摘要
针对轴承故障诊断方法容易忽视故障图像细粒度特征学习的问题,提出了一种基于双线性改进残差神经网络的故障诊断方法。首先,采用格拉姆角场将振动信号转换成二维图像,保证故障信息的时序特征;其次,通过改进的残差网络挖掘深层故障特征;然后,建立双线性特征层学习细粒度特征;最后,经过全连接层完成故障识别。实验首先在凯斯西楚大学轴承数据集进行多种复杂负载下的有效性验证,其次与经典的轴承故障诊断算法对比,证明该方法具备一定的优势,最后在西安交通大学轴承数据集上验证了该方法的泛化性。
Aiming at the problem that bearing fault diagnosis methods tend to ignore the fine-grained feature learning of fault images,a diagnosis method based on bilinear improved residual neural network is proposed in this paper.Firstly,the vibration signal is converted into 2D image by Gram Angle field to ensure the timing characteristics of fault information.Secondly,the improved residual network is used to explore the deep fault characteristics.Then bilinear feature layer is established to learn the fine-grained features.Finally,the fault identification is completed through the full connection layer.In the experiment,the effectiveness of this method is verified under various complex loads on the bearing data set of CWRU,and then compared with the classical bearing fault diagnosis algorithm,which proves that the method has certain advantages.Finally,the generalization of this method is verified on the bearing data set of XJTU-SY.
作者
荀志文
缪小冬
顾寅骥
孙奥云
任万凯
XUN Zhiwen;MIAO Xiaodong;GU Yinji;SUN Aoyun;REN Wankai(School of Mechanical and Power Engineering,Nanjing Tech University,Nanjing 211800,China;College of Computer and Information Engineering,Nanjing Tech University,Nanjing 211800,China)
出处
《组合机床与自动化加工技术》
北大核心
2025年第2期126-130,共5页
Modular Machine Tool & Automatic Manufacturing Technique
基金
国家自然科学基金项目(52175465)。
关键词
滚动轴承
故障诊断
残差神经网络
细粒度
rolling bearing
fault diagnosis
residual neural network
fine-grained
