Machine learning-based surrogate modelling of AOD nozzle using an integrated CFD and ANN approach

Abstract

This study introduces a novel machine learning-based surrogate model that integrates computational fluid dynamics (CFD) with artificial neural networks (ANN) to predict thermal-hydraulic behaviour in argon oxygen decarburization (AOD) nozzles. The objective is to address the limitations of traditional CFD models, which are computationally expensive and time-consuming, by developing an efficient, accurate alternative for nozzle outlet predictions. A two-dimensional CFD nozzle model was used to simulate fluid flow and heat transfer in AOD nozzles, considering variables like inlet pressure, nozzle length, and diameter. The CFD results were then employed to train an ANN, creating the CFD-ANN surrogate model. This model demonstrated high predictive accuracy (R-values > 99%), and can significantly reduce the time and resources required for simulations. The CFD-ANN surrogate model offers a practical solution for real-time decision-making, optimizing nozzle design and advancing sustainable operation in industrial applications.