DPCA-MFF:面向复杂工况的高抗噪性轴承故障诊断模型

doi:10.6040/j.issn.1672-3961.0.2025.130

摘要/Abstract

摘要： 为解决复杂工况下单一轴承故障诊断存在的单模态数据信息片面、抗噪能力弱及跨工况适应性差等问题,提出一种基于动态主成分分析(dynamic principal component analysis, DPCA)与多模态特征融合(multimodal feature fusion, MFF)的增强诊断模型(DPCA-MFF)。利用DPCA对数据进行自适应降维,去除噪声并保留关键特征,结合时域统计量与快速傅里叶变换(fast Fourier transform, FFT)提取时频特征;采用双向门控循环单元(bidirectional gated recurrent unit, BiGRU)和残差网络(residual networks, ResNet)分别提取时频图像的时序依赖和空间特征,通过多头自注意力机制增强特征表达能力;通过交叉注意力机制实现多模态特征动态融合。在CWRU数据集噪声干扰试验中,所提模型对10种不同类型的故障信号平均识别率保持在99%以上;在变转速跨工况试验中,所提模型针对不同转速和健康状况的故障信号平均诊断准确率达98.1%,较单模态模型提升约4~6百分点,较卷积神经网络(convolutional neural network, CNN)-门控循环单元(gated recurrent unit, GRU)融合模型提升1.7百分点,在噪声水平为1.0的情况下准确率仍保持65.8%,在不同类型噪声条件下表现优异。试验结果表明,所提模型通过动态降噪、注意力增强与多模态融合全面提取故障特征,显著提升复杂工况下的抗噪性能与跨工况适应能力,为工业设备智能运维提供有效解决方案。

关键词: 故障诊断, 动态主成分分析, 多模态特征融合, 交叉注意力机制, 双向门控循环单元

Abstract: To address the issues of insufficient information in single-modal data, weak noise resistance, and poor cross-condition adaptability in bearing fault diagnosis under complex working conditions, an enhanced diagnostic model(DPCA-MFF)based on dynamic principal component analysis(DPCA)and multimodal feature fusion(MFF)was proposed. The DPCA was applied to adaptively reduce data dimensionality, removing noise while retaining essential features, and combined time-domain statistics with fast Fourier transform(FFT)to extract time-frequency features. Bidirectional gated recurrent unit(BiGRU)and residual network(ResNet)were employed to capture long-term temporal dependencies and spatial features from time-frequency images, respectively, with multi-head self-attention mechanisms applied to enhance feature representation. Cross-attention mechanism was introduced to dynamically fuse multimodal features. Noise interference experiments on the CWRU dataset showed that the proposed model maintained an average recognition rate above 99% under ten different types of fault signals. Under variable-speed cross-condition scenarios, the proposed model achieved an average diagnostic accuracy rate of 98.1% for fault signals of different rotational speeds and health conditions, outperforming single-modal models by about 4 to 6 percentage points and the convolutional neural network(CNN)-gated recurrent unit(GRU)fusion model by 1.7 percentage points. Even at a noise level of 1.0, the accuracy rate remained at 65.8%, which demonstrated strong performance across various noise types. The experimental results indicated that the proposed model effectively extracted comprehensive fault features through dynamic denoising, attention enhancement, and multi-modal fusion, significantly improving the noise immunity and cross-condition adaptability under complex conditions, offering a robust solution for intelligent maintenance of industrial equipment.

Key words: fault diagnosis, dynamic principal component analysis, multimodal feature fusion, cross-attention mechanism, bidirectional gate recurrent unit

中图分类号:

TP277

王琦淼,林培光,孙玫,刘利达. DPCA-MFF:面向复杂工况的高抗噪性轴承故障诊断模型[J]. 山东大学学报 (工学版), 2026, 56(4): 94-105.

Wang Qimiao, Lin Peiguang, Sun Mei, Liu Lida. DPCA-MFF: high noise resistant bearing fault diagnosis model for complex working conditions[J]. Journal of Shandong University(Engineering Science), 2026, 56(4): 94-105.

