Phase-Resolved Partial Discharge (PRPD) Pattern Recognition Using Image Processing Template Matching

Area of Science:

• Electrical Engineering
• Materials Science
• Signal Processing

Background:

• Partial Discharge (PD) analysis is crucial for diagnosing electrical equipment health.
• Traditional PRPD pattern analysis often requires manual interpretation, which is time-consuming and subjective.
• Deep learning methods for PD pattern recognition demand significant computational resources and large datasets, limiting practical application.

Purpose of the Study:

• To develop an automated method for recognizing and processing Phase-Resolved Partial Discharge (PRPD) patterns.
• To identify specific defect types in electrical equipment without human intervention.
• To provide a computationally efficient alternative to deep learning for on-line PD monitoring.

Main Methods:

• A novel Cosine Cluster Net (CCNet) model was developed as an image processing pipeline.
• The CCNet model extracts and processes patterns from 2D PRPD plots.
• Cosine similarity is used to compare extracted patterns against predefined templates of known defect types.

Main Results:

• The CCNet model successfully identified defect types from manually classified PRPD images in existing literature.
• The model consistently produced similarity scores aligning with manual classifications.
• Identification speed was rapid, typically not exceeding four seconds per analysis.

Conclusions:

• The CCNet model offers an effective and efficient approach for automated PRPD pattern recognition.
• The method's speed and accuracy show strong potential for real-time on-line partial discharge monitoring applications.
• This technique bypasses the need for complex deep learning algorithms, making it more accessible.