Convolutional Neural Networks for Defect Detection on LV cables

Abstract

A convolutional neural network (CNN) is a machine learning algorithm that is particularly well-suited for tasks such as object recognition, image captioning, and speech recognition. CNNs are particularly effective at detecting features in images that are not easily observable by other machine learning algorithms, such as defects in manufacturing. By analyzing large collections of images, CNNs are able to find patterns that are indicative of defects. Power cables are an important part of manufacturing, as they allow machines to be operated and communicated with. Recently, great importance has been offered in making electricity generation, transmissions, distribution and storage smart. However, the shift to smart grid should include intelligent methods of detecting the reliability of electrical connections. Several types of electrical cables are used to transmission and distribution of electrical energy. Due to excellent electrical and mechanical properties, cross-linked polyethylene (XLPE) cables are widely used in power systems. Poor manufacturing techniques in the production and installation of cable joints will cause insulation defects. Some suggest, the use of interdigital capacitive (IDC) for online monitoring on XLPE cables. Others suggest The use of a continuous wave (CW) terahertz (THz) imaging technology could help display and detect interior faults in cross-linked polyethylene (XLPE) plates used for power line insulation. In this paper, I developed models which predominantly use locally collected custom dataset to forecast individual power cable physical safety status. The model is aimed at replacing the physical inspection with computer vision and image processing techniques to classify defective power cable from non-defective ones. The project is implemented using the Python programming language, the Tensorflow library, and a Convolutional neural network.The Convolutional Neural Network (CNN)-based method is purposefully chosen and applied in this project for power cable defect classification. The project culminates by recommending the use of same or additional datasets and provide modalities to detect power cable defect from live video.