架图

Fig. 1 Framework diagram of degradation-driven temporal modeling method for equipment maintenance interval prediction

图2 装备多传感器时序数据的PDI

Fig. 2 PDI for equipment multi-sensor time-series data

图3 MHA-Seq2Seq 的网络结构

Fig. 3 The network structure of MHA-Seq2Seq

表1 TBM数据集传感器信息

Table 1 TBM dataset of sensors information

编号	参数
S1	刀盘滚动角
S2	电机电流
S3	掘进速度
S4	穿透力
S5	贯入度
S6	姿态水平(尾)
S7	刀盘转速
S8	刀盘转矩
S9	姿态垂直(尾)
S10	姿态垂直(首)
S11	推进推力
S12	姿态水平(首)
S13	左侧夹持器的杆侧压力
S14	右侧夹持器的杆侧压力

表2 TBM数据集信息

Table 2 TBM dataset information

数据集	TBM-1	TBM-2
训练样本数量	10423	11126
测试样本数量	428	5784
训练周期数	173	196
测试周期数	126	139

图4 TBM数据集的相关矩阵热图((a) 数据集TBM-1；(b) 数据集TBM-2)

Fig. 4 The correlation matrix heat map of TBM dataset ((a) TBM-1 dataset; (b) TBM-2 dataset)

图5 TBM数据集的PDI趋势图((a) 数据集TBM-1；(b) 数据集TBM-2)

Fig. 5 PDI trend chart of the TBM datasets ((a) TBM-1 datasets; (b) TBM-2 datasets)

图6 预测值与实际值的比较((a) 数据集TBM-1；(b) 数据集TBM-2)

Fig. 6 Comparison of predicted and actual values ((a) TBM-1 dataset; (b) TBM-2 dataset)

表3 TBM数据集上的实验结果

Table 3 Experiment results of TBM dataset

模型	MAE	RMSE
LSTM	11.57	16.21
RNN	10.03	14.01
Attention+LSTM	10.50	14.83
Ours	9.32	12.14

表4 PDI指标消融实验结果

Table 4 PDI index ablation experiment results

模型	MAE	RMSE
Ours without PDI	9.78	13.78
LSTM+PDI	10.67	14.53
RNN+PDI	9.92	13.52
Attention+LSTM +PDI	9.83	13.35

表5 CMPASS数据集传感器信息

Table 5 CMPASS dataset of sensors information

编号	符号	描述
S1	T2	风机入口总温度
S2	T24	低压压气机出口总温度
S3	T30	高压压气机出口总温度
S4	T50	低压涡轮出口的总温度
S5	P2	风机入口压力
S6	P15	旁通管道中的总压力
S7	P30	高压压缩机出口总压力
S8	Nf	物理风扇转速
S9	Nc	物理核心速度
S10	epr	发动机压力比

图7 C-MAPSS数据集的相关矩阵热图((a) 子数据集FD001；(b) 子数据集FD002；(c) 子数据集FD003；(d) 子数据集FD004)

Fig. 7 The correlation matrix heat map of C-MAPSS dataset ((a) FD001; (b) FD002; (c) FD003; (d) FD004)

图8 C-MAPSS数据集的PDI趋势图((a) 子数据集FD001；(b) 子数据集FD002；(c) 子数据集FD003；(d) 子数据集FD004)

Fig. 8 PDI trend chart of the C-MAPSS dataset ((a) FD001; (b) FD002; (c) FD003; (d) FD004)

表6 不同代表性算法的C-MAPSS结果对比

Table 6 Comparison of C-MAPSS results from different representative algorithms

方法名称	FD001		FD002		FD003		FD004
方法名称	RMSE	Score	RMSE	Score	RMSE	Score	RMSE	Score
Hybrid	14.53	322	21.37	3077	13.24	367	27.08	5649
MODBNE	15.04	334	25.05	5585	12.51	421	28.66	6557
AConvLSTM	13.10	262	14.11	737	12.13	276	14.64	1011
KDNet	13.68	362	12.95	929	12.95	327	15.96	1303
GCN	12.58	237	13.78	849	11.92	218	14.44	967
AutoFormer	23.04	1063	16.51	1248	25.40	2034	20.30	2291
DAGN	16.11	595	16.43	1242	18.05	1216	19.04	2321
CNN-LSTM	12.47	257	15.02	835	12.42	252	14.49	870
SCTA-LSTM	12.10	207	16.90	1267	12.14	248	21.93	3310
GAT-DAT	13.83	319	14.85	1163	14.85	438	16.80	1928
Ours	12.02	204	12.86	835	11.92	215	13.74	838

