Use of mask R-CNN for detection and control of slag scraper wear

Abstract

The steel industry presents several problems and opportunities for improvement, from the factory floor to the business management level. Operational procedures are continually improved to reduce failures, create reliable parameters and increase the reliability of the equipment. A highlight is the computer vision, presented in several processes, contributing to the continuous and accelerated advancements of innovations in industrial processes, allowing systems' automation or upgrade and changing their way of operation. This project aims to segment and detect, through convolutional neural networks, the wear of the shovels of the slag scrapers in pig iron pans in a Kambara Reactor of an industrial steel plant. In other words, the goal is to detect the wear of the shovels to control their use and replacement using mask R-CNN (Regionbased Convolutional Neural Network), for instance, segmentation and pixel count for wear control and change forecast.