基于多域信息融合的深度学习轴承故障诊断方法  

Bearing fault diagnosis method based on deep learning of multi-domain information fusion

作　　者：葛卓 夏华猛 王凯亮 徐增丙[3] 丁改革 GE Zhuo;XIA Huameng;WANG Kailiang;XU Zengbing;DING Gaige(Marine Design and Research Institute of China,Shanghai 200011,China;Science and Technology on Water Jet Propulsion Laboratory,Shanghai 200011,China;School of Machinery and Automation,Wuhan University of Science and Technology,Wuhan 430081,China)

机构地区：[1]中国船舶及海洋工程设计研究院,上海200011 [2]喷水推进技术重点实验室,上海200011 [3]武汉科技大学机械自动化学院,武汉430081

出　　处：《振动与冲击》2024年第23期47-55,共9页Journal of Vibration and Shock

摘　　要：针对单一振动信号包含故障信息易被隐藏以及单一深度学习模型诊断能力不强导致轴承故障诊断精度低的问题,提出了一种基于多域信息融合的深度学习故障诊断方法。利用变分模态分解方法(variational mode decomposition,VMD)将原始振动信号分解为多个本征模态函数(intrinsic mode function,IMF)分量,同时对每个IMF分量进行快速傅里叶变换(fast Fourier transformation,FFT)转化为频域样本;然后将多个IMF分量和其对应频域样本分别输入至多个深度度量学习(deep metric learning,DML)模型和深度置信网络(deep belief network,DBN)模型分别进行初步诊断分析,并利用简单软投票法对这些初步诊断结果进行融合从而获取最终诊断结果。最后通过对不同轴承故障的诊断试验分析,结果表明,该研究提出的方法不仅具有较好的诊断效果,而且诊断性能分别优于基于时域和基于频域的信息融合诊断方法。Here,aiming at problems of easy concealment of fault information contained in a single vibration signal and weak diagnosis ability of a single deep learning model to cause low accuracy of bearing fault diagnosis,a bearing fault diagnosis method based on deep learning of multi-domain information fusion was proposed.Using the variational mode decomposition(VMD)method,the original vibration signal was decomposed into multiple intrinsic mode function(IMF)components,and FFT was performed for various IMF components to transform them into frequency-domain samples.Then,these IMF components and their corresponding frequency-domain samples were input into multiple deep metric learning(DML)models and deep belief learning(DBL)models for preliminary diagnosis analysis,and the simple soft voting method was used to fuse these preliminary diagnosis results,and obtain the final diagnosis results.Finally,through diagnosis tests and analyses for different bearing faults,the results showed that the proposed method not only has the better diagnosis effect,but also its diagnosis performance is superior to those of information fusion diagnosis methods based on time-domain and frequency-domain,respectively.

关 键 词：信息融合 深度度量学习(DML) 深度置信网络(DBN) 软投票法 

分 类 号：TH133.33[机械工程—机械制造及自动化] TP165.3[自动化与计算机技术—控制理论与控制工程]