融合注意力机制的残差型双向LSTM汽车电机轴承诊断

doi:10.3969/j.issn.1674-8484.2024.04.007

摘要/Abstract

摘要：

为保障汽车的安全行驶，准确诊断和监测电机轴承故障，该文提出一种融合注意力机制的残差型双向长短期记忆网络（LSTM）汽车电机轴承故障诊断方法。利用特征提取模块结合正反向移动的LSTM组以充分感知汽车电机轴承故障特征；信号诊断模块采用残差型双向LSTM架构，并结合局部增强注意力机制优化权值，获得隐藏状态量；通过故障分类模块采用全局平均池化（GAP）方法与SoftMax模型，有效完成故障检测。结果表明：该方法汽车电机轴承故障检测准确率可达93.1%；在训练样本仅为30的条件下，准确率可达66.3%；当测试集的信噪比从10 dB降低至2 dB时，准确率仅下降8.5%。因此，该方法具有更高的准确性和更强的鲁棒性。

关键词: 汽车安全, 电机轴承, 故障诊断, 注意力机制, 特征提取, 准确率, 信噪比

Abstract:

In order to ensure the safe driving of vehicles and accurately diagnose and monitor motor bearing faults, this paper proposed an automotive motor bearing fault diagnosis method based on residual bidirectional Long Short-Term Memory (LSTM) network with attention mechanism. The feature extraction module combined LSTM groups that move in both forward and backward directions to fully perceive the fault features of automotive motor bearings. The signal diagnosis module adopted a residual bidirectional LSTM architecture and combined the local enhanced attention mechanism to optimize the weights and obtain the hidden state quantity. Global average pooling (GAP) method and SoftMax model are used in fault classification module to effectively detect faults. The results show that the detection accuracy of this method for automotive motor bearing fault detection reaches 93.1%. Under the condition of only 30 training samples, the accuracy reaches 66.3%. When the signal-to-noise ratio of the test set decreases from 10 dB to 2 dB, the accuracy only drops by 8.5%. Therefore, the proposed method has higher accuracy and stronger robustness.

Key words: vehicle safety, motor bearings, fault diagnosis, attention mechanism, feature extraction, accuracy rate, signal-to-noise ratio

中图分类号:

U469

姜健, 王平. 融合注意力机制的残差型双向LSTM汽车电机轴承诊断[J]. 汽车安全与节能学报, 2024, 15(4): 511-519.

JIANG Jian, WANG Ping. Diagnosis of residual bidirectional LSTM automotive motor bearings with attention mechanism[J]. Journal of Automotive Safety and Energy, 2024, 15(4): 511-519.

图/表 16

图1 残差LSTM网络架构

图2 LSTM网络的单元架构

图3 局部增强注意力机制下的LSTM网络序列值模型

图4 轴承构造

图5 PNG-LSTM下的特征提取模块

图6 融合注意力机制的残差型双向LSTM模型

图7 残差型双向LSTM模型下的局部增强注意力机制模型

图8 基于GAP的汽车电机轴承故障分类模块

图9 电机轴承故障诊断测试平台

故障位置	DPD / μm		样本数目	标记	标签
正常	0		1 000	0	J
滚动球	178		1 000	1	GQ07
	356		1 000	3	GQ14
	533		1 000	4	GQ21
内圈故障	178		1 000	5	IT07
	356		1 000	6	IT14
	533		1 000	7	IT21
外圈故障	178	3点钟	330	8	OT07_03
		6点钟	330	9	OT07_06
		12点钟	340	10	OT07_12
	356	3点钟	330	11	OT14_03
		6点钟	330	12	OT14_06
		12点钟	340	13	OT14_12
	533	3点钟	330	14	OT21_03
		6点钟	330	15	OT21_06
		12点钟	340	16	OT21_12

表1 电机轴承振动数据

图10 有代表性的电机轴承状态的部分振动信号图

图11 3种方法随训练标签数目增加的准确率对比

图12 故障诊断多类别混淆矩阵

图13 3种方法的平均损失值比较

图14 不同信噪比下三种方法故障准确率的对比

方法	第1次	第2次	第3次	平均值
ResFDSVD方法	3.75	3.84	3.72	3.77
FDWK方法	1.65	1.75	1.79	1.73
本文方法	1.81	1.85	1.83	1.83

表2 3种方法的故障诊断时间对比 (单位：s)

