Glass production has a great industrial importance and is associated with many technological challenges. Control related problems concern especially the last part of the process, so called glass conditioning. Molten glass is gradually cooled down in a long ceramic channels called forehearths during glass conditioning. The glass temperature in each zone of the forehearth should be precisely adjusted according to the assumed profile. Due to cross-couplings and unmeasured disturbances, traditional control systems based on PID controllers, often do not ensure sufficient control quality. This problem is the main motivation for the research presented in the paper. A Model Predictive Control algorithm is proposed for the analysed process. It is assumed the dynamic model for each zone of the forehearth is identified on-line with the Modulating Functions Method. These continuous-time linear models are subsequently used for two purposes: for the predictive controller tuning, measurable disturbances compensation and for a static set point optimisation. Proposed approach was tested using Partial Differential Equation model to simulate two adjacent zones of the forehearth. The experimental results proved that it can be successfully applied for the aforementioned model.

The paper presents an iterative identification method dedicated for industrial processes. The method consists of two steps. In the first step, a MISO system is identified with the Modulating Functions Method to obtain sub-models with a common denominator. In the second step, the obtained subsystems are re-identified. This procedure enables to obtain the set of models with different denominators of the transfer functions. The algorithmwas used for on-line identification of a glass conditioning process. Identification window is divided into intervals, in which the models can be updated based on recent process data, with the use of the integral state observer. Results of the performed simulations for the identified models are compared with the historical process data.