Assessing Lightweight YOLO Models for Vision-Based Fault Diagnosis in Wind Turbine and Cold Chain Inspection Systems

The reliability of wind energy systems depends on the timely detection of surface and structural defects in turbine blades. This paper compares six YOLO (You Only Look Once) architectures (v8n–v13n) for automatic fault detection in wind turbine inspection images. All models were trained under identical experimental settings and evaluated by precision, recall, mAP@0.50, mAP@[0.50:0.95], and inference latency. Results show that YOLOv12n achieved the highest performance (mAP@0.50 = 0.867, computed as the mean over three seeds using the Top-3 class protocol), while YOLOv10n delivered the lowest inference time of 0.7 ms. These findings support the suitability of lightweight YOLO variants for real-time fault inspection and predictive maintenance in wind energy systems and the same framework can be extended to visual quality inspection tasks in intelligent cold chain logistics systems.