Research metodology:
This research project aims to develop a predictive model for accurately estimating nitrogen content in molten metal produced in both Basic Oxygen Furnaces (BOFs) and Electric Arc Furnaces (EAFs). Real-world operational data from steelmaking plants will be used to build and refine the model.
Nitrogen content will be monitored at four key stages of the steelmaking process: after pig iron desulfurization, before tapping from the BOF, at the start of secondary metallurgy, and at its conclusion. Several hundred samples of both molten pig iron and steel will be collected, with a focus on obtaining samples from all four stages within a single heat to track nitrogen evolution. Nitrogen analysis will be conducted using Thermal Conductivity Detection (ASTM E1019-18) at the certified laboratory, with a measurement range of 0.0005% to 0.50% nitrogen.
Numerous factors interact during the steelmaking process to collectively influence and determine the final nitrogen content of the steel.
Actual nitrogen measurements from molten pig iron and steel will be aligned with a comprehensive dataset of operational parameters from steelworks. This dataset will include detailed information on metal chemistry, temperature, material quantities, gas usage, and other relevant process variables.
Following rigorous data cleaning and alignment of metal samples with their corresponding process parameters, correlation analysis will be conducted to identify the key factors influencing nitrogen content in both molten pig iron and steel.
Advanced statistical techniques, including least squares regression, cointegration analysis, and econometric modeling, will be applied to the operational data to create an initial predictive model for nitrogen content in molten metal. Subsequently, cloud-based machine learning will be employed to enhance model accuracy through training with relevant datasets and validation using a separate test dataset.
The model's accuracy will be assessed using various statistical metrics. To evaluate its practical application in a steel plant environment, the model will be tested on real-world operational data. This evaluation will determine the model's generalizability beyond the specific conditions under which it was developed.