基于多尺度特征融合的滚动轴承寿命预测

火久元; 李昕; 常琛; 李宇峰; 张耀南

doi:10.13700/j.bh.1001-5965.2024.0161

基于多尺度特征融合的滚动轴承寿命预测

doi: 10.13700/j.bh.1001-5965.2024.0161

火久元^{1, 2, ,},
李昕¹,
常琛¹,
李宇峰¹,
张耀南²

1.
兰州交通大学电子与信息工程学院，兰州 730070
2.
国家冰川冻土沙漠科学数据中心，兰州 730000

基金项目:

国家自然科学基金(62262038)；甘肃省重点研发计划工业项目(22YF7GA145)

详细信息

通讯作者:
E-mail：huojy@mail.lzjtu.cn

中图分类号: TH133.3；TP183
计量
- 文章访问数: 612
- HTML全文浏览量: 225
- PDF下载量: 79
- 被引次数: 0
出版历程
- 收稿日期: 2024-03-22
- 录用日期: 2024-05-11
- 网络出版日期: 2024-06-20
- 整期出版日期: 2026-05-26

Rolling bearing life prediction based on multi-scale feature fusion

1.
School of Electronic and Information Engineering，Lanzhou Jiaotong University，Lanzhou 730070，China
2.
National Cryosphere Desert Science Data Center，Lanzhou 730000，China

Funds:

National Natural Science Foundation of China (62262038); Gansu Provincial Key R & D Program-Industrial Projects (22YF7GA145)

More Information

Corresponding author: E-mail：huojy@mail.lzjtu.cn

摘要

摘要:
滚动轴承是机械设备中的常用部件，有效预测滚动轴承的剩余使用寿命(RUL)对于制定合理的维修计划和确保设备的安全性具有重要作用。传统的深度学习方法难以提取滚动轴承的多尺度退化特征，而非平稳信号噪声的存在也使RUL更难预测。因此，提出了一种RUL预测模型EEMD-AFP-FSBLformer，该模型结合了集成经验模态分解(EEMD)、离散小波变换(DWT)、注意力特征金字塔(AFP)和FSBLformer网络。通过EEMD分解与DWT降噪处理，对低频模态函数与降噪处理后的高频模态函数进行滚动轴承时域退化特征处理，以产生更多具有代表性的退化特征；将退化特征输入到AFP网络中以提取多尺度的特征；将退化特征作为FSBLformer模型的输入，FSBLformer模型的编码器引入了特征注意力机制和自注意力机制，解码器使用了双向长短期记忆(BiLSTM)网络，使模型在特征提取与时序预测方面更具优势。分别在PHM2012数据集与XJTU-SY数据集的不同工况下进行实验，结果表明：所提模型决定系数达94%以上，可有效提取滚动轴承的多尺度退化特征并预测其RUL。
- 滚动轴承 /
- 寿命预测 /
- 集成经验模态分解 /
- 注意力特征金字塔 /
- FSBLformer
Abstract:
Rolling bearings are commonly used components in mechanical equipment, and the effective prediction of the remaining useful life (RUL) of bearings plays an important role in formulating a reasonable maintenance plan, avoiding sudden downtime of mechanical equipment, and ensuring the safety of equipment. Traditional deep learning methods are difficult to extract multi-dimensional and multi-scale degradation features, which reduces the accuracy of RUL prediction. Meanwhile, there are uncertainties such as noise and model parameters, which make it difficult to meet the maintenance requirements for point prediction of RUL. In this paper, we propose an RUL prediction model called EEMD-AFP-FSBLformer, which integrates the FSBLformer network, attention feature pyramid (AFP), discrete wavelet transform (DWT), and ensemble empirical mode decomposition (EEMD). The low-frequency modal functions are firstly processed by EEMD decomposition with DWT noise reduction and bearing time-domain degradation features with the noise reduction processed high-frequency modal functions in order to produce more representative degradation features; then the degradation features are inputted into the AFP network in order to extract the multi-scale features; and finally, these degradation features are used as inputs to the FSBLformer model. The FSBLformer model’s encoder incorporates the self-attention and feature attention mechanisms, while the decoder employs the bidirectional long short-term memory (BiLSTM) network, which improves the model’s performance in time prediction and feature extraction. The experiments are conducted in different working conditions of the PHM2012 dataset and XJTU-SY dataset, and the comparative experimental analysis shows that the model has a high coefficient of determination of more than 94%, which can effectively extract the multi-scale degradation features of bearings and predict their RUL.
- rolling bearing /
- life prediction /
- ensemble empirical mode decomposition /
- attention feature pyramid /
- FSBLformer

HTML全文

图 1 EEMD特征提取流程

Figure 1. EEMD feature extraction flowchart

下载: 全尺寸图片幻灯片

图 2 AFP网络模型结构

Figure 2. Architecture of AFP network model

下载: 全尺寸图片幻灯片

图 3 FSBLformer预测模型总体架构

Figure 3. The overall architecture of the prediction model

下载: 全尺寸图片幻灯片

图 4 FSBLformer预测模型编码器结构

Figure 4. Encoder structure of FSBLformer prediction model

下载: 全尺寸图片幻灯片

图 5 FSBLformer预测模型解码器结构

Figure 5. Decoder structure of FSBLformer prediction model

下载: 全尺寸图片幻灯片

图 6 Bearing1_1的IMF局部放大图

Figure 6. Partial enlarged IMF image of Bearing1_1

下载: 全尺寸图片幻灯片

图 7 Bearing1_1的高频IMF降噪前后对比

Figure 7. Comparison of Bearing 1_1 before and after high-frequency IMF denosing

下载: 全尺寸图片幻灯片

图 8 Bearing1_1的IMF1部分信号降噪前后对比

Figure 8. Comparison of IMF1 part signal of Bearing 1_1 before and after noise reduction

下载: 全尺寸图片幻灯片

图 9 时域退化特征

Figure 9. Time domain degradation characteristics

下载: 全尺寸图片幻灯片

图 10 工况1下不同滚动轴承的RUL预测结果

Figure 10. RUL prediction results of different roll bearings under the first operating condition

下载: 全尺寸图片幻灯片

图 11 Bearing 1_3重复运行3次的RUL预测结果

Figure 11. RUL prediction results for three repeated runs of Bearing1_3

下载: 全尺寸图片幻灯片

图 12 工况2下不同滚动轴承的RUL预测结果

Figure 12. RUL prediction results of different roll bearings under the second operating condition

下载: 全尺寸图片幻灯片

图 13 Bearing 2_7重复运行3次的RUL预测结果

Figure 13. RUL prediction results for three repeated runs of Bearing2_7

下载: 全尺寸图片幻灯片

图 14 XJTU-SY数据集的RUL预测结果

Figure 14. RUL prediction results for the XJTU-SY dataset

下载: 全尺寸图片幻灯片

图 15 对高频IMF进行降噪前后的RUL预测结果

Figure 15. RUL prediction results before and after denoising for high-frequency IMF

下载: 全尺寸图片幻灯片

图 16 不同方法下Bearing1_3的RUL预测结果

Figure 16. RUL prediction results of Bearing1_3 for different methods

下载: 全尺寸图片幻灯片

图 17 不同方法下Bearing2_7的RUL预测结果

Figure 17. RUL prediction results of Bearing2_7 for different methods

下载: 全尺寸图片幻灯片

表 1 IEEE 2012 PHM预测挑战的数据集

Table 1. Datasets for the IEEE 2012 PHM prediction challenge

运行工况	转速/ (r·min⁻¹)	载荷/N	训练集	测试集
工况1	1800	4000	Bearing1_1, Bearing1_2	Bearing1_3,Bearing1_4, Bearing1_5,Bearing1_6, Bearing1_7
工况2	1650	4200	Bearing2_1, Bearing2_2	Bearing2_3,Bearing2_4, Bearing2_5, Bearing2_6,Bearing2_7

下载: 导出CSV

表 2 Bearing1_1的IMF样本熵

Table 2. Sample entropy of IMF of Bearing1_ 1

模态分量	样本熵	模态分量	样本熵
IMF1	9.40	IMF5	8.93
IMF2	10.06	IMF6	8.02
IMF3	9.80	IMF7	6.51
IMF4	9.67	IMF8	5.10

