An Evaluation Method for Partial Discharge in Generator Stator Bar Insulation Based on Fiber-Optic Acoustic Detection

Area of Science:

• Electrical Engineering
• Materials Science
• Acoustics

Background:

• Partial-discharge (PD) monitoring is crucial for generator insulation health.
• Conventional methods face electromagnetic interference and deployment challenges in confined spaces.
• Fiber-optic acoustic detection provides electromagnetic immunity but lacks direct discharge magnitude information.

Purpose of the Study:

• To develop and validate a fiber-optic acoustic detection system for PD monitoring in generator stator bars.
• To create a hybrid neural network model for estimating PD magnitude from acoustic signals.
• To assess the system's performance against conventional methods and simpler models.

Main Methods:

• Developed a mandrel-type fiber-optic acoustic sensor for PD acoustic signal acquisition.
• Conducted PD tests on full-scale generator stator bars with internal defects.
• Constructed a hybrid Transformer-CNN-LSTM neural network for time-series modeling of acoustic signals and discharge magnitude.

Main Results:

• Fiber-optic acoustic detection demonstrated sensitive and stable monitoring of weak PD signals.
• Phase-resolved PD (PRPD) patterns correlated with internal defect discharge characteristics.
• The hybrid Transformer-CNN-LSTM model achieved 96.6% overall accuracy in discharge magnitude interval estimation.
• Specific high accuracies of 100% and 99.4% were achieved for low (<100 pC) and high (>2000 pC) discharge intervals, respectively.

Conclusions:

• Fiber-optic acoustic detection is a viable technology for PD monitoring in challenging environments.
• The hybrid Transformer-CNN-LSTM model effectively maps PD acoustic signals to discharge magnitude intervals.
• This approach offers a significant advancement in assessing generator insulation condition with high accuracy.