Classification of Low Frequency Signals Emitted by Power Transformers Using Sensors and Machine Learning Methods

Area of Science:

• Electrical Engineering
• Acoustics
• Machine Learning

Background:

• Power transformers generate low-frequency noise during operation.
• Accurate monitoring and classification of transformer noise are crucial for maintenance and diagnostics.
• Existing methods may lack automation and precision in noise analysis.

Purpose of the Study:

• To develop an automated method for detecting and classifying low-frequency noise from power transformers.
• To optimize signal processing parameters for machine learning algorithms.
• To evaluate the effectiveness of the proposed method across different transformer types.

Main Methods:

• Utilizing sensors to capture sound pressure levels from operating transformers in real environments.
• Applying frequency spectra analysis of the captured sound data.
• Automatically optimizing frequency spectra interval and resolution for selected machine learning algorithms.
• Testing various machine learning algorithms and optimization techniques on different transformer types (indoor and overhead).

Main Results:

• The proposed method achieves over 97% accuracy in detecting and classifying transformer types based on low-frequency noise.
• A preprocessing stage enhanced the method's accuracy by 10%.
• Specific machine learning algorithms were identified as robust solutions for accurate noise classification.

Conclusions:

• Automated detection and classification of transformer low-frequency noise is feasible with high accuracy.
• The developed method enables efficient, non-invasive inspections of working transformers.
• Optimized signal processing and machine learning are key to reliable transformer noise analysis.