基于贝叶斯优化多尺度DenseNet的离心泵声信号故障诊断方法

doi:10.3969/j.issn.1004-132X.2025.09.015

摘要/Abstract

摘要：

由于一维特征向量不能保留时间特征信息，而神经网络对图像识别具有良好效果，因此尝试用离心泵故障声信号构建的图像数据集开展离心泵故障诊断，提出贝叶斯优化多尺度DenseNet的离心泵声信号故障诊断方法。将一维时间序列声信号经过格拉姆角场转化为二维图像，保留其时间信息及故障特征；然后采用多尺度密集块对图像进行特征提取，增强图像特征复用；通过dropout层和L₂正则化方法防止过拟合，采用贝叶斯优化算法确定神经网络超参数，最后利用离心泵声信号进行实验验证，与其他诊断方法进行对比。结果表明，贝叶斯优化多尺度DenseNet的诊断模型对测试集具有99.5%的故障识别率。

关键词: 离心泵, 故障诊断, 格拉姆角场, 贝叶斯优化, 多尺度DenseNet

Abstract:

Since one-dimensional feature vectors might not retain temporal feature information， but neural networks had good effects on image recognition， an image data set constructed by fault sound signals of centrifugal pumps was used to conduct centrifugal pump fault diagnosis. A Bayesian optimized multiscale DenseNet fault diagnosis method was proposed for centrifugal pump sound signals. One-dimensional time series acoustic signals were transformed into two-dimensional image through Gram angle field， and the time information and fault characteristics were preserved. Then multiscale dense blocks were used to extract image features to enhance image feature reuse. The dropout layer and L₂ regularization method were used to prevent overfitting， and Bayesian optimization algorithm was adopted to determine neural network hyperparameters. Finally， experimental verification was performed using centrifugal pump acoustic signals， and comparisons were made with other diagnostic methods. The results show that the Bayesian optimization multiscale DenseNet diagnosis model has a fault recognition rate of 99.5% for the test set.

Key words: centrifugal pump, fault diagnosis, Gram angle field, Bayesian optimization, multiscale DenseNet

中图分类号:

TP183

陈剑, 严明辉, 陈品. 基于贝叶斯优化多尺度DenseNet的离心泵声信号故障诊断方法[J]. 中国机械工程, 2025, 36(09): 2032-2038.

Jian CHEN, Minghui YAN, Pin CHEN. Acoustic Signal Fault Diagnosis Method of Centrifugal Pumps Based on Bayesian Optimization Multiscale DenseNet[J]. China Mechanical Engineering, 2025, 36(09): 2032-2038.

导出引用管理器 EndNote|Ris|BibTeX

链接本文: https://www.cmemo.org.cn/CN/10.3969/j.issn.1004-132X.2025.09.015

https://www.cmemo.org.cn/CN/Y2025/V36/I09/2032

图/表 15

图1 GAF映射过程图

Fig.1 GAF mapping process diagram

图2 DenseNet结构图

Fig.2 DenseNet structure diagram

图3 多尺度Dense块结构图

Fig.3 Multiscale Dense block structure map

表1 多尺度DenseNet网络参数

Tab.1 Multiscale DenseNet network parameters

网络层名称	参数	输出尺寸
卷积层	3×3，步长1	64×64×64
池化层	3×3，步长3	32×32×64
Dense块1	密集连接层×3	32×32×256
过渡层1	过渡层×1	16×16×128
Dense块2	密集连接层×3	16×16×320
过渡层2	过渡层×1	8×8×128
Dense块3	密集连接层×3	8×8×320
过渡层3	过渡层×1	8×8×128
Dense块4	密集连接层×2	4×4×256
全局池化	平均池化	1×1×256
dropout	50%丢弃	1×1×256
全连接层		4

图4 贝叶斯优化多尺度DenseNet的故障诊断模型

Fig.4 Fault diagnosis model of multiscale DenseNet based on Bayesian optimization

图5 离心泵声信号采集试验台

Fig.5 Centrifugal pump acoustic signal acquisition test bench

图6 离心泵信号时域

Fig.6 Time domain signal of centrifugal pump

表2 离心泵故障标签及其样本量

Tab.2 Centrifugal pump fault labels and samples

故障类型	标签	训练集	验证集	测试集
正常	状态1	350	150	100
气蚀	状态2	350	150	100
轴不对中	状态3	350	150	100
螺栓松动	状态4	350	150	100

表3 超参数取值

Tab.3 Hyperparameter value

参数	初始学习率	动量	L₂正则化系数
取值区间	［0.0001，0.1］	［0.7，0.98］	［1×10^-10，1×10^-2］

图7 最小目标函数的观测值与估计值

Fig.7 The observed and estimated values of the minimum objective function

图8 贝叶斯优化多尺度DenseNet训练迭代曲线

Fig.8 Bayesian optimization multiscale DenseNet training iteration curve

图9 测试集诊断结果

Fig.9 Test set diagnostic results

图10 各模型验证集准确率迭代曲线

Fig.10 Accuracy iteration curve of each model verification

图11 各模型故障识别率对比

Fig.11 Comparison of fault recognition rate of each model

表4 各模型验证集和测试集准确率

Tab.4 Accuracies of each model verification set and test set

神经网络	验证集准确率/%	测试集准确率/%
CNN	68.10	68.25
DenseNet	83.54	80.5
多尺度DenseNet	91.72	91.25
本文方法	99.5	99

