Application of a neural architecture to estimate the wear of down and up throats in RH degassers

Resumo

During the metallurgical process of steelmaking, conformity is achieved in the processes of primary and
secondary refining. One of the types of secondary refining takes place in the RH degasser, where the molten steel
from primary refining is fed into a ladle below RH. RH and the ladle are connected by two tubes called "up leg"
and "down leg". The steel is encouraged to circulate in the up leg, while the molten bath flows back into the ladle
through the down leg. The edges of the refractory bricks of both legs, also called "throats," are subject to wear.
This observation is made by an operator who goes to the top of the degassing ladle and uses a cell phone to take a
picture of the condition of the throats on the lower vessel of RH. The image is evaluated to check for throat wear
and to ensure the integrity of the process to avoid perforations and unavailability of RH. To provide a more
meaningful measurement, YOLACT, a neural architecture designed for real-time image segmentation, is used to
extract the coordinates describing the segmentation of the throats in the images, which are later processed to
estimate the actual wear.