一种基于小样本声音信号的托辊故障诊断方法

郝洪涛; 邱园园; 丁文捷

doi:10.13272/j.issn.1671-251x.2022120007

一种基于小样本声音信号的托辊故障诊断方法

doi: 10.13272/j.issn.1671-251x.2022120007

郝洪涛^{1, 2,},
邱园园¹,
丁文捷^{1, 2}

1.
宁夏大学机械工程学院, 宁夏银川　750021
2.
宁夏智能装备CAE重点实验室, 宁夏银川　750021

基金项目: 宁夏自然科学基金项目（2021AAC03046）；2023年宁夏回族自治区重点研发计划项目（2023BDE03005) 。

详细信息

作者简介:
郝洪涛(1976—)，男，宁夏银川人，教授，硕士研究生导师，博士，研究方向为机电设备智能运维、汽车先进制造技术，E-mail：haoht_03@126.com

中图分类号: TD634
计量
- 文章访问数: 678
- HTML全文浏览量: 54
- PDF下载量: 32
- 被引次数: 0
出版历程
- 收稿日期: 2022-12-02
- 修回日期: 2023-08-12
- 网络出版日期: 2023-09-04

A fault diagnosis method for roller based on small sample sound signals

HAO Hongtao^{1, 2
,},
QIU Yuanyuan¹,
DING Wenjie^{1, 2}

1.
School of Mechanical Engineering, Ningxia University, Yinchuan 750021, China
2.
Ningxia Key Laboratory of CAE on Intelligent Equipment, Yinchuan 750021, China

摘要

摘要: 基于深度学习的故障诊断方法对数据集的质量有很高要求，需要大批量数据才能进行良好的模型训练，从而实现准确的故障诊断，而在实际应用中能够采集到的故障信号通常很有限。针对托辊故障声音信号获取困难、样本量少，导致智能故障诊断方法性能受限的问题，提出了一种基于小样本声音信号的托辊故障诊断方法。使用特征转换方法将一维声音信号转换为二维时频图像，将频率域的特征融入进来，以提高数据集对故障特征的表达能力；提出了多种类型时频图结合的数据集扩充方法，将短时傅里叶变换（STFT）、连续小波变换（CWT）、希尔伯特−黄变换（HHT） 3种时频分析方法绘制的时频图相结合，以扩充数据集，增加数据样式；引入了深度迁移学习的思想，使用轴承数据集对模型进行预训练，然后使用托辊数据对预训练模型进行微调，以进一步提升模型的识别准确率。实验结果表明：多种类型时频图结合的数据集扩充方法能有效解决使用小样本数据训练模型时易过拟合的问题；使用迁移学习后，模型的测试准确率达98.81%，相较于不使用迁移学习时提升了7%，且没有出现过拟合现象，说明模型训练良好；相较于生成对抗网络扩充STFT时频图数据集+迁移学习的方法，多种类型时频图结合的数据集扩充+迁移学习的方法准确率提高了4%，且更容易实现，可解释性更强。
- 带式输送机 /
- 托辊 /
- 故障诊断 /
- 小样本 /
- 时频图 /
- 数据集扩充 /
- 迁移学习
Abstract: Fault diagnosis methods based on deep learning have high requirements for the quality of the dataset, requiring a large amount of data for good model training to achieve accurate fault diagnosis. However, the fault signals that can be collected in practical applications are usually limited. A method for diagnosing roller faults based on small sample sound signals is proposed to address the problem of limited performance of intelligent fault diagnosis methods due to the difficulty in obtaining sound signals for roller faults and the small sample size. The feature transformation method is used to convert one-dimensional sound signals into two-dimensional time-frequency images, incorporating features from the frequency domain to improve the dataset's capability to express fault features. A dataset expansion method combining multiple types of time-frequency maps has been proposed. The method combines time-frequency maps drawn by three time-frequency analysis methods: short time fourier transform (STFT), continuous wavelet transform (CWT), and Hilbert Huang transform (HHT) to expand the dataset and increase data styles. The concept of deep transfer learning is introduced, using bearing datasets to pre-train the model, and then using roller data to fine-tune the pre-trained model to further improve the recognition accuracy of the model. The experimental results show that the dataset expansion method combining multiple types of time-frequency maps can effectively solve the problem of overfitting when training models with small sample data. After using transfer learning, the testing accuracy of the model reaches 98.81%, an improvement of 7% compared to not using transfer learning. There was no overfitting phenomenon, indicating that the model is well-trained. Compared to the method of generating adversarial networks to expand the STFT time-frequency map dataset and transfer learning, the method of dataset expansion by combining multiple types of time frequency maps and transfer learning has an accuracy improvement of 4%. It is easier to implement, and has stronger interpretability.
- belt conveyor /
- roller /
- fault diagnosis /
- small sample /
- time-frequency image /
- dataset expansion /
- transfer learning

