The CaO-MgO-SiO2-Al2O3 system, characterised by its intricate phases and thermodynamic
properties, plays a pivotal role in steel secondary refining processes, encompassing desulfurisation,
non-metallic inclusion capture, and refractory protection. Accurate predictions for diverse industrial
applications, including metallurgy, ceramics, and materials science, are imperative. To address this
challenge, a combination of machine learning techniques will be specifically applied to model the
liquid fraction of the slag and the solid fraction of MgO. The development of an artificial intelligence
(AI) system, leveraging various machine learning techniques, has gained momentum in this project.
The focus of this work is on constructing an AI model, based on machine learning techniques, within
the CaO-MgO-SiO2-Al2O3 system, utilising simulation results from FactSage™, version 8.1 (by GTT
Technologies). The primary objective is to train the AI model using these simulation outputs to predict
the percentage of liquid fraction and MgO saturation based on chemical composition parameters.
The AI model will undergo training with a comprehensive data set of simulations within the CaOMgO-
SiO2-Al2O3 system, covering a diverse range of compositional at 1873 K. These simulations,
conducted through FactSage™ 8.1 software, provide a robust foundation for AI model training,
ensuring generalisability and precise predictions for the liquid fraction of the slag and the solid
fraction of MgO, the solid fraction of MgO in this case is determined by the difference between the
total MgO and the MgO in the liquid fraction, so it is not the objective of this study to determine which
phase of MgO is in the solid state.. The predictive capabilities of this AI model hold significant
implications for process optimisation, quality control, and decision-making in CaO-MgO-SiO2-Al2O3-
dependent industries. Precise estimations of the liquid fraction and MgO saturation empower
researchers and engineers to enhance operational efficiency and quality. This paper explores the
methodologies employed for AI model creation and training, achieved results in terms of prediction
accuracy, and potential applications in the field. The development of this AI system signifies a
notable advancement in utilising machine learning for better comprehension and control of complex
chemical systems. Furthermore, to align the study with real-world steel production, we introduce FeO
and MnO at concentrations of 2 per cent and 1 per cent at 1873 K, respectively, following the
validation of model results using the CaO-MgO-SiO2-Al2O3 system. This adjustment aims to bring
the study closer to the observed reality in steel mills globally.