Classification of partial discharge severities of ceramic insulators based on texture analysis of UV pulses

Inspection of partial discharge before contamination flashover is of great importance for preventing exterior insulation accidents. In this study, a new method for identification of discharge severities is proposed. Specifically, a low‐cost ultraviolet (UV) sensor detection system was combined with...

Full description

Saved in:
Bibliographic Details
Published in:High voltage 2021-12, Vol.6 (6), p.986-996
Main Authors: Ma, Danrui, Jin, Lijun, He, Jianming, Gao, Kai
Format: Article
Language:English
Subjects:
Arc discharges
Cameras
ceramic insulators
Ceramics
corona
Cost analysis
Decision trees
Electric arcs
Fault detection
Feature extraction
Flashover
image classification
Insulators
Leakage current
Measurement techniques
Pollutants
Sensors
Signal classification
Spectrograms
Stress concentration
Support vector machines
Texture
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Inspection of partial discharge before contamination flashover is of great importance for preventing exterior insulation accidents. In this study, a new method for identification of discharge severities is proposed. Specifically, a low‐cost ultraviolet (UV) sensor detection system was combined with time–frequency method, texture analysis and support vector machine (SVM) classifier to classify partial discharge severities for ceramic insulators. The visible images and the root‐mean‐square value of leakage currents detected simultaneously are used to classify the UV signals into different discharge faults. The frequency and amplitude integration of UV pulses are minimum in corona discharge and larger in arc discharge. The images of UV signal spectrograms differ significantly at different discharge stages. The density and brightness of image textures are minimal in corona discharge and larger in arc discharge. Valid and reliable features selected by two texture feature extraction methods with SVM classifier have a reliable classification accuracy of 90.6% for ceramic insulators, and outperform a single time feature or other texture features. SVM outperforms k‐Nearest Neighbour, Naive Bayes and Decision Tree. Our new method with computational effectiveness and high practicality can solve the problem of high randomness and low accuracy of UV sensor detection. It can be further applied to the deterioration diagnosis of power facilities.
ISSN:2397-7264
2397-7264
DOI:10.1049/hve2.12081