Robust Tuning of Cross-Directional Model Predictive Controllers for Paper-Making Processes.

This paper studies automated tuning of cross-directional model predictive control for industrial paper-making processes under user-specified model parameter uncertainties. Automated parameter tuning algorithms are developed to reduce the variability of the actuator and measurement profiles in the spatial domain and to achieve satisfactory performance in terms of worst case settling times and worst case control signal overshoots in the temporal domain for given parametric uncertainties. Due to decoupling properties of the spatial and temporal frequency components, the controller design and parameter tuning can be realized separately. For the spatial design and parameter tuning in the presence of parametric uncertainties, the undesirable high-frequency components in the actuator profile are suppressed via an appropriate design of the weighting matrix $S_{b}$ using the real-valued Fourier matrix approach. For the temporal design, a temporal filter is adopted to smooth the reference trajectory, where the parameter in the temporal filter is carefully tuned to achieve a tradeoff between the worst case settling time and the worst case control signal overshoot. Finally, the effectiveness of the proposed tuning algorithms is verified using a system model extracted from the pulp and paper industry. [ABSTRACT FROM AUTHOR]