基于多元分解和深度学习的机器故障检测方法的研究  

Machine Fault Detection Method Based on Multivariate Decomposition and Deep Learning

作　　者：林香[1] LIN Xiang(School of Intelligent Manufacturing,Minxi Vocational and Technical College,Longyan Fujian 364021,China)

机构地区：[1]闽西职业技术学院智能制造学院,福建龙岩364021

出　　处：《菏泽学院学报》2023年第2期63-68,共6页Journal of Heze University

基　　金：福建省中青年教师教育科研项目(JZ181049)。

摘　　要：基于声学的机器故障检测方法是一种新颖的研究方法,但是需要依赖采集足够丰富的故障类型数据.因此,提出一种无监督的机器故障检测方法,即使用机器正常运行状态的数据作为训练样本,从而鉴别出机器异常状态.该检测方法的一个关键问题是信号中的噪声,它会影响检测的性能,针对这个问题,提出了一种基于多元张量分解的准非参数光谱数据去噪策略,即非负正则多重分解,这种策略特别适用于发出稳定声音的机器.通过实验证明了这种声学检测方法,可以使得工业过程的故障监控更加准确.Acoustic-based machine fault detection is a novel research method,but it needs to rely on sufficient data collection of fault types.Therefore,this paper proposes an unsupervised fault detection method,which uses data from the normal operating state of the machine as training samples to identify the abnormal state.A key issue of this method is the noise in the signal,which can affect the performance of detection.Therefore,the paper proposes a quasi nonparametric spectral data de-noising strategy based on multivariate tensor decomposition,namely,non-negative regular multivariate decomposition.This strategy is particularly suitable for machines that produce stable sound.Experiments show that the acoustic detection method can make machine fault monitoring in industrial processes more reliable.

关 键 词：多元分解 深度学习 无监督机器故障检测 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] TH17[自动化与计算机技术—控制科学与工程]