ABSTRAKT

Model Predictive Control (MPC) algorithms offer great perspectives for the optimization of glass furnaces operation. They allow an optimal control of several variables, which offers far better possibilities than classical PID loops, rather limited for the glass producers current objectives. Computational Fluid Dynamics (CFD) models are far too time consuming to be used for real-time control. Faster predictive models of the glass furnace are needed, and until now, identified black-box models have been used. This paper shows our work to develop a fast model based on a first-principle modeling approach for the combustion chamber and on a reduction technique for the glass bath. In the combustion chamber, where high temperature turbulent flames take place and radiate to the enclosure walls and the load underneath, a simplified modeling of combustion coupled with a global enthalpy balance on the combustion gases is considered. A finite volume method is used to solve the flow equations. Gas radiation is treated through the zone method. A particular attention is given to the non-grey aspect of radiation of H₂O-CO₂ mixture. In the glass bath, coupled heat transfer with convection, conduction and radiation are taken into account in a Finite Element model, with the assumption in this feasibility study of a non-thermally dependant velocity field. The Rosseland's diffusion approximation is assumed to model radiation. Starting from simulations of this model, the Modal Identification Method is used to build a reduced model of the glass bath. It differs from a black-box model since the physics are retained in the reduced model structure.