双向建模增强TKAN和全局注意力机制融合的滚动轴承剩余寿命预测

姜蕾; 郑建飞; 胡昌华; 赵瑞星; 韩其辉; 杨立浩

doi:10.16383/j.aas.c250064

双向建模增强TKAN和全局注意力机制融合的滚动轴承剩余寿命预测

doi: 10.16383/j.aas.c250064 cstr: 32138.14.j.aas.c250064

1.
火箭军工程大学导弹工程学院西安 710025
2.
火箭军装备部驻成都地区某军事代表室成都 610000

基金项目: 国家重大科研仪器研制项目(62227814), 国家自然科学基金(62233017, 62373369)资助

详细信息

作者简介:
姜蕾：火箭军工程大学导弹工程学院硕士研究生. 主要研究方向为预测与健康管理, 剩余寿命估计. E-mail: jl00092025@163.com

郑建飞：火箭军工程大学教授. 主要研究方向为预测与健康管理, 剩余寿命估计. 本文通信作者. E-mail: zjf302@126.com

胡昌华：火箭军工程大学教授. 主要研究方向为故障诊断, 可靠性工程和预测与健康管理. E-mail: hch66603@163.com

赵瑞星：火箭军装备部驻成都地区某军事代表室工程师. 主要研究方向为质量管理. E-mail: zhruixing@163.com

韩其辉：火箭军工程大学讲师. 主要研究方向为图像处理, 计算摄影, 故障诊断与预测和模式识别. E-mail: hanqihui2013@163.com

杨立浩：火箭军工程大学讲师. 主要研究方向为可靠性工程, 故障诊断与预测. E-mail: yanglihao003@163.com

计量
- 文章访问数: 16
- HTML全文浏览量: 10
- 被引次数: 0
出版历程
- 收稿日期: 2025-02-24
- 录用日期: 2025-06-09
- 网络出版日期: 2025-07-14

Bidirectional Modeling-enhanced TKAN and Global Attention Mechanism Fusion for Rolling Bearing Remaining Useful Life Prediction

1.
College of Missile Engineering, Rocket Force University of Engineering, Xi＇an 710025
2.
Military Representative Office of the Rocket Force Equipment Department Stationed in Chengdu, Chengdu 610000

Funds: Supported by National Major Scientific Research Instrument Development Project (62227814) and National Natural Science Foundation of China (62233017, 62373369)

More Information

Author Bio:
JIANG Lei　Master student at the College of Missile Engineering, Rocket Force University of Engineering. Her research interest covers prognostics and health management, and remaining useful life estimation

ZHENG Jian-Fei　Professor at Rocket Force University of Engineering. His research interest covers prognostics and health management, and remaining useful life estimation. Corresponding author of this paper

HU Chang-Hua　Professor at Rocket Force University of Engineering. His research interest covers fault diagnosis, reliability engineering, and prognostics and health management

ZHAO Rui-Xing　Engineer at Military Representative Office of the Rocket Force Equipment Department Stationed in Chengdu. His main research interest is quality management

HAN Qi-Hui　Lecturer at Rocket Force University of Engineering. His research interest covers image processing, computational photography, fault diagnosis and prediction, and pattern recognition

YANG Li-Hao　Lecturer at Rocket Force University of Engineering. His research interest covers reliability engineering and fault diagnosis and prediction

摘要

摘要: 滚动轴承剩余使用寿命(RUL)的精准预测是确保设备或系统安全可靠运行的关键. 针对滚动轴承RUL预测中多维退化特征的长期依赖关系难以有效建模的问题, 提出一种双向时间序列建模与注意力机制融合的预测模型——双向时序科尔莫戈洛夫−阿诺尔德注意力网络(Bi-TKAN-Att). 该模型兼具了时序科尔莫戈洛夫−阿诺尔德网络的强时序建模能力和全局注意力机制的关键特征提取能力, 采用双向建模的方式捕捉前后向信息, 最终实现了具有长期依赖多维退化特征的滚动轴承RUL预测. 所提方法在滚动轴承数据集上进行实验验证, 结果表明Bi-TKAN-Att模型在捕获滚动轴承退化特性和提升RUL预测精度方面具有显著优势, 并通过消融实验证明了模型各组件的合理性和有效性, 为滚动轴承的寿命预测提供了全新可行的解决方案.
- 剩余使用寿命预测 /
- 滚动轴承 /
- 时序科尔莫戈洛夫−阿诺尔德网络 /
- 双向建模 /
- 全局注意力机制
Abstract: The accurate prediction of the remaining useful life (RUL) of rolling bearings is crucial for ensuring the safe and reliable operation of equipment or system. To address the challenge of effectively modeling the long-term dependencies of multi-dimensional degradation features in RUL prediction, this paper proposes a prediction model that integrates bidirectional time-series modeling and attention mechanisms——Bidirectional temporal Kolmogorov-Arnold attention networks (Bi-TKAN-Att). This model combines the powerful time-series modeling capability of temporal Kolmogorov-Arnold networks with the key feature extraction capability of global attention mechanisms. It uses a bidirectional modeling approach to capture both forward and backward information, ultimately achieving RUL prediction for rolling bearings with long-term dependent multi-dimensional degradation features. The proposed method is experimentally validated on the rolling bearing dataset. Experimental results demonstrate that the Bi-TKAN-Att model has significant advantages in capturing rolling bearing degradation features and improving RUL prediction accuracy. Ablation experiments further confirm the rationality and effectiveness of each component of the model, providing a novel and feasible solution for the lifetime prediction of rolling bearing.
- Remaining useful life prediction /
- rolling bearing /
- temporal Kolmogorov-Arnold networks /
- bidirectional modeling /
- global attention mechanism

