APPLICATION OF DEEP LEARNING FOR ANALYSIS OF CRACKS IN PELLET FALLING TESTS

Resumo

Iron ore pellets are a prime input for iron production. Therefore there is a need for a rigorous control of
the quality of the pellets to apply them in the industrial process. The pellets are degraded due to impacts caused
by their handling or transport systems. As a result of these degradations many pellet shipments reach the customer
with a proportion of cracks. Laboratory drop test trials are required on wet raw pellets to assess their resistance to
the various drops they suffer in the industrial process. Currently the drop test is performed manually, where the
whole test process, from pellet manipulation and data collection, depends on human action. The present work aims
at the application of Deep Learning to carry out the analysis of pellet cracks, pellet segmentation is initially
presented in this article. A network of light deep learning was designed, generating a data set of the pellet drop
test for training the network for pellet classification. This network will be applied in the autonomous prototype for
the drop test, a technological innovation that is being developed by the Research and Automation Group (GAIn),
of the Federal Institute of Espírito Santo, located in the Municipality of Serra, for the analysis of cracks in the
pellets.