Electrical insulation flaw detection in power equipment by infrared thermography with YOLO algorithm and an attention mechanism

Electrical insulation faults are common in power equipment while epoxy resins (ER) is a common insulation material used. However, detecting imperceptible flaws within the insulation pose a challenge. This paper presents a fault detection method by an infrared thermography technology with an objective detection algorithm under the YOLOv7 framework. Subsurface defects of ER with various insulation flaws have been tested by the infrared thermography so as to obtain the images for reasoning. This study modifies the YOLOv7 architecture by integrating the coordinate attention mechanism in the neck layer to identify the location and depth of a flaw. A key performance indicator, the mean average precision, is used to compare between different methods. The proposed method is proved to be 10.6% and 13% more accurate than general YOLOv7 and YOLOv5, respectively. Moreover, it is validated that the proposed approach has improvements in computational speed.