HTML全文

图 1 TKAN结构图

Fig. 1 TKAN structure diagram

下载: 全尺寸图片幻灯片

图 2 Bi-TKAN结构图

Fig. 2 Bi-TKAN structure diagram

下载: 全尺寸图片幻灯片

图 3 注意力机制结构图

Fig. 3 Structure diagram of attention mechanism

下载: 全尺寸图片幻灯片

图 4 Bi-TKAN-Att结构图

Fig. 4 Bi-TKAN-Att structure diagram

下载: 全尺寸图片幻灯片

图 5 基于Bi-TKAN-Att的时间序列预测框架

Fig. 5 Time-series prediction framework based on Bi-TKAN-Att

下载: 全尺寸图片幻灯片

图 6 特征重要性分布图

Fig. 6 Feature importance distribution map

下载: 全尺寸图片幻灯片

图 7 消融实验预测结果对比

Fig. 7 Comparison of ablation experiment prediction results

下载: 全尺寸图片幻灯片

图 8 Bi-TKAN-Att与其他方法的预测结果对比

Fig. 8 Comparison of the prediction results of Bi-TKAN-Att with those of other methods

下载: 全尺寸图片幻灯片

图 9 不同网络结构的RMSE和MAE比较

Fig. 9 Comparison of RMSE and MAE for different network architectures

下载: 全尺寸图片幻灯片

表 1 Bi-TKAN-Att的结构参数

Table 1 Structural parameters of Bi-TKAN-Att

网络结构超参数	最优值
Bi-TKAN层	2
Bi-TKAN层1	32
Bi-TKAN层2	40
全连接层	(40, 1)
批次大小	128
训练周期	100
学习率	0.0008
滑动窗口大小	300

下载: 导出CSV

表 2 消融实验预测结果评价指标对比

Table 2 Comparison of evaluation metrics for ablation experiment prediction results

评价指标	Bi-TKAN-Att	Bi-TKAN	TKAN-Att	TKAN
R²	0.9373	0.8601	0.8464	0.7796
RMSE (%)	6.3199	9.4381	9.8893	11.8460
MAE (%)	5.0744	8.1298	8.1888	10.5069

下载: 导出CSV

表 3 Bi-TKAN-Att与其他方法预测结果评价指标对比

Table 3 Comparison of evaluation metrics of prediction results between Bi-TKAN-Att and other methods

轴承编号	评价指标	Bi-TKAN-Att	Bi-LSTM	TCN-BiLSTM	Transformer
	R²	0.9373	0.7390	0.8898	0.8667
bearing1-3	RMSE (%)	6.3199	12.8902	8.3768	9.2119
	MAE (%)	5.0744	10.0602	6.0953	6.9968
	R²	0.9270	0.8223	0.8708	0.7084
bearing2-3	RMSE (%)	6.6083	10.3079	8.7897	13.2034
	MAE (%)	4.9011	5.7579	6.8132	10.9906

下载: 导出CSV

表 4 Bi-TKAN-Att与其他预测方法的结果对比

Table 4 Comparison of the results of Bi-TKAN-Att with other prediction methods

	Bi-TKAN-Att	TCN-SABG	DT-DA	Shrinkage-Att-TL	Adaptive-GPR	WTE-Trans
RMSE (%)	6.32	7.7	28.4	15	77.9	17.3
MAE (%)	5.07	6.4	33.5	13	38.3	14.6
文献	本文方法	[23]	[24]	[25]	[26]	[27]

