Global optimization in the estimation, control, and synthesis of chemical processes (Parameter estimation)

In the movement of a process from the laboratory scale to the production scale, two very important questions need to be answered. First, how can the process be accurately modeled from the available experimental data? Secondly, what is the best design, operational conditions, and control method? More often now, with the advent of faster and more economical computational technologies, these questions are being answered using mathematical models and optimization methods. Chemical processes are inherently nonlinear and often described by dynamic equations. These complications make the solution of optimization problems containing these models difficult from both a numerical and theoretical perspective. One of the major problems with these formulations is their nonconvex nature which results in the existence of multiple local minimum in the area of interest. Current optimization techniques for the solution of these formulations attempt to find the best or global minimum but cannot guarantee its determination in a reasonable time.; In this thesis, a deterministic global optimization approach will be presented for the solution of these types of problems. The approach is based on a branch-and-bound framework in which a valid upper bound on the global minimum is determined by solving the original problem to a local optimality. A lower bound is determined by solving a valid convex underestimation to local optimality. ε-convergence to the global minimum is obtained by successively subdividing the region at each level of the branch-and-bound tree. The key to the success of this method is the ability to determine a valid convex tight underestimation of the original formulation. Particular attention in the thesis will be paid to the development of a convex underestimator for the dynamic portion of the problem formulation.; The performance of the approach will be illustrated through its application to series of different problems. These range from the determination VLE parameters to the synthesis of chemical reactor networks. The formulation structure for each class of problems studied will be closely examined and exploited in order to provide a computationally efficient implementation of approach.