Optimal Solvent Design for Batch Separation Based on Economic Performance Apostolos Giovanoglou, Julie Barlatier, Claire S. Adjiman, and Efstratios N. Pistikopoulos Centre for Process Systems Engineering, Dept. of Chemical Engineering and Chemical Technology, Imperial College London, South Kensington Campus, London SW7 2AZ, UK Joan L. Cordiner Global Specialist Technology, Syngenta, Office 2.2 T&P, Grangemouth Manufacturing Centre, Grangemouth, Stirlingshire, FK3 8XG, UK ( ) A mixed-integer dynamic optimization MIDO framework for sol®ent design in batch processes is presented. Performance measures reflecting process economics and com- puted on the basis of process dynamics are used to ®alidate candidate sol®ent structures built from the UNIFAC molecular groups. These define the discrete space in the o®erall material formulation and are combined according to molecular design feasibility rules to guarantee realistic molecular representations. The algorithm is based on the decom- position of the MIDO primal subproblem into se®eral steps that are sol®ed successi®ely. This allows unsuitable sol®ents to be detected and discarded quickly and without signifi - cant computational cost. Emphasis is placed on problem formulation in order to match accuracy of process model and physical property predictions. The algorithm is applied successfully to an industrial case study dealing with a three-phase dehydration column and a decantation unit for sol®ent reco®ery. The proposed algorithm can be regarded as a promising initial step toward an integrated and simultaneous methodology for mate- rial process design in batch separation systems. Introduction Solvent design has attracted significant interest over the last two decades, not only because of the important role of solvents in process operations, but also because of the need to find substitutes for previously used solvents due to envi- ronmental, safety, and health regulations. New solvent selec- tion criteria have been introduced, and more systematic and automated approaches to the problem have been or are be- Ž . ing developed Achenie et al., 2003 . Thanks to technological advances in computer systems and to extensive research on molecular structure-based physical property estimation meth- Ž . ods such as group contribution methods , it is now possible to design simple molecules and computer-aided molecular Ž . design techniques CAMD are being used by some industrial Correspondence concerning this article should be addressed to C. S. Adjiman. practitioners. These molecular design methods are based on the fact that from a small set of structural groups a large number of molecules can be generated and evaluated with respect to a certain performance index. Two types of performance indices can be identified. The first focuses on sol®ent performance, either by considering Ž . physical property targets as defined by Maranas 1996 , or by considering solvent-performance measures that are process- Ž dependent, such as selectivity or capacity Odele and Macchi- . etto, 1993 . Most of the work on solvent design to date falls in this category. The second type of index that can be consid- ered is a measure of process performance. In this approach priority is given to the overall solvent effect on the operation of the process, which allows task-level details and character- istics to be accounted for during the solvent selection proce- dure. It therefore leads to a tighter integration of process December 2003 Vol. 49, No. 12 AIChE Journal 3095