下载: 导出CSV

表 5 工作条件1下预测结果评价指标

Table 5 Evaluation metrics of prediction results under working condition 1

轴承编号	R²	RMSE (%)	MAE (%)
bearing1-1	0.9231	7.7832	4.6578
bearing1-2	0.8976	9.8970	8.9352
bearing1-3	0.9378	6.3178	5.0723
bearing1-4	0.9047	8.4254	7.6311
bearing1-5	0.9207	7.7601	4.5507

下载: 导出CSV

参考文献(27)

[1]	董青, 郑建飞, 胡昌华, 李冰, 牟含笑. 基于两阶段自适应Wiener过程的剩余寿命预测方法. 自动化学报, 2022, 48(2): 539−553 Dong Qing, Zheng Jian-Fei, Hu Chang-Hua, Li Bing, Mu Han-Xiao. Remaining useful life prognostic method based on two-stage adaptive Wiener process. Acta Automatica Sinica, 2022, 48(2): 539−553
[2]	Ye Z S, Revie M, Walls L. A load sharing system reliability model with managed component degradation. IEEE Transactions on Reliability, 2014, 63(3): 721−730 doi: 10.1109/TR.2014.2315965
[3]	Kong Y N, Ye Z S. Goodness-of-fit tests in the multi-state Markov model. Reliability Engineering & System Safety, 2017, 166: 16−24
[4]	Cubillo A, Perinpanayagam S, Esperon-Miguez M. A review of physics-based models in prognostics: Application to gears and bearings of rotating machinery. Advances in Mechanical Engineering, 2016, 8(8): 1−21
[5]	Wang T, Liu Z, Mrad N. A probabilistic framework for remaining useful life prediction of bearings. IEEE Transactions on Instrumentation and Measurement, 2021, 70: Article No. 3503412
[6]	Liu L, Song X, Chen K, Hou B C, Chai X D, Ning H S. An enhanced encoder-decoder framework for bearing remaining useful life prediction. Measurement, 2021, 170: Article No. 108753 doi: 10.1016/j.measurement.2020.108753
[7]	Zhao R, Yan R Q, Chen Z H, Mao K Z, Wang P, Gao R X. Deep learning and its applications to machine health monitoring: A survey. Mechanical Systems and Signal Processing, 2019, 115: 213−237 doi: 10.1016/j.ymssp.2018.05.050
[8]	莫仁鹏, 李天梅, 司小胜, 朱旭. 采用残差网络与卷积注意力机制的设备剩余使用寿命预测方法. 西安交通大学学报, 2022, 56(4): 194−202 doi: 10.7652/xjtuxb202204021 Mo Ren-Peng, Li Tian-Mei, Si Xiao-Sheng, Zhu Xu. Remaining useful life prediction for equipment using residual network and convolutional attention mechanism. Journal of Xi'an Jiaotong University, 2022, 56(4): 194−202 doi: 10.7652/xjtuxb202204021
[9]	Guo L, Lei Y G, Li N P, Yan T, Li N B. Machinery health indicator construction based on convolutional neural networks considering trend burr. Neurocomputing, 2018, 292: 142−150 doi: 10.1016/j.neucom.2018.02.083
[10]	Guo L, Li N P, Jia F, Lei Y G, Lin J. A recurrent neural network based health indicator for remaining useful life prediction of bearings. Neurocomputing, 2017, 240: 98−109 doi: 10.1016/j.neucom.2017.02.045
[11]	郑建飞, 牟含笑, 胡昌华, 赵瑞星, 张博玮. 考虑多性能指标相关性的退化设备剩余寿命预测. 哈尔滨工程大学学报, 2022, 43(5): 620−629 doi: 10.11990/jheu.202109032 Zheng Jian-Fei, Mu Han-Xiao, Hu Chang-Hua, Zhao Rui-Xing, Zhang Bo-Wei. Remaining useful life prediction of degradation equipment considering multiple performance indexes correlation. Journal of Harbin Engineering University, 2022, 43(5): 620−629 doi: 10.11990/jheu.202109032
[12]	Lin R G, Wang H W, Xiong M L, Hou Z G, Che C C. Attention-based gate recurrent unit for remaining useful life prediction in prognostics. Applied Soft Computing, 2023, 143: Article No. 110419 doi: 10.1016/j.asoc.2023.110419
[13]	Xu Z Q, Zhang Y J, Miao Q. An attention-based multi-scale temporal convolutional network for remaining useful life prediction. Reliability Engineering & System Safety, 2024, 250: Article No. 110288