参考文献 15

[1]	柯耀，王琪，苗育茁，等. 基于PARAFAC分析和SVM离心泵故障诊断方法［J］. 噪声与振动控制， 2022， 42（1）：106-111.
	KE Yao， WANG Qi， MIAO Yuzhuo， et al. Fault Diagnosis Method for Centrifugal Pumps Based on PARAFAC Analysis and SVM［J］. Noise and Vibration Control， 2022， 42（1）：106-111.
[2]	范传翰，宋礼威，刘厚林，等. 基于PSO-SVM-RF的离心泵转子故障诊断研究［J］. 中国农村水利水电， 2023（2）：171-176.
	FAN Chuanhan， SONG Liwei， LIU Houlin， et al. Research on the Rotor Fault Diagnosis of the Centrifugal Pump Based on PSO-SVM-RF［J］. China Rural Water and Hydropower， 2023（2）：171-176.
[3]	梁兴，罗远兴，邓飞，等. 多传感器数据下基于MFDFA-BP的离心泵空化致振故障分析［J］. 振动与冲击， 2022， 41（17）：238-243.
	LIANG Xing， LUO Yuanxing， DENG Fei， et al. Cavitation Induced Vibration Fault Analysis of Centrifugal Pump Based on MFDFA-BP under Multi-sensor Data［J］. Journal of Vibration and Shock， 2022， 41（17）：238-243.
[4]	LI Shi， WANG Huaqing， SONG Liuyang， et al. An Adaptive Data Fusion Strategy for Fault Diagnosis Based on the Convolutional Neural Network［J］. Measurement， 2020， 165：108122.
[5]	HASAN M J， RAI A， AHMAD Z， et al. A Fault Diagnosis Framework for Centrifugal Pumps by Scalogram-based Imaging and Deep Learning［J］. IEEE Access， 2021， 9：58052-58066.
[6]	周玉蓉，张巧灵，于广增，等. 基于声信号的工业设备故障诊断研究综述［J］. 计算机工程与应用， 2023， 59（7）：51-63.
	ZHOU Yurong， ZHANG Qiaoling， YU Guangzeng， et al. Review of Acoustic Signal-based Industrial Equipment Fault Diagnosis［J］. Computer Engineering and Applications， 2023， 59（7）：51-63.
[7]	汪欣，毛东兴，李晓东. 基于声信号和一维卷积神经网络的电机故障诊断研究［J］. 噪声与振动控制， 2021， 41（2）：125-129.
	WANG Xin， MAO Dongxing， LI Xiaodong. Motor Fault Diagnosis Using Microphones and One-dimensional Convolutional Neural Network［J］. Noise and Vibration Control， 2021， 41（2）：125-129.
[8]	李少波，姚勇，桂桂，等. 基于CNN与多通道声学信号的齿轮故障诊断［J］. 中国测试， 2019， 45（10）：1-5.
	LI Shaobo， YAO Yong， GUI Gui， et al. Gear Fault Diagnosis Based on CNN and Multi-channel Acoustic Signals［J］. China Measurement & Test， 2019， 45（10）：1-5.
[9]	LI Congyue， HU Yihuai， JIANG Jiawei， et al. Fault Diagnosis of a Marine Power-generation Diesel Engine Based on the Gramian Angular Field and a Convolutional Neural Network［J］. Journal of Zhejiang University：Science A， 2024， 25（6）：470-482.
[10]	ZHANG Ke， GUO Yurong， WANG Xinsheng， et al. Multiple Feature Reweight DenseNet for Image Classification［J］. IEEE Access， 2019， 7：9872-9880.
[11]	WANG Shengwei， WANG Hongkui， XIANG Sen， et al. Densely Connected Convolutional Network Block Based Autoencoder for Panorama Map Compression［J］. Signal Processing：Image Communication， 2020， 80：115678.
[12]	葛战，李兵，孙磊，等. 基于星座图和密集连接网络的QAM信号识别［J］. 电子信息对抗技术， 2023， 38（1）：43-48.
	GE Zhan， LI Bing， SUN Lei， et al. Modulation Classification Based on Constellation Diagrams and DenseNet for QAM Signals［J］. Electronic Information Warfare Technology， 2023， 38（1）：43-48.
[13]	朱汇龙，刘晓燕，刘瑶. 基于贝叶斯新型深度学习超参数优化的研究［J］. 数据通信， 2019（2）：35-38.
	ZHU Huilong， LIU Xiaoyan， LIU Yao. Research on Hyper-parameter Optimization Based on Bayesian Deep Learning［J］. Data Communication，2019（2）：35-38.
[14]	贾寒冰，刘鹏，张雷，等. 基于规则与机器学习融合的换道决策建模方法研究［J］. 机械工程学报， 2022， 58（4）：212-221.
	JIA Hanbing， LIU Peng， ZHANG Lei， et al. Lane-changing Decision Model Development by Combining Rules Abstract and Machine Learning Technique［J］. Journal of Mechanical Engineering， 2022， 58（4）：212-221.
[15]	PENG Kang， JIANG Lizhong， ZHOU Wangbao， et al. A Seismic Response Prediction Method Based on a Self-optimized Bayesian Bi-LSTM Mixed Network for High-speed Railway Track-bridge System［J］. Journal of Central South University， 2024， 31（3）：965-975.