Real-time implementation of new nonlinear neural adaptive generalized predictive speed control for a hot-rolling mill

The real-time application of a new design methodology for an efficient implementation of Adaptive Fuzzy Generalized Predictive Control (AFGPC) using a Radial Basis Function (RBF) based neural-fuzzy model for an experimental hot-rolling mill is presented in this paper. An optimization approach with the Gradient Decent Projection technique is proposed to calculate the predictions of the control actions. AFGPC has been implemented on a simulation platform and validated in real time to provide the mill with good speed control and regulation when steel or aluminium hot-rolling experiments are carried out. From such real time experiments and numerical simulations, it can be concluded that the proposed control scheme performs very well, showing good robustness and disturbance rejection under setpoint and load changes. These successful results will form the basis for future experiments to realise 'right first time' production of metals.