Procesamiento de imágenes usando inteligencia artificial para el desarrollo de aplicaciones industriales.

Abstract

Over the years, different devices have been implemented for the inspection of high voltage electrical networks in order to reduce costs in companies dedicated to the energy industry sector. Currently, the most used device for this inspection process is drones (Sampedro Pérez y cols., 2014). By flying over power lines, they store videos for a long period of time depen ding on the battery life of each drone. These long videos pose a problem as experts need to separate the most important video sections during the inspection process. This separation of videos allows the operator of the high voltage electrical networks to take hours of editing. This research focuses on the development and implementation of image processing algo rithms based on artificial intelligence to facilitate the inspection process of electrical networks, in order to automate processes related to the electrical energy industry. In the process of developing artificial intelligence algorithms, drone flights were carried out over high-voltage electrical infrastructure. 22,000 images were extracted from the captured videos for the training of artificial intelligence systems. Different architectures were trained for machine learning from artificial intelligence systems and the best machine learning archi tecture for their application was implemented. With the algorithms developed, a process of segmentation of videos by type of electrical infrastructure was carried out, showing operators more detailed information about the ins pection process of electrical towers. The video segmentation algorithm achieved the same performance as a human operator by detecting the types of electrical infrastructure. In addi tion to being able to extract images with a close-up view of the insulators found in electrical infrastructures, since electrical insulators are the most important elements in the process of inspecting electrical towers (Pernebayeva y cols., 2017). This research was carried out using the research internship modality at the Technological Center for Industrial Automation of the Pontifical Xavierian University