Automated classification of electrical network high-voltage tower insulator cleanliness using deep neural networks

Abstract

String insulators are components in high-voltage towers responsible for preventing energy dissipation through the tower structure; that is, they are responsible for isolating the high voltage in the electrical network cables. These string insulators must be clean for best performance and to avoid malfunctions. Verifying the necessity for cleaning/washing is most often performed by human visual observation, which can lead to interpretation errors, in addition to bringing risks to the physical integrity of humans in the vicinity of these electrical systems. Thus, this paper aims to develop an algorithm to detect and classify these insulators. The proposed algorithm uses artificial intelligence techniques and analyzes the image, inferring the state of cleanliness of the analyzed insulator. For the development of this algorithm, it was necessary to build a synthetic database using CAD software such as Inventor and Unity-3D due to image limitations available from dirty insulator strings. In this paper, two distinct neural networks are built using supervised learning techniques, where the first one is for detecting the chain of insulators, and the second is for detecting the type of dirt on the disk surface. In the first stage, techniques that use supervised learning are studied, more aimed explicitly at semantic segmentation networks, and in the second stage, classification deep neural networks were used to detect the type of impurities. In detecting insulator strings, an average dice coefficient of 0.95 was achieved for simulated images and 0.92 for natural images, with learning parameters based on a database with only simulated images. The average accuracy obtained in the dirt classification stage was 0.98.