基于深度学习信息融合的喷泵复合故障诊断  

Composite Fault Diagnosis of Water-jet Pump Based on Deep Learning Information Fusion

作　　者：李刚强[1,2] 谢福起 崔永智 徐增丙 LI Gangqiang;XIE Fuqi;CUI Yongzhi;XU Zengbing(Marine Design and Research Institute of China,Shanghai 200011,China;Key Laboratory of Water Jet Propulsion Technology,Shanghai 200011,China;Wuhan University of Science and Technology School of Mechanical Engineering,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]武汉科技大学机械工程学院,湖北武汉430081

出　　处：《电子产品可靠性与环境试验》2024年第S1期93-101,共9页Electronic Product Reliability and Environmental Testing

摘　　要：喷泵作为舰船航行的核心驱动部件,在复杂工况和恶劣环境下存在轴承、叶轮等重要部件故障监测与诊断难题,进而影响舰船正常运行。针对复杂工况下喷泵轴承损伤、叶轮刮擦等复合故障的诊断难题,提出了基于改进软投票多源信息融合的深度学习故障诊断方法。首先,利用快速傅里叶变换将导叶外壳和叶轮外壳振动时域信号分别转换成频域信号,并分别输入基于Yu范数的深度度量学习模型和深度置信模型进行初步诊断分析;然后,利用改进软投票表决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,通过试验台架对喷泵导叶轴承、叶轮刮擦等复合故障展开故障模拟,并采用所提出的诊断方法进行故障分析。结果表明所提出方法的诊断精度高达99.667%,精度明显优于基于单个传感器信号和基于软投票多源信息融合的诊断模型。As the core driving component of ship navigation,the water-jet pump faces challenges in monitoring and diagnosing important components such as bearings and impellers under complex working conditions and harsh environments,which in turn affects the normal operation of the ship.A deep learning fault diagnosis method based on improved soft voting multi-source information fusion is proposed to address the diagnostic challenges of composite faults such as damage to spray pump bearings and impeller scraping under complex working conditions.Firstly,using the FFT method,the time-domain vibration signals of the guide vane casing and impeller casing are respectively transformed into frequency-domain signals,and the signals are input into the Yu norm based deep metric learning model and deep confidence model for diagnostic analysis to obtain preliminary diagnostic results respectively.Then,using the improved soft voting method,these preliminary diagnostic results are fused to obtain the final diagnostic result.Finally,a fault simulation is conducted on the water-jet pump guide vane bearing,impeller scraping and other composite faults through a test bench,and the proposed diagnostic method is used for fault analysis.The results show that the diagnostic accuracy of this proposed method is 99.667%,which is significantly better than that of the diagnostic model based on single sensor signal and multi-source information fusion based on soft voting.

关 键 词：喷泵 多信息融合 改进软投票法 深度学习 复合故障诊断 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论] TB114.39[自动化与计算机技术—计算机科学与技术]