Intelligent diagnostic method for composite insulator surface temperature distribution realized by visible images

Area of Science:

• Electrical Engineering
• Materials Science
• Computer Vision

Background:

• Accurate diagnosis of thermal defects in composite insulators is crucial for transmission line safety.
• Infrared thermography is effective but costly for widespread power inspections.
• A cost-efficient, accurate method for thermal defect diagnosis is needed.

Purpose of the Study:

• To develop a deep learning-based method for diagnosing composite insulator surface temperature distribution using visible images.
• To overcome the limitations of traditional infrared thermography in terms of cost and application scope.

Main Methods:

• Utilized You Only Look Once version 7 (YOLOv7) for precise insulator localization and background cropping.
• Employed an improved U-Net architecture for accurate temperature distribution diagnosis and visualization.
• Fine-tuned the model with a physics-improved U-Net to enhance diagnostic accuracy.

Main Results:

• The model accurately diagnoses surface temperature distribution across various angles and brightness levels.
• Achieved a wide temperature diagnostic range (15-125°C) with high accuracy (MAE: 0.780°C, RMSE: 1.130°C).
• Real-time diagnosis capability with a single image processing time of 30 ms.

Conclusions:

• The proposed deep learning method enables real-time, accurate, and wide-range temperature diagnosis of composite insulators from visible images.
• This approach offers a cost-effective and efficient alternative to traditional infrared thermography for power line inspections.