High-Fidelity Reconstruction Method for Rolling Bearing Fault Features Based on Bessel Series Empirical Wavelet Transform and Attention-Guided Denoising Autoencoder
-
摘要:
针对强背景噪声干扰下滚动轴承故障特征重构波形易失真的问题，提出一种融合贝塞尔级数经验小波变换（BSEWT）与注意力引导降噪自编码器（AGDA）的方法. 首先，通过贝塞尔基展开将时域信号映射至数据驱动的能量谱，结合多尺度峰值检测策略实现对频带边界的自适应识别，分离出包含故障信息的模态分量；其次，构建AGDA模型，利用注意力加权机制重构背景噪声分布，从含噪模态中逆向剥离噪声成分；最后，结合仿真信号与凯斯西储大学（CWRU）公开轴承数据集，研究该方法在5种转速工况和多种信噪比水平下的特征提取性能. 研究结果表明：在频带划分方面，BSEWT相比于传统经验小波变换（EWT），能够自适应收窄频带并精细隔离复杂窄带干扰；在5种不同转速的仿真实验中，当输入信噪比高于10分贝时，BSEWT-AGDA方法的输出信噪比、均方根误差与归一化相关系数指标均全面优于变分模态分解（VMD）、经验模态分解（EMD）和EWT等方法；在输入信噪比低于负五分贝的环境下，BSEWT方法在输出信噪比指标上表现最优；在CWRU数据验证中，经该方法处理后的信号峭度值为
8.1635 、脉冲因子为8.0741 ，高于VMD、EMD等对比算法；BSEWT能够克服传统方法依赖固定频带划分的缺陷，实现复杂频谱的精准分割；结合AGDA模型后，在较高信噪比区间内有效剔除噪声干扰，实现了滚动轴承故障特征的高保真重构.Abstract:To address the waveform distortion of rolling bearing fault features under strong background noise interference, a method combining Bessel series empirical wavelet transform (BSEWT) and attention-guided denoising autoencoder (AGDA) was proposed. First, the time-domain signal was mapped to a data-driven energy spectrum via Bessel basis expansion. By using a multi-scale peak detection strategy, frequency band boundaries were adaptively identified to separate modal components containing fault information. Second, an AGDA model was constructed to reconstruct the background noise distribution by using an attention weighting mechanism, reversely stripping noise components from noisy modes. Finally, the feature extraction property under five rotational speed conditions and various signal-to-noise ratio (SNR) levels was evaluated by using simulated signals and the public bearing dataset of Case Western Reserve University (CWRU). The results show that BSEWT adaptively narrows the frequency bands and finely isolates complex narrow-band interference better than the traditional empirical wavelet transform (EWT) in frequency band division. In simulation experiments with five different rotational speeds, when the input SNR exceeds 10 dB, the output SNR, root mean square error, and normalized correlation coefficient of the BSEWT-AGDA method are comprehensively superior to those of variational mode decomposition (VMD), empirical mode decomposition (EMD), and EWT. For input SNRs below −5 dB, the BSEWT method achieves the optimal output SNR index. In CWRU data validation, the signal processed by the proposed method yields a kurtosis of 8.163 5 and an impulse factor of 8.074 1, surpassing comparative algorithms like VMD and EMD. BSEWT overcomes the defect of relying on fixed frequency band division in traditional methods, achieving precise complex frequency spectrum segmentation. Combined with the AGDA model, it effectively eliminates noise interference in higher SNR ranges and achieves high-fidelity reconstruction of rolling bearing fault features.
