LRF Optimization

For high-quality steel ladle refining furnace is necessary for temperature, deoxidization, desulphurization, and inclusion removal as well as for fine-tuning composition of molten steel grades such as bearing steel where fatigue life is greatly impacted by total oxygen concentration. However, the LRF process is complicated with strong coupling effects, non- linear correlations and changeable input conditions, making precise prediction difficult and control difficult. Traditional methods often result in low precision, increased material consumption, eg, ferroalloys and off-specification heats, necessitating extensive and expensive post-production testing. Oxygen ingress from sources like carryover slag (FeO+MnO) and argon stirring reoxidizes steel, consuming costly deoxidizers like aluminum and reducing their yield This study offers a data-driven strategy to maximize alloy additions in the process at the ladle refining furnace (LRF) stage, which are essential for regulating an ultimate chemical composition quality of steel with the objective of minimizing material cost while ensuring compliance with grade-specific chemical specifications. The study leverages historical plant data, comprising heat-wise opening and final chemistries, ferroalloy addition records, and cost-recovery profiles for grade steel. We explore and compare three mathematical optimization strategies: Linear Programming(LP), Bayesian Optimization (BO) using both Optuna and Scikit-Optimize, and Genetic Algorithms (GA) via the pymoo library This study emphasizes the difficulties in optimizing in actual steelmaking settings and suggests modeling enhancements to match algorithmic results with a metallurgical reality.The findings highlight the need of pre-validating data related to domain expertise, the necessity of hybrid modeling techniques, and the incorporation of physical process behavior with optimization logic.