Ultrasonic Intelligent Diagnostic Method for the Internal Defects in Ultrahigh Voltage Basin Insulators Using Improved YOLOv8

This paper presents an ultrasonic intelligent diagnostic method for internal defects in ultra-high voltage basin insulators using an improved YOLOv8 model. This approach addresses the limitations of manual interpretation in traditional ultrasonic inspections by enabling automatic analysis and intelligent defect identification. The proposed method incorporates a deeper network architecture with optimized loss functions and adaptive feature extraction mechanisms, significantly enhancing adaptability in complex scenarios. Compared to YOLOv5 and YOLOv8n, the improved model achieves a 7% and 3% increase in F1 score, respectively, along with a 13.6% and 10.3% improvement in [email protected]. Experimental validation on standard defect samples and a 550 kV three-phase common box basin insulator demonstrates the model’s capability to detect 1 mm-scale bubbles and cracks with less than 2% localization error, over 80% diagnostic confidence, and 93.5% consistency with industrial Digital Radiography inspections, providing a robust and intelligent solution for defect detection in critical power system components.