A Modified Deep Convolutional Subdomain Adaptive Network Method for Fault Diagnosis of Wind Turbine Systems
Yijun Shen, Bo Chen, Fanghong Guo, Wenchao Meng, Li Yu
Abstract
In most practical situations, there is not enough labeled data when designing the learning algorithm for the fault diagnosis of wind turbine systems (WTSs). Even if the labeled data can be obtained from some similar machines, these labeled data may not be directly used for the target task due to the difference in data distribution. To solve this problem, a deep convolutional subdomain adaptive network (DCSAN) for WTSs fault diagnosis is proposed in this article based on transfer learning. First, the deep convolutional network is used to extract the features of multiscale vibration signals and map them to the same feature space. Then the fully connected neural network is used for fault classification, while the multikernel subdomain local maximum mean discrepancy (MK-LMMD) loss is used to reduce the distance between the source and target domain features in the same subspace. Finally, the effectiveness and advantages of the proposed method are verified by 12 transfer fault diagnosis experiments.