Automatic Detection and Classification of Faults in Wind Turbine Blades Using YOLO Family Models.
Wind energy, predictive fault detection, Electrical Signature Analysis (ESA), operational continuity.
Wind energy originated in Scotland in 1887 and became established in Denmark in 1976. In Brazil, its development began in 1992 on the island of Fernando de Noronha and expanded significantly over the following decades, driven by substantial investments and by the need to diversify the national energy matrix. As a result, wind power became one of the most competitive energy sources in Brazilian electricity auctions, reducing dependence on hydropower and strengthening energy security. In this context, the development of predictive analysis techniques has become essential to avoid high maintenance costs and operational downtime in wind farms. Wind turbines are composed of several components, each one associated with specific failure modes and repair times, and blade-related failures stand out due to their higher maintenance complexity and greater economic impact. To mitigate such costs, preventive inspection methods such as Electrical Signature Analysis (ESA) and the Wavelet transform have been developed for predictive fault detection. In addition, other techniques, including visual inspection, ultrasound, strain transducers, SCADA systems, and drone-based inspection, have been applied to identify and classify blade damage, each presenting specific advantages and limitations. Therefore, these methods contribute to improving the efficiency, reliability, and operational continuity of wind farms by reducing repair time and associated costs.