基于多传感器数据融合的互异网络轴承故障诊断方法  

Heterogeneous Network Bearing Fault Diagnosis Method Based on Multi-Sensor Data Fusion

作　　者：赵小强[1,2,3] 李森 ZHAO Xiaoqiang;LI Sen(School of Electrical Engineering and Information Engineering,Lanzhou University of Technology,Lanzhou 730050,China;Gansu Provincial Key Laboratory of Advanced Industrial Process Control,Lanzhou University of Technology,Lanzhou 730050,China;National Experimental Teaching Center of Electrical and Control Engineering,Lanzhou University of Technology,Lanzhou 730050,China)

机构地区：[1]兰州理工大学电气工程与信息工程学院,兰州730050 [2]兰州理工大学甘肃省工业过程先进控制重点实验室,兰州730050 [3]兰州理工大学国家级电气与控制工程实验教学中心,兰州730050

出　　处：《计算机工程与应用》2025年第5期323-333,共11页Computer Engineering and Applications

基　　金：国家自然科学基金(62263021);甘肃省教育厅产业支撑项目(2021CYZC-02);甘肃省科技专项项目(21YF5GA072)。

摘　　要：为了解决单传感器单一分支网络的输入容易受到外界干扰以及在不同域信号转换过程中丢失特征信息,导致故障诊断效果不佳的问题,提出了基于多传感器数据融合的互异网络轴承故障诊断方法。设计了数据预处理模块,以数据级的融合方式实现来自多传感器的多角度故障特征互补,充分考虑了轴承设备多传感器之间的相关性。同时,将经过快速傅里叶变换(FFT)和频率切片小波变换(FSWT)处理后的信号融合为多域信号作为模型的输入,以多域信号独立作为模型输入的形式确保不同域信号在转换过程中关键的特征信息不会丢失。该方法针对不同的域信号设计了相对应的互异网络结构对多传感器数据高维非线性空间中的低维特征关键提取,这也为设备维修人员提供了更加可靠方便的维修手段。当其中一个分支网络的输入受到外界干扰时,另外两个分支网络会起到纠错的作用,不仅增强了网络的容错能力,同时也会增加网络的特征互补能力。利用记忆单元将特征视为不同的时间步,以此建立不同故障特征之间的依赖关系。为了防止模型陷入局部最优,使用适配于所提模型的学习率余弦退火算法优化模型训练。在两个轴承数据集上进行实验,结果表明,该方法拥有好的故障诊断效果和泛化能力,可以满足基于多传感器数据融合的轴承故障诊断任务。To address the problems that the input of a single-sensor and a single-branch network is easily affected by external interference and that the loss of feature information during the conversion of signals from different domains leads to poor fault diagnosis,this paper proposes a fault diagnosis method for reciprocal network bearings based on multi-sensor data fusion.Firstly,the method designs a data preprocessing module to achieve complementary fault features from multiple sensors with multi-angle fault features by data-level fusion,which fully takes into account the correlation between multi-sensors of bearing equipment.At the same time,the signals processed by fast Fourier transform(FFT)and frequency sliced wavelet transform(FSWT)are fused into multi-domain signals as inputs to the model.Using independent multi-domain signals as model inputs ensures that the key feature information of different domain signals will not be lost during the conversion process.Secondly,the method designs a mutually exclusive network structure for different domain signals to extract lowdimensional features in the high-dimensional nonlinear space of multi-sensor data,which also provides a more reliable and convenient means of maintenance for equipment maintenance personnel.When the input of one of the branch networks is subjected to external interference,the other two branch networks will play the role of error correction,which not only enhances the fault tolerance of the network,but also increases the feature complementary ability of the network.Further,the features are regarded as different time steps by using the memory unit,which establishes the dependency relationship between different fault features.Finally,in order to prevent the model from falling into a local optimum,model training is optimized using a learning rate cosine annealing algorithm adapted to the proposed model.Experiments are conducted on two bearing datasets,and the results show that the method in this paper possesses good fault diagnosis results and generalizatio

关 键 词：滚动轴承 故障诊断 多传感器 互异网络 数据融合 特征互补 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]