HTML全文

图 1 不同时频图局部放大图对比

Figure 1. Comparison of partial enlarged images of different time-frequency images

下载: 全尺寸图片幻灯片

图 2 迁移学习与传统机器学习的区别

Figure 2. The difference between transfer learning and traditional machine learning

下载: 全尺寸图片幻灯片

图 3 托辊故障诊断流程

Figure 3. Flow of roller fault diagnosis

下载: 全尺寸图片幻灯片

图 4 多种类型时频图结合的数据集扩充方法

Figure 4. Dataset enhancements method combining multiple types of time-frequency images

下载: 全尺寸图片幻灯片

图 5 残差卷积神经网络结构

Figure 5. Structure of residual convolutional neural network

下载: 全尺寸图片幻灯片

图 6 故障诊断模型训练流程

Figure 6. Training process of fault diagnosis model

下载: 全尺寸图片幻灯片

图 7 托辊故障模拟平台

Figure 7. Roller fault simulation platform

下载: 全尺寸图片幻灯片

图 8 轴承故障综合模拟器

Figure 8. Bearing fault comprehensive simulation test bench

下载: 全尺寸图片幻灯片

图 9 训练准确率及损失变化曲线

Figure 9. Training accuracy and loss change curves

下载: 全尺寸图片幻灯片

图 10 模型测试混淆矩阵

Figure 10. Model testing confusion matrix

下载: 全尺寸图片幻灯片

表 1 时频分析方法的主要参数

Table 1. Main parameters of time-frequency analysis methods

时频分析方法	参数名称	参数
短时傅里叶变换	窗类型	汉宁窗
	窗长	6 000
	窗之间重叠点数	3 000
	窗内离散傅里叶变换点数	3 000
连续小波变换	母小波函数	Amor小波

下载: 导出CSV

表 2 残差卷积神经网络参数

Table 2. Parameters of residual convolutional neural network

名称	类型	描述	输出特征大小
Conv1	卷积层	64个7×7卷积核，步长为2；激活函数为ReLU；批量归一化	112×112×64
M1	最大池化层	3×3池化核，步长为2	56×56×64
Conv2	卷积层	64个3×3卷积核，步长为2；激活函数为ReLU	56×56×64
Conv3	卷积层	64个3×3卷积核，步长为2；激活函数为ReLU；批量归一化	56×56×64
Add	加法	将M1和Conv3的输出按元素相加	56×56×64
Conv4	卷积层	128个3×3卷积核，步长为2；激活函数为ReLU	28×28×128
M2	最大池化层	3×3池化核，步长为2	14×14×128
FC1	全连接层	128个节点；激活函数为ReLU	1×1×128
FC2	全连接层	5/4个节点；激活函数为Softmax	1×1×5/1×1×4

下载: 导出CSV

表 3 托辊数据集

Table 3. Roller dataset

数据集	分析方法	每类故障所用图像张数
数据集	分析方法	训练	验证	测试
TG−1	STFT	11	5	14
TG−2	CWT	11	5	14
TG−3	HHT	11	5	14
TG−4	STFT，CWT	22	10	28
TG−5	STFT，HHT	22	10	28
TG−6	CWT，HHT	22	10	28
TG−7	STFT，CWT，HHT	34	14	42

下载: 导出CSV

表 4 不同托辊数据集训练RCNN的结果

Table 4. The results of training RCNN using different roller datasets %

数据集	验证准确率	测试准确率
TG−1	100	69.64
TG−2	96.00	75.00
TG−3	100	62.50
TG−4	87.50	80.35
TG−5	87.50	82.14
TG−6	90.00	79.46
TG−7	89.29	91.07