参考文献 19

[1]	邵锦才, 何致远, 王子辉. 轮毂直驱电动汽车速度传感器的故障容错控制[J]. 汽车安全与节能学报, 2022, 13(1): 78-85.
	SHAO Jingcai, HE Zhiyuan, WANG Zihui, et al. Fault-tolerant control of speed sensors for direct drive in-wheel electric vehicle[J]. J Autom Safe Energ, 2022, 13(1): 78-85. (in Chinese)
[2]	张伟涛, 张东江, 纪晓凡, 等. 基于CCA和多通道循环维纳滤波的滚动轴承故障源分析[J]. 振动与冲击, 2023, 42(24): 317-325.
	ZHANG Weitao, ZHANG Dongjiang, JI Xiaodan, et al. Fault analysis of rolling element bearing based on CCA and a multichannel cyclic wiener filter[J]. J Vibr Shock, 2023, 42(24): 317-325. (in Chinese)
[3]	刘浩天, 魏洪乾, 时培成, 等. 基于帧间隔-总线电压混合特征的汽车ECU伪装攻击识别[J]. 汽车工程, 2023, 45(11): 2070-2081.
	LIU Haotian, WEI Hongqian, SHI Peicheng, et al. The masquerade intrusion detection technique for automotive ECUs based on the hybrid feature extraction of frame intervals and bus voltages[J]. Autom Engineering, 2023, 45(11): 2070-2081. (in Chinese)
[4]	张希, 廖宇兰, 李沁逸, 等. 安全行驶下的车用滚动轴承的数字孪生故障诊断[J]. 汽车安全与节能学报, 2023, 14(2): 232-238.
	ZHANG Xi, LIAO Yulan, LI Qinyi, et al. Digital twin fault diagnosis of automotive motor bearings under safe driving[J]. J Autom Safe Energ, 2023, 14(2): 232-238. (in Chinese)
[5]	张鹏博, 陈仁祥, 邵毅明, 等. 纯电动汽车电驱动系统故障诊断研究进展[J]. 汽车工程, 2024, 46(1): 61-74.
	ZHANG Pengbo, CHEN Renxiang, SHAO Yiming, et al. Research review of fault diagnosis for electric drive powertrain system of pure electric vehicles[J]. Autom Engineering, 2024, 46(1): 61-74. (in Chinese)
[6]	CUI Lingli, SUN Mengxin, ZHA Chunqing, et al. Early bearing fault diagnosis based on the improved singular value decomposition method[J]. Int’l J Advan Manufact Tech, 2023, 124(11-12): 3899-3910.
[7]	SU Naiquan, ZHANG Qinghua, ZHOU Lingmeng, et al. A fault diagnosis of rotating machinery based on a mutual dimensionless index and a convolution neural network[J]. IEEE Intel Syst, 2023, 38(4): 33-41.
[8]	ZHU Rui, WANG Mingxin, XU Siyu, et al. Fault diagnosis of rolling bearing based on singular spectrum analysis and wide convolution kernel neural network[J]. J Low Freq Noise V A, 2022, 41(4): 1307-1321.
[9]	Pavithra R, Ramachandran P. Deep convolution neural network for machine health monitoring using spectrograms of vibration signal and its EMD-intrinsic mode functions[J]. J Intel Fuzzy Syst, 2023, 44(6): 8827-8840.
[10]	SONG Xinmin, WEI Weihua, ZHOU Junbo, et al. Bayesian-optimized hybrid Kernel SVM for rolling bearing fault diagnosis[J]. Sensors, 2023, 23(11): No 5137.
[11]	REN Changan, LI He, LEI Jichong, et al. A CNN-LSTM-based model to fault diagnosis for CPR1000[J]. Nucl Tech, 2023, 209(9): 1365-1372.
[12]	ZHANG Shuo, LIU Zhiwen, CHEN Yunping, et al. Selective kernel convolution deep residual network based on channel-spatial attention mechanism and feature fusion for mechanical fault diagnosis[J]. ISA Trans, 2023,133: 369-383.
[13]	朱雪峰, 冯早, 马军, 等. 基于MSCNN-LSTM的注意力机制U型管道缺陷识别模型[J]. 振动与冲击, 2023, 42(22): 293-302.
	ZHU Xuefeng, FENG Zao, MA Jun, et al. Siphon defect recognition model based on the MSCNN-LSTM and attention mechanism[J]. J Vibr Shock, 2023, 42(22): 293-302. (in Chinese)
[14]	YAN Xuyang, Sarkar M, Lartey B, et al. An online learning framework for sensor fault diagnosis analysis in autonomous cars[J]. IEEE Trans Intel Transport, 2023, 24(12): 14467-14479.
[15]	张英杰, 张彩华, 陆碧良, 等. 基于类别域自适应的滚动轴承故障诊断[J]. 振动与冲击, 2023, 42(24): 117-126.
	ZHANG Yingjie, ZHANG Caihua, LU Biliang, et al. Bearing fault diagnosis model based on class domain adaptation[J]. J Vibr Shock, 2023, 42(24): 117-126. (in Chinese)
[16]	SUN Meidi, WANG Hui, LIU Ping, et al. A novel data-driven mechanical fault diagnosis method for induction motors using stator current signals[J]. IEEE Trans Transport Elect, 2023, 9(1): 347-358.
[17]	XU Tao, LV Huan, LIN Shoujin, et al. A fault diagnosis method based on improved parallel convolutional neural network for rolling bearing[J]. Proceed Instit Mech Engi Part G-J Aero Engi, 2023, 237(12): 2759-2771.
[18]	JIA Sixiang, LI Yongbo, MAO Gang, et al. Multi-representation symbolic convolutional neural network: A novel multisource cross-domain fault diagnosis method for rotating system[J]. Struct Heal Monit, 2023, 22(6): 3940-3955.
[19]	刘延伟, 黄志明, 高博麟, 等. 车载视角下基于视觉信息的前车行为识别[J]. 汽车安全与节能学报, 2023, 14(6): 707-714.
	LIU Yanwei, Huang Zhiming, GAO Bolin, et al. Recognition of front vehicle behavior based on visual information from vehicle perspective[J]. J Autom Safe Energ, 2023, 14(6): 707-714. (in Chinese)