下载: 导出CSV

表 3 DWT分解层数与相关性系数

Table 3. DWT decomposition layers and correlation coefficients

层数	相关性系数	平均值
1	0.001001	0.996612
2	0.0001047	0.959676
3	0.001326	0.850582
4	0.002175	0.717504
5	0.002692	0.751011
6	0.002298	0.689021
7	0.001241	0.537907

下载: 导出CSV

表 4 Bearing1_3和Bearing2_7降噪前后信号进行RUL预测后评价指标

Table 4. Evaluation metrics after RUL prediction of signals before and after denoising in Bearing1_3 and Bearing2_7

轴承	降噪前/后	MAE	RMSE	R²
Bearing1_3	降噪前	0.0581	0.0766	0.9291
Bearing1_3	降噪后	0.0466	0.0704	0.8883
Bearing2_7	降噪前	0.0505	0.0691	0.9415
Bearing2_7	降噪后	0.0402	0.0661	0.9313

下载: 导出CSV

表 5 RUL预测后评价指标

Table 5. RUL Forecast Evaluation Indicators

方法	MAE		RMSE		R²
方法	Bearing1_3	Bearing2_7	Bearing1_3	Bearing2_7	Bearing1_3	Bearing2_7
Transformer	0.0446	0.0739	0.0646	0.1061	0.9501	0.8967
CNN-Transformer	0.0743	0.0721	0.0951	0.1043	0.8969	0.9038
TCN-Transformer	0.0581	0.0571	0.0766	0.0803	0.9331	0.9251
GRU	0.0480	0.0601	0.0613	0.0921	0.9486	0.9027
LSTM	0.0796	0.0585	0.0877	0.0803	0.8817	0.9101
BiLSTM	0.0641	0.0591	0.0832	0.0816	0.9236	0.9203
EEMD-AFP-FSBLformer	0.0334	0.0525	0.0448	0.0743	0.9758	0.9419

下载: 导出CSV

参考文献(26)