下载: 导出CSV

表 5 轴承结构参数

Table 5. Structural parameters of bearing

轴承型号	轴承节径/mm	滚子直径/mm	滚子个数	接触角/(°)
ER−12K	33.5	7.9	8	0

下载: 导出CSV

表 6 时频图原图与遮挡灵敏度图对比

Table 6. Comparison between the original time-frequency images and the occlusion sensitivity images

托辊状态	STFT		CWT		HHT
托辊状态	时频图原图	遮挡灵敏度图	时频图原图	遮挡灵敏度图	时频图原图	遮挡灵敏度图
正常托辊
托辊轴承滚珠缺失
托辊掺沙
托辊卡滞

下载: 导出CSV

表 7 不同方法实验结果对比

Table 7. Comparison of experimental results of different methods

方法	方法描述	准确率/%
文献[24]	一维信号+一维卷积神经网络	59.38
文献[25]	STFT时频图数据集+迁移AlexNet	76.79
文献[12]	生成对抗网络扩充STFT时频图数据集+迁移学习	94.35
本文方法	多种类型时频图结合的数据集扩充+迁移学习	98.81

下载: 导出CSV

参考文献(25)

[1]	焦贺彬. 煤矿带式输送机智能化安全监测系统研究[J]. 煤矿机械,2020,41(10):182-185. doi: 10.13436/j.mkjx.202010058 JIAO Hebin. Research on intelligent safety monitoring system of belt conveyor in coal mine[J]. Coal Mine Machinery,2020,41(10):182-185. doi: 10.13436/j.mkjx.202010058
[2]	付朕. 矿用带式输送机托辊远程故障诊断系统[D]. 徐州: 中国矿业大学, 2020. FU Zhen. Remote fault diagnosis system of mine belt conveyor idler[D]. Xuzhou: China University of Mining and Technology, 2020.
[3]	邵思羽. 基于深度学习的旋转机械故障诊断方法研究[D]. 南京: 东南大学, 2019. SHAO Siyu. Methodologies for fault diagnosis of rotary machine based on deep learning[D]. Nanjing: Southeast University, 2019.
[4]	吴文臻,程继明,李标. 矿用带式输送机托辊音频故障诊断方法[J]. 工矿自动化,2022,48(9):25-32. WU Wenzhen,CHENG Jiming,LI Biao. Audio fault diagnosis method of mine belt conveyor roller[J]. Journal of Mine Automation,2022,48(9):25-32.
[5]	贺志军,李军霞,张伟,等. 基于MFCC 特征和GWO−SVM的托辊故障诊断[J]. 机床与液压,2022,50(15):188-193. HE Zhijun,LI Junxia,ZHANG Wei,et al. Roller fault diagnosis based on MFCC feature and GWO-SVM[J]. Machine Tool & Hydraulics,2022,50(15):188-193.
[6]	陈维望,李军霞,张伟. 基于分支卷积神经网络的托辊轴承故障分级诊断研究[J]. 机电工程,2022,39(5):596-603. doi: 10.3969/j.issn.1001-4551.2022.05.004 CHEN Weiwang,LI Junxia,ZHANG Wei. Hierarchical fault diagnosis of idler bearing based on branch convolutional neural network[J]. Journal of Mechanical & Electrical Engineering,2022,39(5):596-603. doi: 10.3969/j.issn.1001-4551.2022.05.004
[7]	WEN Long,LI Xinyu,GAO Liang,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
[8]	宁夏大学. 带式输送机故障巡检载具及其控制系统和控制方法: CN201911098745.5[P]. 2019-11-12. Ningxia University. Belt conveyor fault inspection vehicle and its control system and control method: CN201911098745.5[P]. 2019-11-12.
[9]	WANG Zirui,WANG Jun,WANG Youren. An intelligent diagnosis scheme based on generative adversarial learning deep neural networks and its application to planetary gearbox fault pattern recognition[J]. Neurocomputing,2018,310(8):213-222.
[10]	DING Yu,MA Liang,MA Jian,et al. A generative adversarial network-based intelligent fault diagnosis method for rotating machinery under small sample size conditions[J]. IEEE Access,2019,7:149736-149749. doi: 10.1109/ACCESS.2019.2947194
[11]	何强,唐向红,李传江,等. 负载不平衡下小样本数据的轴承故障诊断[J]. 中国机械工程,2021,32(10):1164-1171,1180. doi: 10.3969/j.issn.1004-132X.2021.10.004 HE Qiang,TANG Xianghong,LI Chuanjiang,et al. Bearing fault diagnosis method based on small sample data under unbalanced loads[J]. China Mechanical Engineering,2021,32(10):1164-1171,1180. doi: 10.3969/j.issn.1004-132X.2021.10.004
