Compound Fault Diagnosis of Planetary Gearbox Using Capsule Network with Multi-Scale Convolutional Neural Network

Capsule network is an innovative deep learning-based model that uses neuron vectors as inputs and outputs of the network. It can achieve independent output of multiple classification labels, and solve the problem of decoupling recognition of compound fault components in planetary gearboxes. However, the single-scale feature extraction cannot fully explore multi-scale feature information in compound fault signals. This study proposes a capsule network combined with multi-scale convolutional feature fusion to diagnose the compound fault of planetary gearbox. The multi-scale features are extracted from STFT images of vibration signals using convolutional layers with different kernel sizes; secondly, a capsule network is designed to classify and identify the fault components by using the dynamic routing algorithm. The experiments are conducted to assess the performance of the proposed method, and the results indicates that it can effectively decoupling identify the compound faults with strong anti-interference and stability.