Application of proper orthogonal decomposition to reduce detailed cfd models of glass furnaces and forehearths

The steady state and dynamic behaviour (heat transfer, temperatures, glass and gas flows) in glass furnaces and forehearths can be described accurately and reliably by computational fluid dynamics (CFD) models such as the TNO Glass Tank Model (GTM X). CFD models are based on the numerical solution of the partial differential equations for conservation of mass, momentum, energy and electric neutrality. Application of these detailed, but also slow models for direct on-line control or optimisation of glass melting processes (controlling fuel input, batch charging, batch composition, pressure, etc) is not possible without strong model reduction. A generic approach, so-called proper orthogonal decomposition (POD), which is able to reduce the complex CFD glass furnace simulation model to no more than approximately 50 equations, while maintaining the required accuracy and level of detail, is presented. The computational speed of the reduced order model is increased drastically to up to 50–1000 times faster than real-time. By following this approach, the resulting reduced models have become so fast, that they can directly be applied in Model based Predictive Control (MPC). The results of different applications based on this technique for the control of input parameters and process performance of glass furnaces and forehearths are shown. The benefits of this type of MPC control systems based upon 3D detailed CFD models will be discussed.