Deep Learning for Insulator and Binding Fault Diagnosis in Smart Grids With YOLOv11 Instance Segmentation

Smart grids heavily depend on insulators and their connecting components for reliability. Faulty components can lead to power outages, equipment damage, and safety hazards. This paper presents a design for a deep-learning-based system that employs the YOLOv11 instance segmentation technique to automatically detect and classify faults on insulators and their binders, central to smart grid reliability. Classical approaches to fault detection require manual recognition and basic image processing techniques, and thus they are tedious and unreliable. YOLOv11, on the other hand, offers better fault recognition accuracy, quicker processing, and segmentation. By using roboflow, a dataset of insulator images affected by faults such as cracks, pollution flashover, and binding failures are build and proved that YOLOv11 surpasses traditional and benchmarked methods by performing at a mAP of 84.1% and allowing for real-time diagnostics of faults under varying conditions. This allows a reactive maintenance approach, minimizing downtime and firing up grid reliability. Further, this research will involve making the model edge-computer friendly and enhancing its fault detection scope.