图9 PDI消融结果可视图

Fig. 9 PDI ablation result visualization

表7 不同传感器数量规模的时序预测结果

Table 7 Prediction results with different sensors

数量	FD001		FD002		FD003		FD004
数量	RMSE	Score	RMSE	Score	RMSE	Score	RMSE	Score
4	15.27	376	15.18	1181	15.59	610	16.37	1261
7	14.64	333	14.57	1109	14.02	334	15.49	1037
10	12.99	241	13.34	949	13.28	266	14.67	952
14	12.24	204	12.92	882	12.48	220	13.92	891

表8 相关性筛选策略结果

Table 8 Results of correlation screening strategy

策略	FD001		FD002		FD003		FD004
策略	RMSE	Score	RMSE	Score	RMSE	Score	RMSE	Score
随机去除	14.64	333	14.57	1 109	14.02	334	15.49	1 037
去除低相关性	14.61	340	13.82	1 004	14.76	358	14.68	910
去除高相关性	13.67	317	15.99	1 398	13.85	453	17.57	1 455
高与低相关性去除	16.12	426	15.35	1 187	13.88	377	15.76	1 032

表9 不同注意力机制对比结果

Table 9 Rresults of different attention mechanisms

策略	FD001		FD002		FD003		FD004
策略	RMSE	Score	RMSE	Score	RMSE	Score	RMSE	Score
单头注意力	13.13	270	12.96	880	12.87	228	14.29	969
多头注意力	12.24	204	12.92	882	12.48	220	13.92	891

图10 TBM多模态时序数据信息采集效果图((a) 围岩地质类型、颗粒分布数量；(b) 岩渣尺寸；(c) 岩渣体积)

Fig. 10 Effect diagram of TBM multimodal time-series data information collection ((a) Surrounding rock geological type, particle distribution quantity; (b) Rock debris size; (c) Rock debris volume)

