A Generic Approach to the Development of Semiempirical Predictive Models for Bath-Type Furnaces

"The modelling of bath type furnaces tends to be either fundamental in nature or total empirical ""black box models"". While models of the first kind are robust, they tend to lack accuracy, in cases where they are to be used for process control, and may take an overly long in development time. On the other hand, black box models tend to lack robustness of prediction for all circumstances and tend to be limited to interpolate based on historical data. While this may very often provide short-term accuracy, the models may have problems with stability in the long term. Moreover, the model structure may not reflect any of the fundamental properties of the system and remains opaque to fundamental analysis. The aim of the model development is in obtaining a model that have robustness, high accuracy, a short development time and which is adaptive to a changing environment. The models are specifically developed for on-line decision support or control of furnaces. The modelling approach is evaluated for a chromite bath smelting furnace. The role of systems analysis, mass balance smoothing, thermodynamic analysis are integrated into one generic algorithm for bath type furnaces. Not only are fundamental process relationships such the mass and energy balances satisfied through a scientific method of data reconciliation; the proposed modelling approach allows adaptive prediction, the first step towards predictive control or metallurgical decision support.It is shown that neither an ARMAX model only, nor a purely equilibrium model only, could predict the dynamics of the process very well. However, a combination of the two approaches more than doubled the accuracy as measured by the R2. It was noted for carbon dissolved in the alloy, that the improvement in R2 in using a hybrid modelling approach was more than the sum of the R2 of the individual traditional approaches. Similar improvements in modelling performance were noted for all other species.It shows further that process dynamics may totally overshadow the effect of thermochemical equilibrium (even when the thermochemical equilibrium prediction includes the recycled alloy and slag heels as additional feeds). The exclusion of process dynamics therefore leads to erroneous results."