基于注意力 LSTM 的齿轮剩余使用寿命预测

中国民航大学学报 ›› 2024, Vol. 42 ›› Issue (6): 67-73.

基于注意力 LSTM 的齿轮剩余使用寿命预测

郭润夏，倪志高

中国民航大学电子信息与自动化学院，天津 300300

收稿日期:2022-10-30 修回日期:2023-01-05 出版日期:2025-04-08 发布日期:2025-04-08
作者简介:郭润夏（1981— ），男，山东诸城人，教授，博士，研究方向为机载设备故障诊断与健康管理、飞行器控制
基金资助:
国家自然科学基金项目（62173331）；天津市特支计划青年拔尖人才基金项目（TJTZJH-QNBJRC-2-19）；航空科学基金项目（2019ZD067007）；

天津市教委科研计划项目（2018KJ238）

Prediction of gear remaining useful life based on attention LSTM

GUO Runxia, NI Zhigao

College of Electronic Information and Automation, CAUC, Tianjin 300300, China

Received:2022-10-30 Revised:2023-01-05 Online:2025-04-08 Published:2025-04-08

摘要/Abstract

摘要：

针对旋转机械设备中齿轮剩余使用寿命（RUL，remaining useful life）的预测问题，本文利用长短期记忆

（LSTM，long short-term memory）网络在处理时间序列数据方面的独特优势，提出一种将注意力机制和LSTM

融合的齿轮 RUL 预测算法。首先，从齿轮振动信号中分解出最能反映其健康状况的 4 种时域特征（均方根

值、峭度、方差和裕度指标）作为 RUL 预测网络的输入；其次，以提升 RUL 预测结果的精度为目标，把

LSTM 和注意力机制相结合，设计了一种新型的 RUL 预测网络；最后，使用实验室齿轮全寿命加速疲劳实

验台生成的真实数据进行了模型验证，结果表明，本文所提出的注意力 LSTM 算法在进行齿轮 RUL 预测

时具有较高的预测精度。

关键词:

齿轮, 剩余使用寿命（RUL）预测, 长短期记忆（LSTM）网络, 注意力机制, 时域特征

Abstract:

Aiming at the problem of predicting the gear remaining useful life (RUL) in rotating machinery, using the unique

advantages of long short- term memory (LSTM) network in processing time series data, a gear remaining useful life

prediction algorithm combining attention mechanism and LSTM is proposed in this paper. Firstly, four kinds of

time domain features (root mean square value, kurtosis, variance, and margin index) that can better reflect

the health status are decomposed from gear vibration signal and taken as inputs to the RUL prediction network.

Secondly, with the goal of improving the accuracy of RUL prediction results, a novel RUL prediction network is

designed by combining LSTM and attention mechanism. Finally, the model was validated using real data generated from gear full-life accelerated fatigue test bench of the laboratory. The results show that the attention LSTM

algorithm proposed in this paper has high prediction accuracy in predicting the gear RUL.

Key words:

gear, prediction of remaining useful life (RUL), long short-term memory (LSTM) network, attention mechanism;

time domain feature

中图分类号:

V249
TH133

郭润夏, 倪志高.

基于注意力 LSTM 的齿轮剩余使用寿命预测

[J]. 中国民航大学学报, 2024, 42(6): 67-73.

GUO Runxia, NI Zhigao.

Prediction of gear remaining useful life based on attention LSTM

[J]. Journal of Civil Aviation University of China, 2024, 42(6): 67-73.

参考文献 12

	[1]
	PENG S D, DING H, ZHANG G, et al． New determination to loaded
	transmission error of the spiral bevel gear considering multiple elastic
	deformation evaluations under different bearing supports[J]． Mechanism
	and Machine Theory, 2019, 137: 37-52．
	[2]
	PATIL V, CHOUHAN V, PANDYA Y． Geometrical complexity and
	crack trajectory based fatigue life prediction for a spur gear having tooth
	root crack[J]． Engineering Failure Analysis, 2019, 105: 444-465.
	[3]
	FERNANDEZ-DEL-RINCON A, GARCIA P, DIEZ-IBARBIA A, et
	al． Enhanced model of gear transmission dynamics for condition moni－
	toring applications: effects of torque, friction and bearing clearance[J]．
	Mechanical Systems and Signal Processing, 2017, 85: 445-467.
	[4]
	SOUALHI M, NGUYEN K T P, SOUALHI A, et al． Health monitoring of
	bearing and gear faults by using a new health indicator extracted from
	current signals[J]． Measurement, 2019, 141: 37-51.
[5]	马海龙. 基于主元特征融合和 SVM 的轴承剩余寿命预测[J]. 工矿自
	动化, 2019, 45(8): 74-78.
[6]	吕明珠, 苏晓明, 刘世勋, 等. 基于 AFCM-SVM 的滚动轴承退化状
	态评估与剩余寿命预测[J]. 组合机床与自动化加工技术, 2020(3):
	65-69.
[7]	邹旺, 江伟, 冯俊杰, 等．基于 ANN 和 SVM 的轴承剩余使用寿
	命预测[J]．组合机床与自动化加工技术, 2021(1): 32-35.
	[8]
	PAWAR P, BUTPAT A A． Analysis of composite material spur gear un－
	der static loading condition [J]． Materials Today: Proceedings, 2015, 2
	(4/5): 2968-2974.
[9]	程谭．基于深度学习的设备剩余寿命预测技术研究[D]．太原: 中北
	大学, 2021.
[10]	WANG Y D, ZHAO Y F, ADDEPALLI S． Remaining useful life predic－
	tion using deep learning approaches: a review[J]． Procedia Manufactur－
	ing, 2020, 49: 81-88.
[11]	付强, 王华伟. 基于多层 LSTM 的复杂系统剩余寿命智能预测[J].
	兵器装备工程学报, 2022, 43(1): 161-169.
[12]	姚德臣, 李博阳, 刘恒畅, 等. 基于注意力 GRU 算法的滚动轴承剩
	余寿命预测[J]. 振动与冲击, 2021, 40(17): 116-123.