Deep Learning-Based Fault Diagnosis in Wind Turbine Bearings and Gearboxes Using Vibration Signals: Survey, Challenges, and Recommendations
Tales Moreira Tavares, Mateus Giesbrecht
Abstract
Wind power plays an important role in current global electric power generation scenario. With the increasing number of installed wind power generation units worldwide, efficient and cost-effective condition monitoring strategies become more relevant. Gearboxes and bearings are critical components in wind turbine (WT) operation, and enhancing the monitoring of these components contributes to improving wind power technology reliability and overall performance. In recent years, Deep Learning (DL) methods have emerged as tools for fault diagnosis in gearboxes and bearings in many different rotating machines, with the capacity to extract features from data and perform fault diagnosis in a more automated way. Even though DL methods can achieve high accuracy rates in diagnosis tasks, proper model training and application pose several challenges. In rotating machines fault diagnosis, DL models learn from labeled datasets; thus, these models can show poor generalization to classify data from unseen conditions. Such problem is concerning for WTfault diagnosis, where public available labeled datasets are scarce. In addition, DL models have inherently low interpretability, raising concerns about its application to real scenarios. In this context, this paper presents a literature review of the recent research on DL methods applied to fault diagnosis of WT bearings and gearboxes. The review focuses on the main challenges faced in this field, presenting and discussing the main strategies proposed to address them, including Transfer learning and explainable artificial intelligence techniques. This work aims to support future research and development on WT condition monitoring by providing a critical overview of current methodologies.