LightFD: Real-Time Fault Diagnosis with Edge Intelligence for Power Transformers.

Area of Science:

• Acoustic Signal Processing
• Artificial Intelligence in Power Systems
• Edge Computing for Industrial Monitoring

Background:

• Power fault monitoring using acoustic waves is crucial for industry.
• Current methods face challenges with real-time analysis due to transmission latency and limited edge device computing power.
• Effective feature extraction for acoustic signals on edge devices remains a significant hurdle.

Purpose of the Study:

• To develop a lightweight and fast fault diagnosis method for power transformers suitable for edge devices.
• To overcome the limitations of existing methods in terms of real-time response and on-device processing.
• To enhance the accuracy and efficiency of acoustic-based power fault detection.

Main Methods:

• Proposed a novel asymmetric Hamming-cosine window function to minimize signal spectrum leakage and ensure data integrity.
• Developed a multidimensional spatio-temporal feature extraction technique for acoustic signals.
• Designed a parallel, dual-layer, dual-channel lightweight neural network for on-device fault classification.

Main Results:

• The proposed LightFD method achieved a diagnostic precision of 94.64% and a recall of 95.33%.
• Demonstrated significant improvements over the traditional Support Vector Machine (SVM) method, with a 4% increase in precision and 1.6% in recall.
• Successfully validated the method's effectiveness through extensive simulations and experimental results on edge devices.

Conclusions:

• LightFD offers an effective solution for real-time power transformer fault diagnosis on edge devices with limited computational resources.
• The integrated approach of window function, feature extraction, and lightweight neural network provides high diagnostic performance.
• This method paves the way for more efficient and reliable industrial power fault monitoring systems.