参考文献 33

[1]	HUANG X, CHEN W W, QU D R, et al. Remaining useful life prediction method based on multisensor fusion under time-varying operating conditions[J]. IEEE Transactions on Instrumentation and Measurement, 2024, 73: 3516413.
[2]	JAVID A, ZARKESH M, BOROUMAND Y, AL-FAWA’REH M. Enhancing safety and minimizing risk in mining processes with artificial intelligence[M]//RAZMJOU A, ASADNIA M, eds. Artificial Intelligence in Future Mining. Cambridge: Academic Press, 2025: 355-382.
[3]	WEST J, SIDDHPURA M, EVANGELISTA A, HADDAD A. Improving equipment maintenance—switching from corrective to preventative maintenance strategies[J]. Buildings, 2024, 14: 3581. DOI URL
[4]	鲜思渔, 赵泽田, 吴轩宇, 等. 基于改进红鸢优化算法与LSTM的核电换热器寿命预测方法[J]. 图学学报, 2025, 46(5): 1085-1093. DOI
	XIAN S Y, ZHAO Z T, WU X Y, et al. Life prediction method of nuclear power heat exchanger based on improved red kite optimization algorithm and LSTM[J]. Journal of Graphics, 2025, 46(5): 1085-1093 (in Chinese). DOI
[5]	EANG C, LEE S. Predictive maintenance and fault detection for motor drive control systems in industrial robots using CNN-RNN-based observers[J]. Sensors, 2025, 25(1): 25. DOI URL
[6]	GE R H, ZHAI Q Q, WANG H, et al. Wiener degradation models with scale-mixture normal distributed measurement errors for RUL prediction[J]. Mechanical Systems and Signal Processing, 2022, 173: 109029. DOI URL
[7]	LI Z, ZHU R, VERWIMP T, et al. Estimation of remaining useful life of rolling element bearings based on the adaptive kernel Kalman filter[J]. Mechanical Systems and Signal Processing, 2025, 229: 112493. DOI URL
[8]	KUNDU P, DARPE A K, KULKARNI M S. Weibull accelerated failure time regression model for remaining useful life prediction of bearing working under multiple operating conditions[J]. Mechanical Systems and Signal Processing, 2019, 134: 106302. DOI URL
[9]	ZHU K P, LI X, LI S S, et al. Physics-informed hidden Markov model for tool wear monitoring[J]. Journal of Manufacturing Systems, 2024, 72: 308-322. DOI URL
[10]	GARCIA J, RIOS-COLQUE L, PEÑA A, et al. Condition monitoring and predictive maintenance in industrial equipment: an NLP-assisted review of signal processing, hybrid models, and implementation challenges[J]. Applied Sciences, 2025, 15(10): 5465. DOI URL
[11]	LIN T T, REN Z J, ZHU L B, et al. A systematic review of multi-sensor information fusion for equipment fault diagnosis[EB/OL]. [2025-07-10]. https://ieeexplore.ieee.org/abstract/document/10840346.
[12]	DONG S J, XIAO J F, HU X L, et al. Deep transfer learning based on Bi-LSTM and attention for remaining useful life prediction of rolling bearing[J]. Reliability Engineering & System Safety, 2023, 230: 108914. DOI URL
[13]	WAN A P, ZHANG H, CHEN T, et al. Aeroengine life prediction and status evaluation based on sequential multitask learning and health indicators[J]. IEEE Transactions on Reliability, 2025, 74(3): 3833-3846. DOI URL
[14]	LIU L, SONG X, ZHOU Z T. Aircraft engine remaining useful life estimation via a double attention-based data-driven architecture[J]. Reliability Engineering & System Safety, 2022, 221: 108330. DOI URL
[15]	SUTSKEVER I, VINYALS O, LE Q V. Sequence to sequence learning with neural networks[C]// The 28th International Conference on Neural Information Processing Systems. Cambridge: MIT Press, 2014: 3104-3112.
[16]	VASWANI A, SHAZEER N, PARMAR N, et al. Attention is all you need[C]// The 31st International Conference on Neural Information Processing Systems. Red Hook: Curran Associates Inc., 2017: 6000-6010.
[17]	吴沛宸, 袁立宁, 胡皓, 等. 基于注意力特征融合的视频异常行为检测[J]. 图学学报, 2024, 45(5): 922-929. DOI
	WU P C, YUAN L N, HU H, et al. Video anomaly detection based on attention feature fusion[J]. Journal of Graphics, 2024, 45(5): 922-929 (in Chinese). DOI
[18]	AWAISI K S, YE Q, SAMPALLI S. A survey of industrial AIoT: opportunities, challenges, and directions[J]. IEEE Access, 2024, 12: 96946-96996. DOI URL
[19]	何庆, 荆传玉, 孙华坤, 等. 基于BIM和语义网的轨道智能运维管理方法[J]. 图学学报, 2024, 45(3): 601-612. DOI
	HE Q, JING C Y, SUN H K, et al. An intelligent railway operation and maintenance management approach based on BIM and semantic web[J]. Journal of Graphics, 2024, 45(3): 601-612 (in Chinese). DOI
[20]	ZHANG Q, LIU J, LIU Q. A novel multi-source domain transfer learning method for remaining useful life prediction using multi-scale attention-based temporal convolutional network[C]// The 30th International Conference on Mechatronics and Machine Vision in Practice. New York: IEEE Press, 2024: 1-5.
[21]	REN H J, DAI Z X, CHEN J X, et al. Natural gas triethylene glycol dehydration equipment digital twin and condition evaluation application[J]. IEEE Transactions on Industrial Informatics, 2024, 20(10): 12147-12156. DOI URL
[22]	XIAO X Q, LI C S, HE H X, et al. Rotating machinery fault diagnosis method based on multi-level fusion framework of multi-sensor information[J]. Information Fusion, 2025, 113: 102621. DOI URL
[23]	MIKOŁAJEWSKA E, MIKOŁAJEWSKI D, MIKOŁAJCZYK T, et al. Generative AI in AI-based digital twins for fault diagnosis for predictive maintenance in industry 4.0/5.0[J]. Applied Sciences, 2025, 15(6): 3166. DOI URL
[24]	ABED A I, PING L W. Implementing data mining techniques for gas-turbine (GT) health tracking and life management: the bibliographic perspective[J]. Expert Systems with Applications, 2024, 252: 124077. DOI URL
[25]	WANG X Q, LIU M Z, LIU C H, et al. Data-driven and knowledge-based predictive maintenance method for industrial robots for the production stability of intelligent manufacturing[J]. Expert Systems with Applications, 2023, 234: 121136. DOI URL
[26]	LI Y S, ZHOU Z, HU C Y, et al. Sequence to sequence network with Bayesian attention and state transition for self-data-driven remaining useful life estimation[J]. Expert Systems with Applications, 2025, 286: 128165. DOI URL
[27]	LU O H T. A Seq2Seq transformation strategy for generalizing a pre-trained model in anomaly detection of rolling element bearings[J]. Expert Systems with Applications, 2024, 254: 124297. DOI URL
[28]	ZHANG T C, FU T, NI T, et al. Data-driven excavation trajectory planning for unmanned mining excavator[J]. Automation in Construction, 2024, 162: 105395. DOI URL
[29]	XU D, XIAO X Q, LIU J, et al. Spatio-temporal degradation modeling and remaining useful life prediction under multiple operating conditions based on attention mechanism and deep learning[J]. Reliability Engineering & System Safety, 2023, 229: 108886. DOI URL
[30]	ZHANG Y, LI C B, SU L, et al. Degradation trend prediction for centrifugal blowers based on multi-sensor information fusion and attention mechanism[J]. Expert Systems with Applications, 2025, 276: 127195. DOI URL
[31]	WANG Y W, DENG L, ZHENG L Y, et al. Temporal convolutional network with soft thresholding and attention mechanism for machinery prognostics[J]. Journal of Manufacturing Systems, 2021, 60: 512-526. DOI URL
[32]	XIAO Y T, YIN H S, ZHANG Y D, et al. A dual-stage attention-based Conv-LSTM network for spatio-temporal correlation and multivariate time series prediction[J]. International Journal of Intelligent Systems, 2021, 36(5): 2036-2057. DOI URL
[33]	YANG T G, LI G C, LI K T, et al. The LPST-Net: a new deep interval health monitoring and prediction framework for bearing-rotor systems under complex operating conditions[J]. Advanced Engineering Informatics, 2024, 62: 102558. DOI URL

