Optimised path planning and environment-adaptive defect detection for wind turbine blades based on improved ant colony algorithm and EP-YOLOv12n

Wind turbine blades are prone to surface defects due to prolonged exposure to complex environments, posing significant challenges for Non-Destructive Testing (NDT). To overcome the limitations of manual inspection and the inadequate accuracy of existing automated methods under complex weather conditions, this study addresses both inspection path planning and defect detection within the framework of Structural Health Monitoring (SHM). For path optimisation, an improved dynamic ant colony algorithm is proposed, which incorporates dynamic parameter adjustment and path smoothing. The generated optimal path measures 2777.87 metres with 180 turns, representing a reduction of 577.22 metres in length and 25 fewer turns compared to the original algorithm. For defect detection, a weather-condition dataset is constructed, and an Environment-Adaptive Multimodal Attention Module (EAMM) is designed based on YOLOv12n. The EAMM integrates an environment perception sub-module and introduces the Inner-Distance Intersection over Union (Inner-DIoU) loss. The resulting model, termed EP-YOLOv12n, achieves a mAP50 of 99.04% and a Frames Per Second (FPS) rate of 157.58, enabling high-precision, real-time blade defect detection for intelligent wind farm operations. Finally, the generalisation experiment across diverse weather scenarios validates that the EP-YOLOv12n model possesses excellent cross-domain generalisation capability and detection robustness under different complex weather environments.