A Computer Vision System for Monitoring Disconnect Switches in Distribution Substations

Knowing the state of the disconnect switches in a power distribution substation is important, since incorrect operation may lead to problems such as accidents, outages, and damaged equipment. Although human errors when reporting switch positions are rare, they can have a large impact when they occur...

Full description

Saved in:
Bibliographic Details
Published inIEEE transactions on power delivery Vol. 37; no. 2; pp. 833 - 841
Main Authors Nassu, Bogdan Tomoyuki, Marchesi, Bruno, Wagner, Rafael, Gomes, Victor Barpp, Zarnicinski, Vanderlei, Lippmann, Lourival
Format Journal Article
LanguageEnglish
Published New York IEEE 01.04.2022
The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
Subjects
Cameras
Computer vision
Deep learning
disconnect switches
distribution substation
Electric power distribution
High-voltage techniques
Human error
Image classification
Machine learning
Maintenance engineering
Monitoring
Substations
Switches
Training
Vision systems
Online AccessGet full text

Cover

More Information
Summary:Knowing the state of the disconnect switches in a power distribution substation is important, since incorrect operation may lead to problems such as accidents, outages, and damaged equipment. Although human errors when reporting switch positions are rare, they can have a large impact when they occur, making automatic monitoring an attractive proposition, especially in substations without permanent staff. In this paper, we describe a non-intrusive computer vision system for monitoring the state of disconnect switches in distribution substations. The system employs regular surveillance cameras, which can also be used for other purposes, leading to lower cost and simpler installation and maintenance compared to individual per-switch sensors. Several challenges were addressed: cluttered backgrounds, occlusions, variations in lighting, weather and switch aspect, and the fact that it is often not possible to maneuver the switches during installation. We addressed these challenges by combining data provided by a human operator during installation with multiple machine learning models, relying on techniques such as SVMs for classifying SIFT or HOG features, and deep learning. By combining multiple outputs, the system achieved a 99.7365% success rate in experiments performed at a real substation over several days.
ISSN:0885-8977
1937-4208
DOI:10.1109/TPWRD.2021.3071971