[12]	吴定会,方钦,吴楚宜. 基于数据生成与迁移学习的轴承小样本故障诊断[J]. 机械传动,2020,44(11):139-144. doi: 10.16578/j.issn.1004.2539.2020.11.023 WU Dinghui,FANG Qin,WU Chuyi. Bearing small sample fault diagnosis based on data generation and transfer learning[J]. Journal of Mechanical Transmission,2020,44(11):139-144. doi: 10.16578/j.issn.1004.2539.2020.11.023
[13]	蒋杰. 基于深度学习的车型识别算法研究[D]. 北京: 北方工业大学, 2018. JIANG Jie. Vehicle recognition algorithm based on deep learning[D]. Beijing: North China University of Technology, 2018.
[14]	ZHANG Hongyi, CISSE M, DAUPHIN Y N, et al. Mixup: beyond empirical risk minimization[EB/OL]. [2022-11-25].https://arxiv.org/abs/1710.09412.
[15]	DVORNIK N, MAIRAL J, SCHMID C. Modeling visual context is key to augmenting object detection datasets[EB/OL]. [2022-11-25]. https://arxiv.org/abs/1807.07428.
[16]	LEE W J,WU Haiyue,HUANG Aihua,et al. Learning via acceleration spectrograms of a DC motor system with application to condition monitoring[J]. The International Journal of Advanced Manufacturing Technology,2020,106(3/4):1-14.
[17]	BERA A, DUTTA A, DHARA A K. Deep learning based fault classification algorithm for roller bearings using time-frequency localized features[C]. International Conference on Computing, Communication, and Intelligent Systems, Greater Noida, 2021.
[18]	LI Pengfei,YUAN Hejin,WANG Yu,et al. Pumping unit fault analysis method based on wavelet transform time-frequency diagram and CNN[J]. International Core Journal of Engineering,2020,6(1):182-188.
[19]	XU Yang, LI Zhixiong, WANG Shuqing, et al. A hybrid deep-learning model for fault diagnosis of rolling bearings[J]. Measurement, 2021, 169（6）. DOI: 10.1016/j.measurement.2020.108502.
[20]	YUAN Zhuang, ZHANG Laibin, DUAN Lixiang, et al. Intelligent fault diagnosis of rolling element bearings based on HHT and CNN[C]. Prognostics and System Health Management Conference, Chongqing, 2018.
[21]	PHAM M T, KIM J M, KIM C H. Accurate bearing fault diagnosis under variable shaft speed using convolutional neural networks and vibration spectrogram[J]. Applied Sciences, 2020, 10（18）. DOI: 10.3390/app10186385.
[22]	HE Zhiyi,SHAO Haidong,ZHANG Xiaoyang,et al. Improved deep transfer auto-encoder for fault diagnosis of gearbox under variable working conditions with small training samples[J]. IEEE Access,2019,7:115368-115377. doi: 10.1109/ACCESS.2019.2936243
[23]	WANG Chongyu,XIE Yonghui,ZHANG Di. Deep learning for bearing fault diagnosis under different working loads and non-fault location point[J]. Journal of Low Frequency Noise,Vibration and Active Control,2021,40(1):588-600. doi: 10.1177/1461348419889511
[24]	WANG Xin, MAO Dongxing, LI Xiaodong. Bearing fault diagnosis based on vibro-acoustic data fusion and 1D-CNN network[J]. Measurement, 2021, 173. DOI: 10.1016/j.measurement.2020.108518.
[25]	MINERVINI M, HAUSMSN S, FROSINI L, et al. Transfer learning technique for automatic bearing fault diagnosis in induction motors[C]. IEEE 13th International Symposium on Diagnostics for Electrical Machines, Power Electronics and Drives, Dallas, 2021.