基于特征融合与 ResNet 的行星齿轮箱故障诊断

针对行星齿轮箱振动信号相互耦合和故障诊断不准确等问题，提出一种基于特征融合与深度残差网络(ResNet)的行星齿轮箱故障诊断方法。首先，对采集到的行星轮裂纹、磨损，太阳轮断齿及复合故障等模拟故障振动信号应用多维集成经验模态分解(MEEMD)和VMD进行分解，分别筛选确定有效分量。然后，将筛选出的有效特征进行融合，分别应用传统卷积神经网络(CNN)和深度残差网络对其进行分类识别。结果发现，深度残差网络，分类准确度更高，可达95%以上。最后，应用深度残差对特征融合前后数据的分类准确度进行了比较。融合前准确度最高只达91.16%,低于融合的97.18%。可见，该方法对行星齿轮箱耦合振动信号的处理和故障诊断非常有效。

Fault diagnosis of planetary gearboxes based on feature fusion and ResNet

Aiming at the coupling of vibration signals and inaccurate fault diagnosis of planetary gearbox, a fault diagnosis method of planetary gearbox based on feature fusion and ResNet is proposed. Firstly, the collected analog fault vibration signals such as planetary gear crack, wear, sun gear broken tooth and composite fault are decomposed by MEEMD and VMD to screen and determine the effective components respectively. Then, the selected effective features are fused and classified by using traditional CNN network and ResNet. The results show that the ResNet has higher classification accuracy, up to more than 95%. Finally, the classification accuracy of data before and after feature fusion is compared by using ResNet. The accuracy before fusion was only 91. 16%, which was lower than 97. 18% of after fusion. Thus, this method is very effective for coupling vibration signal processing and fault diagnosis of planetary gearbox.