Monitoring of the oxygen lance, in the steel fabrication process by LD conveter, using a Mask R-CNN deep learning network
Palavras-chave:
Linz-Donawitz converter, computer vision, convolutional neural networks, deep learning, Segmentation, Mask R-CNNResumo
A critical problem in the manufacture of steel by converting pig iron into a Linz-Donawitz (LD) con-
verter is the accumulation of skull, a mixture of slag and steel, in the body of the lance used for oxygen injection.
This accumulation of skull can cause serious problems, among them prevent the movement of the lance through
the flanges located in the converters. Currently, in the steel industry, monitoring is done manually, based on an
operator’s experience. Traditional computer vision methods are not effective due to the hostile environment of the
steelmaking process, due to its object detection algorithm. Problems such as light, smoke, and others, hamper the
task of image recognition. To overcome these deficiencies, this article proposes a method to monitor and mea-
sure, in real time, the thickness of the skull on the lance using deep learning, more specifically a Mask R-CNN
framework. Our method consists of installing a high resolution camera to monitor the lance in real time, sending
images to a computer, equipped with a Mask R-CNN framework already trained to identify and measure the skull
deposited on the lance. The results of the experiments performed show the feasibility of using the system to assist
the operator in monitoring of lance skull.
