Recognition of multiple partial discharge patterns by multi-class support vector machine using fractal image processing technique

• Published in: IET science, measurement & technology measurement & technology, 2018-11, Vol.12 (8), p.1031-1038
• Main Authors: Basharan, Vigneshwaran, Maria Siluvairaj, Willjuice Iruthayarajan, Ramasamy Velayutham, Maheswari
• Format: Article
• Language: English
• Subjects: box‐counting fractal image compression technique; condition monitoring; data compression; edge detection; edge detection methods; feature extraction; fractal feature extraction; fractal features; fractal image processing technique; fractals; high‐voltage power apparatus; HV laboratory; image coding; insulation properties; insulation quality; multiclass nonlinear support vector machine; multiple partial discharge patterns; partial discharge measurement; partial discharges; PD recognition; phase‐resolved PD patterns; power apparatus; power engineering computing; power industry; power system reliability; PRPD patterns; Research Article; support vector machines
• ISSN: 1751-8822; 1751-8830; 1751-8830
• DOI: 10.1049/iet-smt.2018.5020

Partial discharge (PD) measurement is an efficient method for condition monitoring of insulation in high-voltage (HV) power apparatus. Generally, phase-resolved PD (PRPD) patterns are commonly used to identify the PD sources. It is clearly recognised that there is a correlation between the PD patterns and the insulation quality. However, in the case of multiple PDs, the PRPD patterns partially overlapped in nature, which results in difficult to identify the types of partial discharges. In this proposed methodology, a combined algorithm of different edge detection methods with box-counting fractal image compression technique is used for fractal feature extraction. The extracted features used as the input vector for the classifiers for PD recognition. To evaluate the performance of the proposed methodology, artificially multiple PD sources are simulated in HV laboratory. The result of this proposed work shows better recognition for canny edge detected fractal features implemented with user define kernel multi-class nonlinear support vector machine which can be further used to assess the insulation properties for practical implementation in power industry.