Modelling reactive distillation

5218 R. Taylor, R. Krishna / Chemical Engineering Science 55 (2000) 5183}5229di!erence between an arbitrary composition and a singularpoint. The use of these di!erence points in RD processdesign is the subject of two recent papers by Hauan,Ciric, Westerberg and Lien (2000a) and by Lee, Hauan,Lien and Westerberg (2000c).Pistikopolous and co-workers (see Papalexandri &Pistikopolous, 1996; Ismail, Pistikopolous &Papalexandri, 1999) have developed a phenomena-basedmodelling framework that is able to capture hybrid reactive/separationprocesses. In their approach an RD process(here meaning more than just a single RD column) issynthesised from a collection of modules that involvereaction, reaction and separation, or only separation. Inthis context a module does not necessarily representa single model stage, but a collection of such stages.Either EQ or NEQ stage models could be used in thisapproach. Papalexandri and Pistikopolous (1996) illustratetheir methodology with the design of an ethyleneglycol column, while Ismail et al. (1999) considered theproduction of ethyl acetate from acetic acid and ethanol.Bessling, Schembecker and Simmrock (1997a,b) combineheuristic rules, numerical computation, and RD linediagrams to study the feasibility of RD processes. Theydevise a method for identifying RD processes in whichthe RD column needs to be divided into reactive andnon-reactive sections. Bessling, Loning, Ohligschlager,Schembecker and Sundmacher (1998) use these RD linediagrams to identify the main process parameters and toprovide a basis for an experimental study in the methylacetate process. The signi"cant in#uence of the re#uxratio on conversion is shown by simulation using an EQstage model and by experiments in a small-scale columnthat employed packing in the form of Raschig rings.Sera"mov and co-workers (Balashov & Sera"mov,1980; Balashov, Patlasov & Sera"mov, 1981; Pisarenko& Sera"mov, 1988, 1992; Pisarenko et al., 1988a, b;Pisarenko, Danilov & Sera"mov, 1995, 1996, 1997;Giessler et al., 1998, 1999; Sera"mov, Pisarenko& Kulov, 1999b; Sera"mov et al., 1999a) have developeda method they call static analysis. It is the result of anextensive investigation of the thermodynamic andtopological structure of distillation diagrams. Themethod requires very little fundamental data on the systemand is based on an assumption that the vapor andliquid #ow rates in the column are very large. As a resultof this assumption the composition change caused by thereaction is small on each stage and can be neglected. Thedesired product compositions, extent of reaction, andnumber of stages need not be "xed a priori. The mostaccessible work from this group is the paper by Giessleret al. (1998) who outline the steps of the method and illustrateit with a case study of the MTBE process. Giessleret al. (1999) applied the method to "ve other RD processes:the production of acetic acid, cumene, ethylene glycol,ethyl or methyl acetate, and butyl acetate. The in#uenceof inerts can also be considered with this approach.6.2. Graphical design methodsA graphical (Ponchon}Savarit type) design procedurefor RD columns based on the transformed compositionvariables of Barbosa & Doherty (1988) is described byEspinosa, Scenna and Perez (1993). The MTBE process isused to illustrate their method.Lee, Hauan, Lien and Westerberg (2000a,b) and Leeet al. (2000d) have recently presented an in-depth analysisof the classical Ponchon-Savarit and McCabe}Thielegraphical methods for RD systems.6.3. Design via optimisation methodsRD design via optimisation methods has been thesubject of a few studies. Ciric and Gu (1994) formulateda mixed integer nonlinear programming model, solution ofwhich yields the optimal number of EQ stages, feed ratesand re#ux ratios at a minimal annual cost. To the best ofour knowledge this was the "rst paper that combines RDcolumn design and cost estimation, thereby emphasizingthe fact that the usual optimisation techniques for ordinarydistillation processes may be inappropriate for RD.Gumus and Ciric (1997) used a bilevel optimisation procedureto explore the design of RD columns that couldhave more than one liquid-phase. The number of phasesand the phase equilibria were determined by minimisingthe Gibbs free energy. Their paper was illustrated byconsidering the liquid-phase hydrogenation of nitrobenzeneto aniline, in which the resulting aniline}nitrobenzene}watersystem can split into two liquid-phases.Eldarsi and Douglas (1998b) used the optimisationcapability in Aspen Plus to optimise an MTBE column.Their study indicated that the market price of the MTBEproduct, the cost of the butylenes, the isobutylene compositionand utility costs all in#uence the MTBE columnpro"t. Pekkanen (1995b) described a local optimisationmethod for the design of RD. Duprat, Gassend & Gau(1988) constructed a set of empirical formulae that wasused to optimise an RD process for the separation of theclose-boiling mixture of 3-picoline and 4-picoline. Thelatter method leads to the most rapid optimisation calculationsonce the empirical formulae have been constructed(by "tting the results of rigorous model simulations).However, the di!erences between RD processes meansthat the formulae developed for one process have novalue for modelling others.6.4. From conceptual design to column designUntil recently, a procedure to go from a viable EQstage design to an actual column design was not availablein the literature. An important paper by Subawalla andFair (1999) addresses this issue with a lengthy discussionof the importance of several key design parameters: pressure,reactive zone location, reactant feed ratio, feed