-
Key words:
- Bessel series /
- empirical wavelet transform /
- feature extraction /
- autoencoder /
- denoising
-
图 7 BSEWT和EWT的频带划分对比
Figure 7. Comparison of frequency band division between BSEWT and EWT
图 9 不同转速下各方法在3个信噪比下的性能对比
Figure 9. Property comparison of different methods at three SNRs under various rotational speeds
表 1 网络结构参数表
Table 1. Network structure parameters
模块 层类型 输入通道 输出通道 核大小/
步长激活函数 编码器 Conv1d 1 32 11/1 ReLU MaxPool1d 32 32 2/2 Conv1d 32 16 11/1 ReLU MaxPool1d 16 16 2/2 解码器 ConvT1d 16 16 11/2 ReLU ConvT1d 16 32 11/2 ReLU Conv1d 32 1 11/1 Linear 
下载: 导出CSV
表 2 信号参数与模型训练超参数
Table 2. Signal parameters and model training hyperparameters
参数 值 参数 值 采样率 （FS）/Hz 12500 批处理大小 32 训练样本长度/s 0.4 训练轮次 30 评估样本长度/s 0.4 学习率 1e-5 故障频率（BPFO）/Hz 108 优化器 Adam BSEWT展开项数 3000 损失函数 MSE
下载: 导出CSV
表 3 CWRU数据处理性能指标对比
Table 3. CWRU data processing property index comparison
方法 峭度 脉冲因子 原故障信号 8.1635 8.0675 EMD 8.1130 8.0265 VMD 6.2235 6.6900 EWT 6.6553 6.8470 BSEWT 2.7344 3.5351 BSEWT-AGDA 8.1635 8.0741
下载: 导出CSV
-
[1] 李睿智, 杨芳华, 张伟, 等. 基于ShuffleNet-DELM的轴承故障诊断研究[J]. 中国测试, 2024, 50(6): 42-48. doi: 10.11857/j.issn.1674-5124.2022050075Li Ruizhi, Yang Fanghua, Zhang Wei, et al. Research on bearing fault diagnosis based on ShuffleNet-DELM[J]. China Measurement & Testing Technology, 2024, 50(6): 42-48. doi: 10.11857/j.issn.1674-5124.2022050075 [2] Li Q. New approach for bearing fault diagnosis based on fractional spatio-temporal sparse low rank matrix under multichannel time-varying speed condition[J]. IEEE Transactions on Instrumentation and Measurement, 2022, 71: 3502012. doi: 10.1109/tim.2022.3146601 [3] Choudhary A, Goyal D, Letha S S. Infrared thermography-based fault diagnosis of induction motor bearings using machine learning[J]. IEEE Sensors Journal, 2021, 21(2): 1727-1734. doi: 10.1109/JSEN.2020.3015868 [4] Li Y, Zhou J W, Li H G, et al. A fast and adaptive empirical mode decomposition method and its application in rolling bearing fault diagnosis[J]. IEEE Sensors Journal, 2023, 23(1): 567-576. doi: 10.1109/JSEN.2022.3223980 [5] Yang Y, Liu H, Han L J, et al. A feature extraction method using VMD and improved envelope spectrum entropy for rolling bearing fault diagnosis[J]. IEEE Sensors Journal, 2023, 23(4): 3848-3858. doi: 10.1109/JSEN.2022.3232707 [6] 刘秀丽, 王鸽, 吴国新, 等. VMD及PSO优化SVM的行星齿轮箱故障诊断[J]. 电子测量与仪器学报, 2022(1): 54-61. doi: 10.13382/j.jemi.B2104376Liu Xiuli, Wang Ge, Wu Guoxin, et al. Fault diagnosis method of planetary gear box based on variationalmodal decomposition and particle swarmoptimization support vector machine[J]. Journal of Electronic Measurement and Instrumentation, 2022(1): 54-61. doi: 10.13382/j.jemi.B2104376 [7] Gilles J. Empirical wavelet transform[J]. IEEE Transactions on Signal Processing, 2013, 61(16): 3999-4010. doi: 10.1109/TSP.2013.2265222 [8] 盛嘉玖, 陈果, 康玉祥, 等. 一种用于滚动轴承故障诊断的改进EWT方法[J]. 