OPTIMAL SYNTHESIS OF FLEXIBLE HEAT-INTEGRATED DISTILLATION TRAINS

This paper presents an algorithmic approach to the synthesis of a heat-integrated distillation train that accounts for prespecified uncertainties in the composition of the feed stream. It is intended to search for flexible distillation structures at which operability and least utility cost (LUC) are both guaranteed throughout the uncertainty range. In addition to reach such goals through a minimum number of heat transfer units, the sought design involving columns making only sharp separations should also minimize an estimate of the average LUC. This comes from the fact that the least utility cost can change significantly with the feed composition and, consequently, the mean energy efficiency of the distillation train becomes an important issue. To circumvent the problem's inherent complexities, an equivalent one defined in terms of maximum/minimum column heat loads is really tackled. Described through an MILP formulation, the equivalent synthesis problem only requires accounting for a couple of feed compositions providing the highest and the lowest heat duties at each column. Such feed compositions are established through either a low-size NLP or simple rules-of-thumb. Built-in feasibility constraints assure operability of the train everywhere. In the equivalent problem formulation, the original design objectives are all achieved by sequentially reaching the following targets: (i) a minimum global bound on the least utility cost; (ii) a minimum bound on the utility cost at every pinch temperature; and (iii) a minimum number of units. Two examples involving the separation of four-component mixtures into pure chemical species have been successfully solved in a reasonable CPU time on a VAX 11/780. In each case, several near-optimal designs have also been discovered.