Autonomous Drone-Based Powerline Insulator Inspection via Deep Learning

Abstract

Accumulation of pollutants on ceramic insulators is one of the major causes of dry band arcing, a predecessor to flashovers, which may further cause major outages of electricity. It is critical to know locations of polluted insulators to prevent flashovers to make the power-grid reliable. This paper proposes a solution to detect the location of polluted insulators along an overhead transmission line using a quadcopter. Once provided with the GPS locations of the electrical powerline transmission towers, the quadcopter autonomously hovers along the line. And while doing so, it sends a live video feed of the transmission line to the ground station. A pre-trained neural network on the ground station then detects insulators in the video and classifies the detected insulators as polluted or clean. Only if the insulator detected is polluted, its location is recorded and reported back to the ground station. The novelty of this work is the use of a drone to automate the process of insulator inspection via a deep learning based neural network approach. Experiments show that accurate inspection results are obtained. This work is an initial step in the direction of achieving completely autonomous drone-based powerline insulator inspection.