Convolutional Neural Network-Based Transformer Fault Diagnosis Using Vibration Signals.

Area of Science:

• Electrical Engineering
• Signal Processing
• Artificial Intelligence

Background:

• Transformer safety and cost-effectiveness rely on rapid and accurate fault diagnosis.
• Vibration analysis is a promising, low-cost method for transformer fault diagnosis, but faces challenges from complex operating conditions.
• Dry-type transformers require specialized diagnostic techniques due to their unique operating environments.

Purpose of the Study:

• To develop a novel deep-learning-enabled method for diagnosing faults in dry-type transformers using vibration signals.
• To enhance the accuracy and efficiency of transformer fault diagnosis through advanced signal processing and machine learning.
• To address the challenges posed by complex operating environments in transformer fault detection.

Main Methods:

• An experimental setup was created to simulate various transformer faults and collect vibration data.
• Continuous Wavelet Transform (CWT) was employed for feature extraction, converting vibration signals into time-frequency (RGB) images.
• An improved Convolutional Neural Network (CNN) model was developed and optimized for image recognition of transformer faults.

Main Results:

• The proposed CNN model achieved an outstanding overall accuracy of 99.95% in fault diagnosis.
• The deep learning method demonstrated superior performance compared to other machine learning techniques.
• Feature extraction using CWT effectively captured fault-specific information from vibration signals.

Conclusions:

• The developed deep-learning-enabled method provides a highly accurate and effective solution for dry-type transformer fault diagnosis.
• Vibration analysis combined with CWT and CNN offers a robust approach for intelligent transformer monitoring.
• This intelligent diagnosis system significantly improves transformer safety and operational reliability.