航空动力学报, 2024, 39(9): 364-376. doi: 10.13224/j.cnki.jasp.20220677Sheng Jiajiu, Chen Guo, Kang Yuxiang, et al. An improved EWT method for fault diagnosis of rolling bearings[J]. Journal of Aerospace Power, 2024, 39(9): 364-376. doi: 10.13224/j.cnki.jasp.20220677 [9] 张锐, 刘婷婷, 王燕, 等. 基于FBSE-ESEWT的齿轮故障诊断方法[J]. 电子测量与仪器学报, 2025, 39(4): 234-246. doi: 10.13382/j.jemi.B2407639Zhang Rui, Liu Tingting, Wang Yan, et al. Gear fault diagnosis method based on FBSE-ESEWT[J]. Journal of Electronic Measurement and Instrumentation, 2025, 39(4): 234-246. doi: 10.13382/j.jemi.B2407639 [10] Bhattacharyya A, Singh L, Pachori R B. Fourier–Bessel series expansion based empirical wavelet transform for analysis of non-stationary signals[J]. Digital Signal Processing, 2018, 78: 185-196. doi: 10.1016/j.dsp.2018.02.020 [11] Chaudhary P K, Gupta V, Pachori R B. Fourier-Bessel representation for signal processing: a review[J]. Digital Signal Processing, 2023, 135: 103938. doi: 10.1016/j.dsp.2023.103938 [12] 王圣博, 姜孝谟, 陈丙炎, 等. 基于增强时变形态学滤波的铁路车辆轴箱轴承故障诊断[J/OL]. 西南交通大学学报, 2025, [13] 赵光权, 葛强强, 刘小勇, 等. 基于DBN的故障特征提取及诊断方法研究[J]. 仪器仪表学报, 2016, 37(9): 1946-1953. doi: 10.3969/j.issn.0254-3087.2016.09.004Zhao Guangquan, Ge Qiangqiang, Liu Xiaoyong, et al. Fault feature extraction and diagnosis method based on deep belief network[J]. Chinese Journal of Scientific Instrument, 2016, 37(9): 1946-1953. doi: 10.3969/j.issn.0254-3087.2016.09.004 [14] Wen L, Li X Y, Gao L, et al. A new convolutional neural network-based data-driven fault diagnosis method[J]. IEEE Transactions on Industrial Electronics, 2018, 65(7): 5990-5998. doi: 10.1109/TIE.2017.2774777 [15] Gu Y, Cao J W, Song X, et al. A denoising autoencoder-based bearing fault diagnosis system for time-domain vibration signals[J]. Wireless Communications and Mobile Computing, 2021, 2021(1): 9790053. doi: 10.1155/2021/9790053 [16] 梅瀚雨, 廖海黎, 王昌将. 基于自编码器的非线性气动力辨识及非线性颤振分析[J]. 西南交通大学学报, 2025, 60(3): 599-607. doi: 10.3969/j.issn.0258-2724.20230261Mei Hanyu, Liao Haili, Wang Changjiang. Nonlinear aerodynamic force identification and nonlinear flutter analysis based on autoencoder[J]. Journal of Southwest Jiaotong University, 2025, 60(3): 599-607. doi: 10.3969/j.issn.0258-2724.20230261 [17] 朱前坤, 谢辰辉, 张琼, 等. 基于计算机视觉和深度学习的古桥裂缝识别方法[J]. 西南交通大学学报, 2026, 61(2): 529-540. doi: 10.3969/j.issn.0258-2724.20250134Zhu Qiankun, Xie Chenhui, Zhang Qiong, et al. Method for crack detection of ancient bridges based on computer vision and deep learning[J]. Journal of Southwest Jiaotong University, 2026, 61(2): 529-540. doi: 10.3969/j.issn.0258-2724.20250134 [18] 张泽, 周斌, 董伟, 等. 基于CNN的汽轮机故障特征提取方法[J]. 发电设备, 2025, 39(6): 362-366, 374. doi: 10.19806/j.cnki.fdsb.2025.06.003Zhang Ze, Zhou Bin, Dong Wei, et al. A fault feature extraction method for steam turbines based on CNN[J]. Power Equipment, 2025, 39(6): 362-366, 374. doi: 10.19806/j.cnki.fdsb.2025.06.003 [19] 徐波, 魏艺君, 曾晗, 等. 