Fault diagnosis method for wind turbine current signal based on relation networks and margin learning

Abstract The gearbox plays a crucial role in wind turbines, as they often operate in harsh environments for extended periods, resulting in a high failure rate for gearboxes. Therefore, fault diagnosis of gearboxes is essential. However, predicting these faults is challenging, especially when fault data is limited. To address this challenge, this study proposes a relation margin-learning neural network. The model uses the current signals of the gearbox as input and integrates two different branches—a relation module (RM) and a fully connected layer—to learn the features of the samples. The RM tightens within-class clusters, whereas the fully-connected branch enlarges between-class margins. Experiments were carried out on a real wind turbine to verify the proposed method. The results indicate that the model can achieve a high-accuracy of fault classification for gearbox faults. Ablation and comparison experiments were both conducted, which further confirms the superiority of the proposed method.