UCLA

Large-scale chemical process systems are characterized by highly nonlinear behavior and the coupling of physico-chemical phenomena occurring at disparate time scales. Examples include fluidized catalytic crackers, distillation columns, biochemical reactors as well as chemical process networks in which the individual processes evolve in a fast time-scale and the network dynamics evolve in a slow time-scale.

Traditionally, the design of advanced model-based control systems for chemical processes has followed the centralized paradigm in which one control system is used to compute the control actions of all manipulated inputs. While the centralized paradigm to model-based process control has been successful, when the number of the process state variables, manipulated inputs and measurements in a chemical plant becomes large - a common occurrence in modern plants -, the computational time needed for the solution of the centralized control problem may increase significantly and may impede the ability of centralized control systems (particularly when nonlinear constrained optimization-based control systems like model predictive control-MPC are used), to carry out real-time calculations within the limits set by process dynamics and operating conditions. One feasible alternative to overcome this problem is to utilize cooperative, distributed control architectures in which the manipulated inputs are computed by solving more than one control (optimization) problems in separate processors in a coordinated fashion.

Motivated by the above considerations, this dissertation presents rigorous, yet practical, methods for the design of distributed model predictive control systems for nonlinear and two-time-scale process networks. Beginning with a review of results on the subject, the first part of this dissertation presents the design of two, sequential and iterative, distributed MPC architectures via Lyapunov-based control techniques for general nonlinear process systems. Key practical issues like the feedback of asynchronous and delayed measurements as well as the utilization of cost functions that explicitly account for economic considerations are explicitly addressed in the formulation and design of the controllers and of their communication strategy. In the second part of the dissertation, we focus on the design of model predictive control systems for nonlinear two-times-scale process networks within the framework of singular perturbations. Both centralized and distributed MPC designs are presented. Throughout the thesis, the applicability, effectiveness and computational efficiency

of the control methods are evaluated via simulations using numerous, large-scale chemical process networks.