Intelligent State Evaluation and Fault Diagnosis Methods for Wind Turbines

Wind power is an important green and sustainable source of power generation. However, the construction of wind farms needs a large amount of initial investment and a highly expensive maintenance cost for wind turbines (WTs). Therefore, it is crucially important to accurately assess the state of WTs and keep WTs in good operation conditions. This thesis proposes a framework for the novel intelligent state evaluation and maintenance arrangement (iSEMA) system based on digital twin (DT) and deep learning (DL) technology, which can accurately evaluate the state of WTs and detect faults in the early stage. With the proposed iSEMA system, operators can better monitor WT conditions and use the information provided by the iSEMA system to schedule maintenance work more effectively, thereby minimizing downtime and boosting the economic efficiency of WTs.
In addition, this thesis creatively gives the concept of the sub-healthy state of WTs by introducing some relevant quantities, which is very useful for designing the iSEMA system and better describing the state of WTs. Based on the historical data of WTs, the iSEMA system can automatically select the suitable value of the parameters.
In the research about bearing fault diagnosis methods, this thesis proposed a novel hybrid framework by combining traditional machine learning (ML) and advanced DL models. The proposed hybrid framework can not only realize the automatic feature extraction for random forest (RF), but also significantly improve the accuracy for handling multi-classification problems.
Based on the auxiliary classifier generative adversarial networks (ACGANs) which is an advanced variant of the generative adversarial network (GAN), this thesis proposed an improved bearing fault diagnosis method for handling the imbalanced data problem. This method addresses the issue of insufficient fault data in real-world scenarios, which leads to low accuracy of data-driven methods in fault diagnosis.