Fingerprint recognition of partial discharge signals in deep learning enhanced Rydberg atomic sensors

Area of Science:

• Electrical Engineering
• Materials Science
• Sensor Technology
• Artificial Intelligence

Background:

• Partial discharge (PD) is a key indicator of insulation degradation in high-voltage equipment.
• Traditional PD detection methods struggle with limited bandwidth and require manual feature extraction, hindering the analysis of complex transient signals.
• Developing advanced, non-invasive diagnostic techniques for electrical insulation systems is crucial for preventing equipment failure.

Purpose of the Study:

• To develop a novel approach for capturing and analyzing partial discharge signals using Rydberg atomic sensors.
• To apply deep learning for automated recognition of different partial discharge types from time-domain data.
• To assess the robustness and effectiveness of this integrated sensing and AI approach for early-warning diagnostics.

Main Methods:

• Utilized a Rydberg atomic sensor for direct, broadband time-domain capture of partial discharge emissions.
• Constructed unique spectral fingerprints from the captured time-domain signals.
• Employed a 1D ResNet deep learning model for automated classification of partial discharge types without manual feature engineering.

Main Results:

• Achieved approximately 94% recognition accuracy across four partial discharge categories, demonstrating robustness against signal attenuation and noise at increased distances.
• Successfully validated the approach in a simulated early-warning scenario, with the model generating predictive alarms for noisy PD signals.
• The deep learning model effectively recognized spectral fingerprints from time-domain signals, bypassing the need for manual feature extraction.

Conclusions:

• The integration of Rydberg-based broadband sensing and deep learning offers a powerful tool for non-invasive diagnostics of electrical insulation.
• This method provides high sensitivity and robustness, outperforming conventional techniques in complex and noisy environments.
• The approach shows significant potential for real-time, early-warning systems for critical high-voltage apparatus.