参考文献

[1] Yan R Q, Shang Z G, Xu H, et al. Wavelet transform for rotary machine fault diagnosis: 10 years revisited[J]. Mechanical Systems and Signal Processing, 2023, 200: 110545.
[2] 苏红, 朱勇, 刘金华, 等. 旋转机械健康状态评估方法研究现状与展望[J]. 排灌机械工程学报, 2024, 42(3): 304-318. Su Hong, Zhu Yong, Liu Jinhua, et al. Research status and prospect of health status assessment methods for rotating machinery[J]. Journal of Drainage and Irrigation Machinery Engineering, 2024, 42(3): 304-318.
[3] 唐佳敏, 邓耀华. 基于多模态数据的转动部件故障诊断方法研究[J]. 机电工程技术, 2024, 53(12): 67-71. Tang Jiamin, Deng Yaohua. Research on fault diagnosis method of rotating components based on multimodal data[J]. Mechanical & Electrical Engineering Technology, 2024, 53(12): 67-71.
[4] Xia M, Li T, Xu L, et al. Fault diagnosis for rotating machinery using multiple sensors and convolutional neural networks[J]. IEEE/ASME Transactions on Mechatronics, 2018, 23(1): 101-110.
[5] 李恒, 张氢, 秦仙蓉, 等. 基于短时傅里叶变换和卷积神经网络的轴承故障诊断方法[J]. 振动与冲击, 2018, 37(19): 124-131. Li Heng, Zhang Qing, Qin Xianrong, et al. Fault diagnosis method for rolling bearings based on short-time Fourier transform and convolution neural network[J]. Journal of Vibration and Shock, 2018, 37(19): 124-131.
[6] Liu H M, Zhang J C, Cheng Y J, et al. Fault diagnosis of gearbox using empirical mode decomposition and multi-fractal detrended cross-correlation analysis[J]. Journal of Sound and Vibration, 2016, 385: 350-371.
[7] Zhang Q, Wei X H, Wang Y, et al. Convolutional neural network with attention mechanism and visual vibration signal analysis for bearing fault diagnosis[J]. Sensors, 2024, 24(6): 1831.
[8] Zhang X H, Wang H, Wang C Z, et al. Time-segment-wise feature fusion transformer for multi-modal fault diagnosis[J]. Engineering Applications of Artificial Intelligence, 2024, 138: 109358.
[9] Xu Z Z, Chen X, Xu J T. Multi-modal multi-sensor feature fusion spiking neural network algorithm for early bearing weak fault diagnosis[J]. Engineering Applications of Artificial Intelligence, 2025, 141: 109845.
[10] Sun Y J, Li S H, Wang X H. Bearing fault diagnosis based on EMD and improved Chebyshev distance in SDP image[J]. Measurement, 2021, 176: 109100.
[11] Long Z, Zhang X F, Song D Y, et al. Motor fault diagnosis using image visual information and bag of words model[J]. IEEE Sensors Journal, 2021, 21(19): 21798-21807.
[12] 叶壮, 余建波. 基于多通道一维卷积神经网络特征学习的齿轮箱故障诊断方法[J]. 振动与冲击, 2020, 39(20): 55-66. Ye Zhuang, Yu Jianbo. Gearbox fault diagnosis based on feature learning of multi-channel one-dimensional convolutional neural network[J]. Vibration and Shock, 2020, 39(20): 55-66.
[13] 曹月, 王衍学, 姚家驰, 等. 基于多模态特征融合的并行双通道轴承故障诊断[J]. 机床与液压, 2025, 53(13): 71-76. Cao Yue, Wang Yanxue, Yao Jiachi, et al. Parallel dual-channel bearing fault diagnosis based on multi-modal feature fusion[J]. Machine Tool & Hydraulic, 2025, 53(13): 71-76.
[14] Zhang H F, Ge H W, Li T, et al. Combining decomposition and graph capsule network for multi-objective vehicle routing optimization[J]. Intelligent Data Analysis, 2025, 29(1): 116-140.
[15] Sun X M, Yang Y H, Chen C Z, et al. A multi-branch convolution and dynamic weighting method for bearing fault diagnosis based on acoustic-vibration information fusion[J]. Actuators, 2025, 14(1): 17.
[16] Yin J J, Wang T, Du Y L, et al. SLIC superpixel segmentation for polarimetric SAR images[J]. IEEE Transactions on Geoscience and Remote Sensing, 2022, 60: 5201317.
[17] Wang Z H, Chen C Q, Li J H, et al. ST-CapsNet: linking spatial and temporal attention with capsule network for P300 detection improvement[J]. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 2023, 31: 991-1000.
