Design and Implementation of Transmitter Unit and AI-Driven Monitoring System for High-Voltage Insulator Condition Assessment

Abstract

This capstone project presents the design and implementation of a lightweight UAV-mounted monitoring system for real-time condition assessment of high-voltage insulators. The system addresses limitations of manual, helicopter-based, and offline UAV inspection methods by integrating RGB visual inspection with edge AI inference on a Raspberry Pi 5 platform. The final prototype includes a camera, LTE communication module, regulated battery subsystem, active cooling, UPS HAT, and 3D-printed UAV enclosure, mounted on a DJI Air 2S platform for controlled testing. The selected deployment pipeline uses YOLO26n exported to NCNN, chosen for its balance of accuracy, inference speed, power consumption, and embedded deployability. Experimental validation showed that the system achieved mAP50-95 scores of 0.8984 for mechanical defects and 0.9315 for contamination, with real-time inference of approximately 15.3–15.4 FPS on Raspberry Pi 5. The integrated payload mass was 380 g, average power consumption was approximately 6.2 W, and end-to-end LTE-backed monitoring latency remained below the 600 ms requirement. Overall, the project demonstrates a low-cost, regulation-aware, and deployable edge-AI inspection prototype for safer and more scalable high-voltage insulator monitoring.