Model predictive control of flat sheet processes: robustness and auto-tuning

点击翻译按钮获取中文摘要

In papermaking, a flat sheet process is a large-scale two-dimensional system with spatial and dynamic components; they must be continuously controlled and monitored to ensure product quality. Spatial control, also known as cross directional (CD) control, is used to maintain uniform product quality in the direction perpendicular to the sheet travel. Existing industrial control technology is based on model predictive control methods, which require solving constrained quadratic optimization problems online for every sampling interval. With the increase in scale and size in these processes and the demand for high product quality, how to robustly solve the optimization problems in face of modeling and disturbance uncertainties and how to tune controller parameters for good closed-loop performance are critical; the challenges include high dimensions of input and output variables involved, active actuator constraints, and short control intervals. Our objectives are to develop, evaluate, and apply advanced methods for cross directional control of large flat sheet processes, addressing robustness and controller tuning in the model predictive control framework; in particular, adopting the model predictive approach, we will investigate existing solutions, develop new algorithms for the robust control design and controller auto-tuning, considering feasibility and closed-loop stability issues and exploring sparse matrix structures in cross directional models, and study their efficient implementation in Honeywell control systems. Honeywell is a main provider of papermaking equiments and associated control systems. Successful completion of this project would enhance Honeywell's control technology for papermaking, benefiting the Canadian pulp and paper industry in general. Also, strong industry-university collaboration would strengthen Honeywell Vancouver's employee retention and attraction programs.