[18] Li G N, Xiong C L, Gao J J, et al. Fault detection, diagnosis and calibration of heating, ventilation and air conditioning sensors by combining principal component analysis and improved Bayesian inference[J]. Journal of Building Engineering, 2024, 82: 108230.
[19] 史宗帅, 亚森江·加入拉, 崔鹏飞, 等. 特征降维下基于LSSA-SVM的转子系统故障诊断模型[J]. 机电工程, 2025, 42(3): 463-471. Shi Zongshuai, Yasenjiang Jiarula, Cui Pengfei, et al. Rotor system fault diagnosis model based on LSSA-SVM under feature dimensionality reduction[J]. Journal of Mechanical & Electrical Engineering, 2025, 42(3): 463-471.
[20] 高强, 扈书亮, 董超, 等. 基于主元分析的FPSO故障检测与诊断[J]. 化工自动化及仪表, 2008, 35(4): 7-11. Gao Qiang, Hu Shuliang, Dong Chao, et al. FPSO fault monitor and diagnosis based on principal components analysis[J]. Control and Instruments in Chemical Industry, 2008, 35(4): 7-11.
[21] 刘肖杰. 某无人机动力分系统故障诊断关键技术的研究与实现[D]. 成都: 电子科技大学, 2024: 29-52. Liu Xiaojie. Research and implementation of key technologies for fault diagnosis for aero-power subsystem[D]. Chengdu: University of Electronic Science and Technology of China, 2024: 29-52.
[22] Ding C, Wang F, LI S T, et al. Three-core cable fault line identification based on ground wire current and BiGRU-ResNet-MA[J]. IEEE Access, 2024, 12: 136120-136130.
[23] Moussa M A, Boucherma M, Khezzar A.A detection method for induction motor bar fault using sidelobes leakage phenomenon of the sliding discrete Fourier transform[J]. IEEE Transactions on Power Electronics, 2017, 32(7): 5560-5572.
[24] Lin H C, Ye Y C. Reviews of bearing vibration measurement using fast Fourier transform and enhanced fast Fourier transform algorithms[J]. Advances in Mechanical Engineering, 2019, 11: 1687814018816751.
[25] Fu Y C, Gao Z W, Liu Y H, et al. Actuator and sensor fault classification for wind turbine systems based on fast Fourier transform and uncorrelated multi-linear principal component analysis techniques[J]. Processes, 2020, 8(9): 1066.
[26] Xu R Q, Chang K, Xia Y D, et al. A semi-supervised fault diagnosis method for ventilation fan using multi-head attention-enhanced generative adversarial network[J]. Journal of Vibration and Control, 2026, 32(5/6): 1265-1280.
[27] Jiang L, Li X J, Wu L, et al. Bearing fault diagnosis method based on a multi-head graph attention network[J]. Measurement Science and Technology, 2022, 33(7): 075012.
[28] Yan J, Liao J B, Gao J Y, et al. Fusion of audio and vibration signals for bearing fault diagnosis based on a quadratic convolution neural network[J]. Sensors, 2023, 23(22): 9155.
[29] Zhang Z W, Jiao Z H, Li Y J, et al. Intelligent fault diagnosis of bearings driven by double-level data fusion based on multichannel sample fusion and feature fusion under time-varying speed conditions[J]. Reliability Engineering & System Safety, 2024, 251: 110362.
[30] Liu X R, Chen Z H, Hu D F, et al. A novel framework based on complementary views for fault diagnosis with cross-attention mechanisms[J]. Electronics, 2025, 14(5): 886-886.
[31] Zhao Y J, Wei S C, Xu X Z. Bearing fault diagnosis based on wavelet adaptive threshold filtering and multi-channel fusion cross-attention neural network[J]. Review of Scientific Instruments, 2024, 95(11): 114703.
[32] Sehri M, Dumond P, Bouchard M. University of Ottawa constant load and speed rolling-element bearing vibration and acoustic fault signature datasets[J]. Data in Brief, 2023, 49: 109327.

多维度评价

Viewed

Full text

Abstract

Cited

Shared

Discussed