Thermodynamic analysis of novel ammonia addition to raw iron making followed by neural network development prediction

This study aimed to simulate the thermodynamics of ammonia addition to iron oxide reduction and to use the results of the thermodynamic analysis to develop a neural network that predicts iron oxide reduction performance from input parameters. The results showed that ammonia addition can increase iron (Fe) reduction when carbon and air are proportionally subtracted. However, iron oxide reduction assisted by ammonia is more endothermic, indicating ammonia may have insufficient exothermic activity to maintain the endothermic reduction of iron oxides. Additionally, the simulation suggests ammonia addition can limit the carbon dioxide (CO₂) produced by conventional iron oxide reduction using coke. Further, ammonia addition can reduce oxides of nitrogen (NO_X) and nitrous oxide (N₂O) emissions, depending on the addition method. A neural network was developed using one hidden layer with 10 neurons to predict the system enthalpy, the extent of iron oxide reduction, carbon dioxide production, NO_X concentration, and CO concentration from the temperature, ammonia addition percent, and method of ammonia addition. After training for 43 epochs, the neural network achieved an R² value of 0.9985. Although there are some limitations to the Gibbs minimization method employed in this study, the results indicate ammonia has the potential to supplement coke in ironmaking. More research is required to determine the extent ammonia could be employed to reduce greenhouse gas emissions from ironmaking processes.