Planetary Gearbox Fault Diagnosis based on Dynamic Adversarial Network

Abstract

Aiming at the problem of low fault diagnosis rate of planetary gearbox caused by insufficient fault data in actual industry, a method was proposed based on dynamic adversarial network. The domain-shared one dimensional feature extraction network was first built from the fault data. Secondly, the global discriminator and subdomain discriminator was used to align adaptively the distribution of fault features, and the weight index was used to evaluate the relative weight of those discriminators. Then, the joint loss of weight index, discriminator loss and classification loss is used as the objective function for training. Finally, the trained model was used to identify the target domain health conditions. The experimental results show that the proposed dynamic adversarial network may achieve better accuracy in a small amount of fault data.