Fault diagnosis method of rolling bearing based on VAE and manifold learning
-
摘要: 轴承作为转向架的关键部件,对列车的安全运行有着重要的影响。针对传统轴承故障诊断算法提取的特征在故障诊断模型中表现较差,且降维过程常丢失重要结构信息,提出一种结合变分自编码器(variational autoencoder, VAE)与流形学习的新型滚动轴承故障诊断算法——SFVD (SQuaDMDS-FItSNE and variational autoencoder-based diagnosis)。通过在β-VAE中引入残差结构和SimAM模块进行特征提取,并使用SQuaMDS与FIt-SNE的梯度混合方法对高维特征集进行数据降维,解决传统特征表征故障信息较差以及降维时数据结构保持性差等问题。基于上海地铁某线电机的滚动轴承振动信号数据进行验证,并与多种方法进行对比。结果表明,SFVD结合最小二乘支持向量机(least squares support vector machine, LS-SVM)的故障识别准确率显著优于其他方法,同时保持较优的时间性能。Abstract: As a key component of the bogie, bearings have a significant impact on the safe operation of trains. The features extracted by traditional bearing fault diagnosis algorithms were found to perform poorly in fault diagnosis models, and important structural information was often lost during the dimensionality reduction process. A novel rolling bearing fault diagnosis algorithm, termed SFVD (SQuaDMDS-FItSNE and variational autoencoder-based diagnosis), which integrates a variational autoencoder (VAE) with manifold learning, was proposed. Feature extraction was enhanced by introducing residual structures into the β-VAE for better gradient flow and incorporating a parameter-free SimAM attention module to focus on salient features while suppressing noise. Subsequently, the dimensionality of the high-dimensional feature set was reduced using a gradient hybrid method combining SQuaDMDS and FIt-SNE, in which gradient normalization and independent learning rates were employed to balance global and local structure preservation, thereby addressing the problems of poor fault information representation in traditional features and poor data structure preservation during dimensionality reduction. Validation experiments were conducted using vibration signal data from rolling bearings of a Shanghai Metro line motor, and comparisons were made with multiple existing methods. The results demonstrate that SFVD combined with a least squares support vector machine (LS-SVM) achieves significantly higher fault recognition accuracy than other methods while maintaining competitive time performance.
-
表 1 SKF-6016轴承参数
Table 1. Bearing parameters of SKF-6016
内圈故障特征
频率f/Hz外径D
/mm内径d
/mm滚球直径
r /mm接触角
α/ (˚)滚子数目 64 125 80 13.49 0 14 表 2 不同方法对比实验结果
Table 2. Experimental comparison results of different methods
算法
模型RACC最大值/
%RACC平均值/
%特征提取
时间/s训练时间/
sSFVD 100 99.8150 12.58 1.01 UMAP 97 91.8667 16.99 0.93 t-SNE 94 91.6000 11.98 1.77 PCA 97 86.9333 11.28 1.34 MDS 95 86.3333 13.18 1.83 表 3 消融实验结果
Table 3. Results of ablation
MODEL RACC RREC T R S SQ F √ √ √ √ 0.997 0.996 13.59 √ - √ √ 0.863 0.772 13.59 - √ √ √ 0.723 0.772 12.30 √ √ √ - 0.873 0.881 13.43 √ √ - √ 0.901 0.890 13.62 -
[1] 张伟. 基于卷积神经网络的轴承故障诊断算法研究[D] . 哈尔滨: 哈尔滨工业大学, 2017. [2] 孟静. 基于稀疏表示理论的轴承复合故障诊断方法研究[D] . 南京: 东南大学, 2021. [3] 翟正利, 梁振明, 周炜, 等. 变分自编码器模型综述[J] . 计算机工程与应用, 2019, 55(3): 1 − 9. doi: 10.3778/j.issn.1002-8331.1810-0284 [4] Van der Maaten L, Hinton G. Visualizing data using t-SNE[J] . Journal of Machine Learning Research, 2008, 9(86): 2579 − 2605. [5] Lambert P, De Bodt C, Verleysen M, et al. SQuadMDS: a lean Stochastic Quartet MDS improving global structure preservation in neighbor embedding like t-SNE and UMAP[J] . Neurocomputing, 2022, 503: 17 − 27. doi: 10.1016/j.neucom.2022.06.108 [6] 刘俊利, 缪炳荣, 张盈, 等. 一种基于改进VMD和UMAP的滚动轴承故障特征提取方法[J] . 机械传动, 2023, 47(6): 130 − 138. doi: 10.16578/j.issn.1004.2539.2023.06.019 [7] 杨青, 刘彦俏, 吴东升, 等. 基于DR-VAE与自注意力机制的电机轴承故障诊断[J] . 轴承, 2023(9): 72 − 79. doi: 10.19533/j.issn1000-3762.2023.09.013 [8] 赵连伟, 罗四维, 赵艳敞, 等. 高维数据流形的低维嵌入及嵌入维数研究[J] . 软件学报, 2005, 16(8): 1423 − 1430. [9] 肖婷, 汤宝平, 秦毅, 等. 基于流形学习和最小二乘支持向量机的滚动轴承退化趋势预测[J] . 振动与冲击, 2015, 34(9): 149 − 153. doi: 10.13465/j.cnki.jvs.2015.09.027 [10] De Leeuw J, Mair P. Multidimensional scaling using majorization: SMACOF in R[J] . Journal of Statistical Software, 2009, 31(3): 1 − 30. doi: 10.18637/jss.v031.i03 [11] Nesterov Y. A method of solving a convex programming problem with convergence rate $ O\left(\frac{1}{k^{2}}\right) $[J] . Doklady Akademii Nauk SSSR, 1983, 269(3): 543 − 547. [12] 陈俊芬, 韩金池, 谢博鋆, 等. 基于变分自编码器与流形特征的聚类算法[J] . 山西大学学报(自然科学版), 2024, 47(1): 69 − 80. doi: 10.13451/j.sxu.ns.2023139 [13] 涂望明, 宋执环, 陈运涛, 等. 基于小波变换和LS-SVM的雷达故障诊断[J] . 控制工程, 2013, 20(2): 309 − 312. doi: 10.3969/j.issn.1671-7848.2013.02.027 [14] 夏均忠, 苏涛, 张阳, 等. 基于EEMD能量熵及LS-SVM滚动轴承故障诊断[J] . 噪声与振动控制, 2014, 34(3): 170 − 175. doi: 10.3969/j.issn.1006-1335.2014.03.036 -
下载: