Data Mining analysis of the relationship betwen input
variables and hot metal silicon in a blast furnace

Abstract: The purpose of this report is to get a better understanding of the
relationships between input variables and silicon's behaviour in BlueScope
Steel's Port Kembla No. 5 Blast Furnace. Over a period of 13 years
stretching from 1991 to 2004, potentially important variables affecting the
hot metal silicon content of the hot metal produced in the furnace were
collected on a daily basis. A process analysis was performed using the data
mining program, 'Weka', which is a freeware program developed by the
University of Waikato, New Zealand. The data collected were altered, and
different datasets compiled, in order to analyse the data without
interferences and to reveal previously unknown and hidden relationships. A
post-pulverized coal injection dataset was compiled to remove historical
variations of limited relevance to current operations. An important variable
affecting the hot metal silicon was found to be the amount of Quartz charged
in to the furnace. After finding the importance of Quartz charge, a short
term, high frequency dataset was collected and analysed to find the true
correlations. Quartz is charged in to the furnace to change the slag
basicity, but this project shows that a significant fraction of the Quartz
transfers into the hot metal as silicon instead of transferring in to the
slag as silica as intended. The original data was compiled into a day-to-day
difference dataset where each instance contained a difference between two
consecutive days. In the difference dataset analysis, the long term
variation was removed and the impacts of operational variables on hot metal
silicon were revealed. The most significant variable affecting hot metal
silicon was hot metal temperature. Hot metal silicon was corrected by hot
metal temperature and the analysis was continued. The analysis showed that
the relationship between hot metal sulphur and hot metal silicon was not
causal, and could be explained by the impact of hot metal temperature. A
data mining framework brought valuable new insight into hot metal silicon
control in the blast furnace and further effort in this area is strongly
recommended.