DEFECT DETECTION IN POWER DISTRIBUTION LINES USING MASK R-CNN

ABSTRACT

Distribution lines which are used in power transmission are prone to certain defects which might develop over time and use. Since overhead lines are exposed, they are also vulnerable to damage caused by outside forces. One of such defects occurs when a line stretches such that the gap between it and the adjacent line or the ground falls below acceptable standards. This is known as an Overhead Line Sag Defect. This leads to reduced ground clearance and the increased chance of conductors swinging into contact. Left unattended, such faults can cause problems including power outages and equipment damage so there is need for them to be quickly detected and attended to by maintenance personnel. This paper proposed software for detecting sagging defects on a distribution line using Mask RCNN computer vision algorithm. A total of 556 images on defective and non-defective conductors were captured from various locations from which a training dataset was constructed. The Mask R-CNN model was constructed using the Python library, PyTorch and trained on this dataset. Having obtained the trained model, software was written to make use of the model in classifying images of distribution lines and to display the results in a graphical interface. The proposed methodology, design process, expected results, project timeline and cost analysis are also discussed.