基于特征融合和ResNet的滚动轴承故障诊断

Fault diagnosis of rolling bearing based on feature fusion and ResNet

由于滚动轴承信号非平稳、非线性,导致轴承的故障信息提取困难,并且采用传统故障诊断方法诊断精度低,又过度依赖专家经验以及深度学习的故障诊断方法需海量的训练数据,为了解决上述问题,提出了一种基于特征融合和深度残差神经网络(ResNet)的滚动轴承故障诊断方法。首先,利用变分模态分解(VMD)和经验模态分解(EMD)的方法分解了原始信号;然后,根据方差贡献率和相关系数筛选确定了有效分量,对筛选出的有效分量进行了特征融合,组成数据集输入到ResNet模型中,并进行了故障诊断;最后,利用开源数据集对基于特征融合和深度残差神经网络(ResNet)的滚动轴承故障诊断方法进行了可行性和有效性验证,并通过滚动轴承实例数据验证了其泛化能力和鲁棒性。研究结果表明:在开源数据集中,采用该方法所获得的故障识别率达到了99.8%,相比于传统卷积神经网络(CNN)90%的故障识别率,其故障识别率更高;在滚动轴承实例数据集中,采用该方法所获得的故障识别率达到了97%以上,进一步证明了特征融合结合深度残差神经网络的故障诊断方法可有效应用于滚动轴承故障诊断中。

In order to solve the problems that it was difficult to extract fault features caused by non-stationary and nonlinear vibration signals of rolling bearings,traditional fault diagnosis methods relied too much on expert experience and deep learning fault diagnosis methods required massive training data,etc.,a rolling bearing fault diagnosis method based on feature fusion and deep residual neural network(ResNet)was proposed.Firstly,the original signal was decomposed by variational modal decomposition(VMD)and empirical mode decomposition(EMD).Then,the effective components were determined by variance contribution rate and correlation coefficient screening,and the selected effective components were subjected to feature fusion to form a data set and input them into the ResNet model for fault diagnosis.Finally,the feasibility and effectiveness of the fault diagnosis method for rolling bearings based on feature fusion and depth residual neural network(ResNet)were verified by using open source data sets,and the rolling bearing example was used to verify its generalization ability and robustness.The test results show that the method has a fault recognition rate of 99.8%in the open-source data set,which is higher than the 90%fault recognition rate of the traditional convolutional neural network(CNN).In the rolling bearing example data set,the fault recognition rate reaches more than 97%,which further proves that the fault diagnosis method of feature fusion combined with deep residual neural network can be effectively applied to rolling bearing fault diagnosis.