Metamodeling of the Electrical Conditions in Submerged Arc Furnaces

Physic–based Finite Element Methods (FEM) models of submerged arc furnaces (SAF) are capable to accurately estimate the induced currents in the steel shell, the alternating current distributions in the material layers, and the active and reactive power densities within the furnace. However, a physics–based model is generally very demanding in terms of computational time and resources, and therefore difficult to employ during control operations and in fast prototyping. In this paper we analyze to which level one can obviate these inconveniences through metamodels that are grounded on physics-based modelling. We find that with properly tuned metamodels it is possible to establish: a) forward relationships, to effectively show the relative importance of each input parameters for a given output, b) inverse relationships, to infer the input parameters given a set of outputs. We moreover show how to construct metamodels of the SAF electrical conditions that retain the same generalization capabilities as the original model, while being computationally lightweight.