ABSTRACT The cold crushing strength (CCS) of iron ore pellets produced in a straight grate indurating machine is critical to their quality and performance. This study explored two modelling approaches to accurately predict CCS: an artificial neural network (ANN) model and a physics-informed hybrid model. The ANN model was designed to predict CCS directly from parameters such as pellet chemistry, Blaine number, and measured induration conditions. In contrast, the hybrid model combined a first-principles simulation of the induration process with the ANN model, using the simulated pellet temperatures and the exit gas temperatures as inputs to the ANN component, while keeping the other input parameters the same as the standalone ANN model. The models were validated using 1730 datasets from a 6 MTPA iron ore pelletizing plant of Tata Steel India Ltd. The hybrid model demonstrated superior performance, achieving a mean absolute percentage error (MAPE) of 3.7% compared to 4.1% for the ANN-only model. The hybrid model offers two key advantages: (1) it can provide early insights into CCS, enabling pre-emptive adjustments to the induration process, and (2) it can be used as a powerful optimization tool to enhance the overall quality and consistency of the iron ore pellets.
Read full abstract