In the quest to maintain the reliability of power grids, researchers have developed a novel approach to detect insulation defects in large oil-immersed transformers, a critical component of electrical infrastructure. The challenge lies in the closed structure of these transformers, which makes visual inspection difficult. To overcome this, a team led by Hongxin Ji from the School of Electrical Engineering at China University of Mining and Technology has introduced a self-developed micro-robot equipped with advanced image processing capabilities.
The key innovation lies in a new network called CDEP-CTSeg, designed to accurately and quickly identify carbon traces—insulation defects that can compromise transformer performance. “The defects we’re dealing with have varying shapes, sizes, and complex backgrounds,” explains Ji. “Accurate detection is crucial for the micro-robot to complete internal inspections effectively.”
The CDEP-CTSeg network integrates two key modules: the C2f-DySnake module and the Efficient Pyramid module. The C2f-DySnake module enhances the detection of complex carbon traces, such as branched or elongated structures, by focusing on local details. Meanwhile, the Efficient Pyramid module addresses the challenge of detecting small carbon traces, improving the model’s ability to capture relevant features and reduce missed detections.
To further enhance image quality, the researchers developed an improved Retinex image enhancement algorithm. This algorithm addresses issues like overexposure and color distortion caused by variable imaging distances and fluctuating lighting conditions, ensuring clearer and more detailed images for analysis.
The results are promising. Compared to the traditional YOLOv8-seg model, the CDEP-CTSeg network showed significant improvements in recall rate, precision, and mean Average Precision (mAP50), with increases of 2.3%, 2.8%, and 2.8% respectively. Notably, the network also improved the Average Precision of clustered and dendritic carbon traces by 1.9% and 3.7% respectively.
The implications for the energy sector are substantial. Accurate and efficient detection of insulation defects can lead to more proactive maintenance, reducing downtime and preventing costly failures. “This technology has the potential to revolutionize how we inspect and maintain oil-immersed transformers,” says Ji. “It’s a significant step towards ensuring the stable operation of our power grids.”
Published in the journal *Nature Scientific Reports*, this research opens new avenues for advancing transformer inspection technologies. As the energy sector continues to evolve, such innovations will be crucial in maintaining the reliability and efficiency of electrical infrastructure. The work by Ji and his team not only addresses current challenges but also paves the way for future developments in the field.