Decomposed algorithmic synthesis of reactor-separation-recycle systems

Abstract

More robust methods are needed in the synthesis of reaction-separation-recycle systems. Previously reported approaches using mathematical programming have attempted to solve the reactor-separator-recycle synthesis problem simultaneously, modeling it as a complex MINLP problem (e.g. Floudas, 1990). We propose to take advantage of the inherent hierarchical structure of this class of problems: When the reactor conversion and the recycle purge fraction are fixed, it becomes substantially easier to synthesize the separation system. The reactor synthesis problem, the separation problem and the recycle problem may each be posed as Mixed Integer Programming problems, but each of these will be smaller and less complex than a corresponding formulation for the entire system; — this is illustrated in figure 1. Taking advantage of the fact that the two most important decision variables in the system — conversion and purge fraction — are continuous variables, we do therefore advocate the use of a two-level optimization approach where a response surface method (Simplex) is used as the upper coordinating level and the actual structure of the individual subsystems are determined on the lower level. This algorithmic procedure we have called REREDOS — REactor-RE-cycle-Distillation Optimization System. To illustrate the approach in terms of the onion diagram described by Smith and Linnhoff (1988), the layers of the onion will have to be redefined as shown in fig. 2. In the inner layer of the onion the recycle rate and the total conversion are chosen. These basic variables will dictate the synthesis of the recycle structure, and thereafter the reactor network and the heat integrated separation system. After having synthesized the full system, we may go back to the inner layer of the onion and make new guesses for the basic variables. The guess of the new basic variables arc based on the cost of the total system for the previously chosen basic variables. The procedure will terminate when the basic variables are sufficiently close to an optimal point. The proposed approach will be demonstrated by applying it to the synthesis of the HDA process ( Douglas, 1988).