基于VMD-CWT和改进CNN的直升机轴承故障诊断

由于直升机自动倾斜器滚动轴承振动信号具有非平稳、非线性特点,并夹杂非敏感故障特征信息,导致网络模型对周期信号过于敏感,不能充分利用故障信息的问题;针对此问题,提出一种变分模态分解(VMD)与连续小波变换(CWT)联合提取敏感故障特征的方法。研究表明:在相同模型训练下,该方法相对其他方法最高可提升模型准确率20.8%。为了解决卷积神经网络(CNN)进一步提高故障识别精度难的问题,提出一种基于K最近邻(KNN)改进的CNN的模型,在课题组和西储大学公开轴承数据集验证,测试精度达到99.8%和100%,可有效实现直升机自动倾斜器滚动轴承的故障诊断。

Fault diagnosis of helicopter bearing based on VMD-CWT and improved CNN

The vibration signal of the helicopter automatic tilter rolling bearing of non-stationary and non-linear characteristics was mixed with non-sensitive fault characteristic information, causing the network model to be too sensitive to periodic signals to make full use of the information of fault. To solve this problem, a variational modal decomposition (VMD) and continuous wavelet transform (CWT) were combined to extract sensitive fault features. The experimental result showed that the accuracy of the model can be improved by 20.8% compared with other methods, under the training of the same model. In order to solve the problem that Convolutional neural network (CNN) has the difficulty to further improve the fault recognition accuracy, a model based on CNN improved by K-nearest neighbor (KNN) was proposed, and the test accuracy was 99.8% and 100%, respectively, in the research group and the public bearing data set of Western Reserve University, which can effectively realize the fault diagnosis of the helicopter automatic tilter rolling bearing.