基于退化感知时序建模的装备维保时机预测方法

Degradation-driven temporal modeling method for equipment maintenance interval prediction

RichHTML

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

图/表 19

参考文献 33

相关文章 15

编辑推荐

Metrics

本文评价

[1]	琚晨, 丁嘉欣, 王泽兴, 李广钊, 管振祥, 张常有. 面向有限元法的图神经网络形函数近似方法[J]. 图学学报, 2025, 46(6): 1161-1171.
[2]	易斌, 张立斌, 刘丹楹, 唐军, 方俊俊, 李雯琦. 基于AMTA-Net的卷制过程激光打孔通风率预测模型[J]. 图学学报, 2025, 46(6): 1224-1232.
[3]	赵振兵, 欧阳文斌, 冯烁, 李浩鹏, 马隽. 基于类内稀疏先验与改进YOLOv8的绝缘子红外图像检测方法[J]. 图学学报, 2025, 46(6): 1247-1256.
[4]	贺蒙蒙, 张小艳, 李洪安. 基于Mamba结构的轻量级皮肤病变图像分割网络[J]. 图学学报, 2025, 46(6): 1257-1266.
[5]	李星辰, 李宗民, 杨超智. 基于可信伪标签微调的测试时适应算法[J]. 图学学报, 2025, 46(6): 1292-1303.
[6]	樊乐翔, 马冀, 周登文. 基于退化分离的轻量级盲超分辨率重建网络[J]. 图学学报, 2025, 46(6): 1304-1315.
[7]	王海涵. 基于YOLOv8-OSRA的钢拱塔表观病害多目标检测方法[J]. 图学学报, 2025, 46(6): 1327-1336.
[8]	朱泓淼, 钟国杰, 张严辞. 基于均值漂移与深度学习融合的小语义点云语义分割[J]. 图学学报, 2025, 46(5): 998-1009.
[9]	汪子宇, 曹维维, 曹玉柱, 刘猛, 陈俊, 刘兆邦, 郑健. 基于类内区域动态解耦的半监督肺气管分割[J]. 图学学报, 2025, 46(4): 763-774.
[10]	王道累, 丁子健, 杨君, 郑劭恺, 朱瑞, 赵文彬. 基于体素网格特征的NeRF大场景重建方法[J]. 图学学报, 2025, 46(3): 502-509.
[11]	孙浩, 谢滔, 何龙, 郭文忠, 虞永方, 吴其军, 王建伟, 东辉. 多模态文本视觉大模型机器人地形感知算法研究[J]. 图学学报, 2025, 46(3): 558-567.
[12]	翟永杰, 王璐瑶, 赵晓瑜, 胡哲东, 王乾铭, 王亚茹. 基于级联查询-位置关系的输电线路多金具检测方法[J]. 图学学报, 2025, 46(2): 288-299.
[13]	潘树焱, 刘立群. MSFAFuse：基于多尺度特征信息与注意力机制的SAR和可见光图像融合模型[J]. 图学学报, 2025, 46(2): 300-311.
[14]	张天圣, 朱闽峰, 任怡雯, 王琛涵, 张立冬, 张玮, 陈为. BPA-SAM：面向工笔画数据的SAM边界框提示增强方法[J]. 图学学报, 2025, 46(2): 322-331.
[15]	孙禾衣, 李艺潇, 田希, 张松海. 结合程序内容生成与扩散模型的图像到三维瓷瓶生成技术[J]. 图学学报, 2025, 46(2): 332-344.