Transmission Line Isolator Fault Detection Based on Deep Learning and UAV Imageries

Transmission line inspection is critical in a power transmission system to guarantee the protection and continuous, dependable operation of the power supply. The role played by the basic components of the electricity transmission system between the substation the and final consumer is crucial. The power line components are exposed to extreme weather conditions, like extreme temperature due to high voltages and they suffer high mechanical tension. They are therefore prone to physical defects and should be visually maintained periodically. The main conventional methods of inspection methods are a manual visual inspection by a trained artisan who visually looks for defects. The goal of this study is to look at the feasibility of merging optical and thermal images in drone inspection of transmission lines. Convolutional neural networks strategy of deep learning was selected after gathering sufficient data encompassing both optical and thermal images of substation isolators under different environments and implementing augmentation methods on the data. The model learns the separation between defective and non-defective substation isolators in experimental analysis using TensorFlow object detection API. A comparison was made between a custom-made convolutional neural model and the pre-trained VGG16 and ResNet50 models. The results reveal that the VGG16 pre-trained model accurately detected images with a maximum accuracy of 100%.