[1]	王焱, 丁华, 孙晓春, 等. 基于改进ECANet-TCN和迁移学习的轴承剩余寿命预测[J]. 振动与冲击, 2023, 42(21): 149-159. WANG Y, DING H, SUN X C, et al. Bearing residual life prediction based on improved ECANet-TCN and transfer learning[J]. Journal of Vibration and Shock, 2023, 42(21): 149-159(in Chinese).
[2]	HENG A, ZHANG S, TAN A C C, et al. Rotating machinery prognostics: state of the art, challenges and opportunities[J]. Mechanical Systems and Signal Processing, 2009, 23(3): 724-739.
[3]	LE SON K, FOULADIRAD M, BARROS A. Remaining useful lifetime estimation and noisy gamma deterioration process[J]. Reliability Engineering & System Safety, 2016, 149: 76-87.
[4]	CARPINTERI A, PAGGI M. The effect of crack size and specimen size on the relation between the Paris and Wöhler curves[J]. Meccanica, 2014, 49(4): 765-773.
[5]	LOUTAS T H, ROULIAS D, GEORGOULAS G. Remaining useful life estimation in rolling bearings utilizing data-driven probabilistic E-support vectors regression[J]. IEEE Transactions on Reliability, 2013, 62(4): 821-832.
[6]	RATHORE M S, HARSHA S P. Prognostics analysis of rolling bearing based on bi-directional LSTM and attention mechanism[J]. Journal of Failure Analysis and Prevention, 2022, 22(2): 704-723.
[7]	LI J X, WANG C, DING H B, et al. EMD and spectrum-centrobaric-correction-based analysis of vortex street characteristics in mist annular flow of wet gas[J]. IEEE Transactions on Instrumentation and Measurement, 2018, 67(5): 1150-1160.
[8]	GAO Z H, LIU Y, WANG Q J, et al. Ensemble empirical mode decomposition energy moment entropy and enhanced long short-term memory for early fault prediction of bearing[J]. Measurement, 2022, 188: 110417.
[9]	LI Q, YAN C F, CHEN G Y, et al. Remaining useful life prediction of rolling bearings based on risk assessment and degradation state coefficient[J]. ISA Transactions, 2022, 129: 413-428.
[10]	GUO R X, WANG Y, ZHANG H C, et al. Remaining useful life prediction for rolling bearings using EMD-RISI-LSTM[J]. IEEE Transactions on Instrumentation and Measurement, 2021, 70: 3509812.
[11]	JIN S Y, SU Y, GUO C J, et al. Offshore ship recognition based on center frequency projection of improved EMD and KNN algorithm[J]. Mechanical Systems and Signal Processing, 2023, 189: 110076.
[12]	LI S S, YANG Y, LI C, et al. Research on signal processing technology of ultrasonic non-destructive testing based on EEMD combined with wavelet packet[J]. IEEJ Transactions on Electrical and Electronic Engineering, 2023, 18(5): 686-700.
[13]	LUO J, LING C, ZHANG L. Fault diagnosis of CRDM based on EEMD-wavelet threshold denoising and WOA-KELM[C]//Proceedings of the 29th International Conference on Nuclear Engineering. New York: ASME, 2022: 1-8.
[14]	ZHAO K, JIA Z, JIA F, et al. Multi-scale integrated deep self-attention network for predicting remaining useful life of aero-engine[J]. Engineering Applications of Artificial Intelligence, 2023, 120: 105860.
[15]	ZHU Y C, YANG S, TONG J G, et al. Multi-scale detector optimized for small target[J]. Optoelectronics Letters, 2024, 20(4): 243-248.
[16]	赵志宏, 李晴, 杨绍普, 等. 基于BiLSTM与注意力机制的剩余使用寿命预测研究[J]. 振动与冲击, 2022, 41(6): 44-50. ZHAO Z H, LI Q, YANG S P, et al. Remaining useful life prediction based on BiLSTM and attention mechanism[J]. Journal of Vibration and Shock, 2022, 41(6): 44-50(in Chinese).
[17]	SHU Z Y, GAO L, YI S, et al. Context-aware 3D points of interest detection via spatial attention mechanism[J]. ACM Transactions on Multimedia Computing, Communications, and Applications, 2023, 19(6): 1-19.
[18]	CHEN M, ZANG S R, AI Z H, et al. RFA-Net: residual feature attention network for fine-grained image inpainting[J]. Engineering Applications of Artificial Intelligence, 2023, 119: 105814.
[19]	ZHAO J W, NIE G Z, WEN Y H. Monthly precipitation prediction in Luoyang city based on EEMD-LSTM-ARIMA model[J]. Water Science and Technology, 2023, 87(1): 318-335.
[20]	SANG Y F, SUN F B, SINGH V P, et al. A discrete wavelet spectrum approach for identifying non-monotonic trends in hydroclimate data[J]. Hydrology and Earth System Sciences, 2018, 22(1): 757-766.
[21]	ZHANG W H, ZENG Y Q, WANG J P, et al. Multi-scale feature pyramid approach for melt track classification in laser powder bed fusion via coaxial high-speed imaging[J]. Computers in Industry, 2023, 151: 103975.
[22]	LIU S T, HUANG D, WANG Y H. Learning spatial fusion for single-shot object detection[EB/OL]. (2019-11-25)[2024-03-01]. https://arxiv.org/abs/1911.09516.
[23]	EL-NOUBY A, TOUVRON H, CARON M, et al. XCiT: cross-covariance image Transformers[EB/OL]. (2021-06-18)[2024-03-01]. https://arxiv.org/abs/2106.09681.
[24]	NECTOUX P, GOURIVEAU R, MEDJAHER K, et al. PRONOSTIA: an experimental platform for bearings accelerated degradation tests[C]//Proceedings of the IEEE International Conference on Prognostics and Health Management. Piscataway: IEEEE Press, 2012: 1-8.
[25]	曾大懿, 蒋雨良, 邹益胜, 等. 一种新的轴承寿命预测特征评价指标构建与验证[J]. 振动与冲击, 2021, 40(22): 18-27. ZENG D Y, JIANG Y L, ZOU Y S, et al. Construction and verification of a new evaluation index for bearing life prediction characteristics[J]. Journal of Vibration and Shock, 2021, 40(22): 18-27(in Chinese).
[26]	KANCHANAMALA P, LAKSHMANAN R, KUMAR B M, et al. AACO: aquila anti-coronavirus optimization-based deep LSTM network for road accident and severity detection[J]. International Journal of Pattern Recognition and Artificial Intelligence, 2023, 37(5): 2252030.