Modelling of metallurgical reactors with Scilab

Lea Florentin, Eloy Crespo, Bruno Laboudigue
Eramet Research

Metallurgical reactors are characterized by their complexity, especially in terms of chemistry, heat transfer, transport phenomena, and more generally thermodynamics. Because of the large amount of parameters to consider, modelling seems the most suitable way of studying such a system.
Two of the main general purposes for modelling are (1) process understanding and (2) optimizing the performances of a reactor. The models, once created, can help in the decision making process and can contribute to avoid expensive experiments. Scilab provides an adequate flexible modelling framework to achieve such an objective. The studied reactor is a converter, blowing air into a 1200°C Ni-Fe-S melt for several hours. The aim of the model is to simulate the chemical evolution of the different species, phases and temperature in the reactor as a function of time. This work presents two different approaches of modelling this metallurgical reactor. The first approach uses a thermodynamic software as a “slave” component for computing thermodynamic equilibrium at each time step. A standard Scilab code implements the coupling of other transport phenomena. The model realizes automatically calculations on a thermodynamic database and can set a new calculation based on the previous results. It is also able to use the results to calculate other data or to save graphs and excel files. In addition to the use of a large thermodynamic database, Scilab answers the need for a dynamic model, implementing other transport equations using standard Scilab functionalities. The second approach is based on the construction of a meta-model taking into account the thermodynamic information. It is thus fully embedded in Scilab, making it possible to incorporate easily the thermodynamic data using the meta-model. Furthermore, this model is independent of the large thermodynamic database and can be more easily used for industrial process control.
Development of both models has been done in parallel, allowing cross-validation. Moreover, the two approaches are compatible, thus allowing to obtain, on one hand, a process simulator answering the need for process understanding, and, on the other hand, a compact and better constrained model, putting emphasis on the process control issue.