基于EMATE和POA-ELM的声音信号故障诊断方法

扫码查看

原文链接

NETL
NSTL
万方数据

中文摘要：常规的工程机械故障诊断方法一般需对振动信号进行分析,但采集振动信号时需要使振动传感器与工程机械相接触,在某些情况下工程机械表面不适合安装传感器,如设备的温度较高或者传感器的安装空间有限.针对这些问题,以声音信号作为故障诊断对象,提出了一种基于增强多尺度注意熵(EMATE)和鹈鹕优化算法优化极限学习机(POA-ELM)的工程机械故障诊断方法.首先,利用声音传感器采集了工程机械不同故障的声音信号,避免了振动传感器存在的接触式采集缺陷;然后,利用EMATE提取了声音信号中的故障信息,建立了表征工程机械不同故障状态的特征向量;接着,鉴于ELM的参数需要优化的问题,采用POA对ELM的关键参数进行了寻优,建立了参数自适应设置的ELM分类模型;最后,利用POA-ELM分类器对故障特征进行了辨识,实现了工程机械的故障识别,并利用往复压缩机和滚动轴承的声音信号数据集对基于EMATE-POA-ELM的故障诊断方法的有效性进行了验证.研究结果表明:将EMATE方法作为故障特征提取指标能够取得 100%和 99.23%的识别准确率,且特征提取的时间仅为53.88 s和172.47 s;与多尺度注意熵、复合多尺度注意熵、时移多尺度注意熵等指标相比,EMATE的平均故障识别准确率更高,并具有更好的综合性能.

外文标题：Fault diagnosis method of sound signal based on EMATE and POA-ELM

外文摘要：The conventional fault diagnosis method of engineering machinery generally analyzes the vibration signal,but the vibration sensor needs to contact with the engineering machinery when the vibration signal is collected.In some cases,the engineering machinery surface is not suitable for installing the sensor,such as the temperature of the equipment is high or the installation space of the sensor is limited.To solve these problems,a fault diagnosis method for the engineering machinery based on the enhanced multiscale attention entropy(EMATE)and the pelican optimization algorithm optimized extreme learning machine(POA-ELM)was proposed,and the sound signal was taken as the fault diagnosis object.Firstly,the sound sensor was used to collect the sound signals of different faults of the engineering machinery to avoid the contact acquisition defects existing in the vibration sensor.Then,EMATE was used to extract fault information from the sound signal and establish feature vectors that represent different fault states of engineering machinery.Subsequently,considering the issue of optimizing the parameters of ELM,POA was used to optimize the key parameters of ELM,and an ELM classification model with adaptive parameter settings was established.Finally,the POA-ELM classifier was used to identify fault features and achieve fault identification of engineering machinery.The effectiveness of EMATE-POA-ELM based fault diagnosis methods was validated using the sound signal dataset of reciprocating compressor and rolling bearing.The research results indicate that using the EMATE method as a fault feature extraction indicator can respectively achieve a recognition accuracy of 100%and 99.23%,and the feature extraction time is only 53.88 s and 172.47 s.Comparing with indicators such as multi-scale attention entropy,composite multi-scale attention entropy,and time-shifted multi-scale attention entropy,EMATE has higher average fault recognition accuracy and better comprehensive performance.

外文关键词：

engineering machineryreciprocating compressorrolling bearingfault data setenhanced multiscale attention entropy(EMATE)fault diagnosispelican optimization algorithm optimized extreme learning machine(POA-ELM)

作者：

徐浙君、王凯、罗少杰、崔炳荣

展开 >

作者单位：

重庆大学计算机学院,重庆 400030

浙江邮电职业技术学院人工智能学院,浙江绍兴 312366

国网浙江省电力有限公司设备管理部,浙江杭州 310063

国网杭州市余杭区供电公司科技数字化部,浙江杭州 311199

北京智芯微电子科技有限公司数字芯片设计中心,北京 100192

展开 >

关键词：

工程机械往复压缩机滚动轴承故障数据集增强多尺度注意熵故障诊断鹈鹕优化算法优化极限学习机

基金：

国家电网浙江省电力公司科技项目

项目编号：

B311HZ220003

出版年：

2024

DOI：

10.3969/j.issn.1001-4551.2024.06.002

机电工程

浙江大学　浙江省机电集团有限公司

机电工程

CSTPCD北大核心

影响因子：0.785

ISSN：1001-4551

年,卷(期)：2024.41(6)