[14]	Zhou J H, Qin Y, Chen D L, Liu F Q, Qian Q. Remaining useful life prediction of bearings by a new reinforced memory GRU network. Advanced Engineering Informatics, 2022, 53: Article No. 101682 doi: 10.1016/j.aei.2022.101682
[15]	Zhao B, Zhang W G, Zhang Y R, Zhang C P, Zhang C, Zhang J W. Lithium-ion battery remaining useful life prediction based on interpretable deep learning and network parameter optimization. Applied Energy, 2025, 379: Article No. 124713 doi: 10.1016/j.apenergy.2024.124713
[16]	康守强, 周月, 王玉静, 谢金宝. 基于改进SAE和双向LSTM的滚动轴承RUL预测方法. 自动化学报, 2022, 48(9): 2327−2336 Kang Shou-Qiang, Zhou Yue, Wang Yu-Jing, Xie Jin-Bao. RUL prediction method of a rolling bearing based on improved SAE and Bi-LSTM. Acta Automatica Sinica, 2022, 48(9): 2327−2336
[17]	夏然, 苏春. 基于健康因子和混合Bi-LSTM-NAR模型的锂离子电池剩余寿命预测. 中国机械工程, 2024, 35(5): 851−859 doi: 10.3969/j.issn.1004-132X.2024.05.010 Xia Ran, Su Chun. Remaining useful life prediction for lithium-ion batteries based on health indicators and hybrid Bi-LSTM-NAR model. China Mechanica Engineering, 2024, 35(5): 851−859 doi: 10.3969/j.issn.1004-132X.2024.05.010
[18]	Liu Z M, Wang Y X, Vaidya S, Ruehle F, Halverson J, Soljacic M. KAN: Kolmogorov-Arnold networks. arXiv preprint arXiv: 2404.19756, 2024.
[19]	Genet R, Inzirillo H. TKAN: Temporal Kolmogorov-Arnold networks. arXiv preprint arXiv: 2405.07344v2, 2024.
[20]	He J, Xiao Z G, Zhang C F. Predicting the remaining useful life of rails based on improved deep spiking residual neural network. Process Safety and Environmental Protection, 2024, 188: 1106−1117 doi: 10.1016/j.psep.2024.06.008
[21]	Lei Y G, Li N P, Guo L, Li N B, Yan T, Lin J. Machinery health prognostics: A systematic review from data acquisition to RUL prediction. Mechanical Systems and Signal Processing, 2018, 104: 799−834 doi: 10.1016/j.ymssp.2017.11.016
[22]	梁浩鹏, 曹洁, 赵小强. 基于并行双向时间卷积网络和双向长短期记忆网络的轴承剩余使用寿命预测方法. 控制与决策, 2024, 39(4): 1288−1296 Liang Hao-Peng, Cao Jie, Zhao Xiao-Qiang. Remaining useful life prediction method for bearing based on parallel bidirectional temporal convolutional network and bidirectional long and short-term memory network. Control and Decision, 2024, 39(4): 1288−1296
[23]	刘业峰, 王帅, 刘晶晶, 马祎航. 基于TCN-SA和Bi-GRU的轴承剩余寿命预测. 计算机集成制造系统, DOI: 10.13196/j.cims.2024.0260 Liu Ye-Feng, Wang Shuai, Liu Jing-Jing, Ma Yi-Hang. Bearing remaining life prediction based on TCN-SA and Bi-GRU. Computer Integrated Manufacturing Systems, DOI: 10.13196/j.cims.2024.0260
[24]	Li X, Zhang W, Ma H, Luo Z, Li X. Data alignments in machinery remaining useful life prediction using deep adversarial neural networks. Knowledge-Based Systems, 2020, 197: Article No. 105843 doi: 10.1016/j.knosys.2020.105843
[25]	Li W X, Shang Z W, Gao M S, Qian S Q, Feng Z H. Remaining useful life prediction based on transfer multi-stage shrinkage attention temporal convolutional network under variable working conditions. Reliability Engineering and System Safety, 2022, 226: Article No. 108722 doi: 10.1016/j.ress.2022.108722
[26]	Hou W J, Peng Y Z. Adaptive ensemble Gaussian process regression-driven degradation prognosis with applications to bearing degradation. Reliability Engineering and System Safety, 2023, 239: Article No. 109479 doi: 10.1016/j.ress.2023.109479
[27]	Zhang M Y, He C, Huang C X, Yang J H. A weighted time embedding Transformer network for remaining useful life prediction of rolling bearing. Reliability Engineering and System Safety, 2024, 251: Article No. 110399 doi: 10.1016/j.ress.2024.110399