The blast furnace iron-making process is currently the dominant process for providing steelmaking raw materials worldwide due to its reliability and cost effectiveness. In a blast furnace, fuel, ores, and flux are continuously fed through the top of the furnace, while a hot blast of air is blown into the lower section of the furnace. The end products of blast furnace iron-making are molten metal (Pig Iron) and slag which are tapped from the bottom of the furnace. The molten metal tapped from the blast furnace is used as a raw material for producing Ductile Iron which is a type of graphite-rich cast iron which provides better ductility properties than cast iron, due to its nodular graphite inclusions.

Trace describes metallic or non-metallic elements that are not specified in the alloy grade, but an acceptable amount (trace) can be present without any detrimental effect on the alloy’s performance, however its presence above the acceptable amount can cause major problems in the alloy’s manufacturing. For Ductile Iron Pipe manufacturing, the following elements are classified as trace elements: Mo, V, Cu, Sn, Pb, Sb, B, Bi, Cr, Co, Nb, W, Ce, Ca and As.

These elements enter the Hot Metal from the raw materials charged into the Blast Furnace. Raw Materials charged into Blast Furnace are analyzed to determine the chemical composition of the burden and then fed into a mass balance model to estimate the chemistry of output Hot Metal. However, it is not possible to measure amount of trace elements entering into the blast furnace as the measuring instrument (XRF) does not have sensitive detection limits for most of the trace elements and requires costlier instrumentation such as ICP-OES/ICP-MS/CVAAS.

An alternative approach to understand the impact of input raw materials charged on the output Hot Metal trace element chemistry is suggested. The effect of different raw materials charged in Blast Furnace was studied by analyzing the historical data and a Machine Learning Model was developed to predict whether the Hot Metal will have high or low trace element percentages, so that preemptive control measures to combat high trace elements may be taken during the Ductile Iron Pipe manufacturing. This model uses different quantities of raw material charged in the Blast Furnace as input and provides high or low trace element as output. Data for six months of operation of Blast Furnace was considered for the study. Data was preprocessed to remove any data pertaining to abnormal operating conditions of the Blast Furnace. Several Binary Classification models were applied and the best model was selected based on performance on a validation data with 10-fold cross-validation scheme.

Decision Tree model was selected as the final model, due to its explainability and possibility of generating business rules for operators to follow. This model gave a Precision & Recall of around 82% on test data. With this model, it is possible to predict the possibility of high trace element in Hot Metal during burden preparation itself, leading to corrective action in downstream processes.

Menu

Mr. Barun Kumar