A multi-learner neural network approach to wind turbine fault diagnosis with imbalanced data

The data imbalance problem extensively exists in wind turbine fault diagnosis, resulting in the compromise between learning attention to majority and minority classes. In this paper, a deep neural network method is proposed to resolve the mentioned problem. Specifically, convolutional and recurrent neural networks are designed to extract spatial and temporal features within supervisory control and data acquisition (SCADA) measurements. To improve the reliability of fault diagnosis results by collective decision, a coarse learner and multiple fine learners are established. With the consideration of data imbalance and learning diversity, fault-related information can be revealed. Moreover, a learner selection scheme is designed to ensure high compu-tational efficiency. The effectiveness of the proposed method is demonstrated by experiments based on simulated data and real-world SCADA measurements from a wind farm. Experimental results show that the accuracy in identifying health conditions can be improved by the proposed method regardless of the data imbalance. On the two datasets, the proposed method outperforms four benchmark approaches as the learning attention to all classes can be enhanced. Therefore, the proposed method is a promising solution to wind turbine fault diagnosis.

Automated summary

A deep neural network method is proposed to address the data imbalance problem in wind turbine fault diagnosis. Convolutional and recurrent neural networks are used to extract spatial and temporal features from SCADA measurements. By employing a coarse learner and multiple fine learners, the reliability of fault diagnosis results is improved. A learner selection scheme is also designed to ensure computational efficiency. Experimental results demonstrate the effectiveness of the proposed method in improving accuracy and enhancing learning attention to all classes, making it a promising solution to wind turbine fault diagnosis.