基于多源数据融合和优化深度置信网络的变压器故障诊断方法[J]. 计算机与现代化, 2025(11): 106-111. doi: 10.3969/j.issn.1006-2475.2025.11.013Xu Bo, Wei Yijun, Zeng Han, et al. Transformer fault diagnosis method based on multi-source data fusion and optimized deep belief network[J]. Computer and Modernization, 2025(11): 106-111. doi: 10.3969/j.issn.1006-2475.2025.11.013 [20] 安春丽, 张碧玲, 赵国安, 等. 基于FFT-CNN-GCN的电网故障诊断[J]. 浙江大学学报(工学版), 2025, 59(10): 2205-2212.An Chunli, Zhang Biling, Zhao Guoan, et al. Power grid fault diagnosis based on FFT-CNN-GCN[J]. Journal of Zhejiang University (Engineering Science), 2025, 59(10): 2205-2212. [21] 罗金, 童靳于, 郑近德, 等. 基于EEMD和堆叠稀疏自编码的滚动轴承故障诊断方法[J]. 噪声与振动控制, 2020, 40(2): 115-120. doi: 10.3969/j.issn.1006-1355.2020.02.021Luo Jin, Tong Jinyu, Zheng Jinde, et al. Fault diagnosis method for rolling bearings based on EEMD and stacked sparse auto-encoder[J]. Noise and Vibration Control, 2020, 40(2): 115-120. doi: 10.3969/j.issn.1006-1355.2020.02.021 [22] 张磊, 张勇, 陈永强, 等. 基于VDM-BP神经网络的电动机泵故障诊断技术研究[J]. 机械工程师, 2025(11): 171-175. doi: 10.3969/j.issn.1002-2333.2025.11.037Zhang Lei, Zhang Yong, Chen Yongqiang, et al. Research on motor fault diagnosis technology based on VDM-BP neural network[J]. Mechanical Engineer, 2025(11): 171-175. doi: 10.3969/j.issn.1002-2333.2025.11.037 [23] 胡宇宏. 基于CEEMD与CNN的数控车床刀具磨损智能监测方法研究[J]. 湖南工业职业技术学院学报, 2025, 25(5): 13-18. doi: 10.13787/j.cnki.43-1374/z.2025.05.003Hu Yuhong. Research on intelligent monitoring method of CNC lathe tool wear based on CEEMD and CNN[J]. Journal of Hunan Industry Polytechnic, 2025, 25(5): 13-18. doi: 10.13787/j.cnki.43-1374/z.2025.05.003 [24] 田晶, 谢肇阳, 吕静宜, 等. 基于改进变模态分解和宽核深度卷积神经网络的中介轴承故障诊断方法[J]. 推进技术, 2026(1): 309-320. doi: 10.3724/1001-4055.202504015Tian Jing, Xie Zhaoyang, Lyu Jingyi, et al. A fault diagnosis method for intermediate bearings based on improved variable mode decomposition and wide-kernel deep convolutional neural networks[J]. Journal of Propulsion Technology, 2026(1): 309-320. doi: 10.3724/1001-4055.202504015 [25] Olver, Frank WJ. NIST handbook of mathematical functions hardback and CD-ROM. [S. l. ]: Cambridge University Press, 2010. [26] Canny J. A computational approach to edge detection[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 1986, PAMI-8(6): 679-698. [27] Hu J, Shen L, Sun G. Squeeze-and-excitation networks[C]//2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Salt Lake City: IEEE, 2018: 7132-7141. [28] Randall R B, Antoni J. Rolling element bearing diagnostics: a tutorial[J]. Mechanical Systems and Signal Processing, 2011, 25(2): 485-520. doi: 10.1016/j.ymssp.2010.07.017 [29] Wu Z D, Zhang Z, Zheng L, et al. The denoising method for transformer partial discharge based on the whale VMD algorithm combined with adaptive filtering and wavelet thresholding[J]. Sensors, 2023, 23(19): 8085. doi: 10.3390/s23198085 -
下载:
点击查看大图
计量
- 文章访问数: 2
- HTML全文浏览量: 1
- PDF下载量: 0
- 被引次数: 0