Pollution Monitoring on Polymeric Insulators Adopting Laser-Induced Breakdown Spectroscopy, Computer Vision, and Machine Learning Techniques

Abstract

Timely detection of pollution deposit on insulator is crucial for safety operation of the power system network. In this study, a novel sensing technique is proposed, which combines laser-induced breakdown spectroscopy (LIBS) with computer vision and machine learning techniques, for accurate classification of type of pollutant deposits and severity of the pollution on outdoor insulators. The method involves 1) analyzing silicone rubber insulators coated with various pollutants using LIBS; 2) converting spectral data into images; and 3) processing them using a corner detection algorithm. By identifying strong corners that correspond to significant spectral wavelengths, which has enabled accurate pollutant-type classification, additionally, spectral information combined with National Institute of Standards and Technology (NIST) database is used to assess pollution severity. A multitask learning vector quantization (LVQ) neural network model is employed to achieve simultaneous classification, which significantly improves accuracy from 90.2% to 98.89% compared to full raw LIBS spectral data, indicating reduced reliance on human expertise. This nondestructive assessment eliminates the need for insulator removal during operation, and the proposed technique has high accuracy and efficiency.