Multicriteria Machine Learning Approach for Wind Turbine Fault Detection Using SCADA Data

The detection of faults in wind turbines is vital for maintaining efficiency, reducing unplanned downtime, and optimizing maintenance strategies. However, traditional fault detection methods often fail to address the complexity and multidimensional nature of turbine data. This paper presents an advanced data-driven approach to fault diagnosis in wind turbines, integrating machine learning techniques with multicriteria performance evaluation. By leveraging data from SCADA subsystem and IoT -enabled sensors, it proposed framework that utilizes deep learning models to automatically extract relevant features from raw sensor data and detect faults with high accuracy. The study also discussed challenges such as class imbalance and sensor noise by employing self-supervised learning and hybrid architecture. This research demonstrates the effectiveness of our approach through several case studies, comparing various machine learning models, including Support Vector Machines, LightGBM, and Deep Neural Networks, for their classification performance and computational efficiency. Results show that models like the Weighted Multi-Scale Global-Local Feature Fusion Network and Convolutional Temporal-Spatial Attention Networks offer substantial improvements in fault detection accuracy and early warning capabilities. Furthermore, a decision-making framework was introduced to help select the most suitable fault detection approach, considering turbine configuration, data availability, and operational priorities. The outcomes of this study provide a roadmap for enhancing fault diagnosis in wind turbines through the deployment of robust, data-driven models that consider multiple operational metrics.