Improved YOLOv11 Detection Method to Study the Stability Analysis of Wind Turbine Tower Cranes

Abstract

To address the issues of low efficiency, poor accuracy, and weak environmental adaptability in detecting pitting on the wear-resistant epoxy mortar surfaces of the intake tower flow channels at the Xiaolangdi Project on the Yellow River, an intelligent epoxy mortar paving system was designed. Concurrently, for tunnel simulation paving experiments, an improved YOLOv11s detection model incorporating the CBAM attention mechanism was proposed. A hybrid dataset was constructed, combining transfer learning with incremental training to enhance model generalization capabilities, and deployed on Jetson Nano edge devices. Experiments demonstrated that the model achieved mAP of 92.1% and 91.5% in conventional road and epoxy mortar scenarios, respectively. Following TensorRT optimization, inference speed increased from 89.6 FPS to 101.2 FPS, with a single-frame detection time of 9.9 ms, meeting real-time requirements. This technology effectively achieves automated high-precision detection of potholes in simulated environments, providing reliable support for anti-abrasion surface construction in hydraulic engineering projects.