Optimisation-Based Design of Reactive Distillation Columns
Systematic methods and tools for managing the complexity
Process Synthesis & Design - II (T4-1b)
Keywords: optimisation-based conceptual design, reactive distillation
In a reactive distillation column a chemical reaction is coupled with the distillation of the reaction mixture. This offers the opportunity to overcome chemical and thermodynamic limitations, such as the chemical equilibrium or distillation boundaries. On the other hand the feasibility of such an integrated process strongly depends on the design of the column. In the conceptual design of reactive distillation columns, the operating parameters, the dimensions and the structure of the column are determined. Structural variables correspond to decisions on the number of column trays, the location of the reaction zone and the distribution of the feed streams. In principle, these decisions can be modelled via discrete variables. The basis of the evaluation of a design is its annualised cost. The total costs which have to be minimised are composed of operating costs and investment costs that depend nonlinearly on the design parameters. The conceptual design constitutes an optimisation problem with complex constraints due to chemical and thermodynamic equilibria and reaction kinetics and can be tackled by mixed-integer nonlinear programming techniques. In the literature, different approaches to the modelling of the existence of individual structural variables in the optimal solution have been proposed, e.g. the multiplication of constraints or structural variables by binary variables [1, 2] or disjunctive programming [3]. Besides the nonlinearities of the model, the existence of binary or integer variables causes significant difficulties in the computation of feasible and optimal solutions.
We discuss different approaches to the modelling of the structural decisions here. Besides a conventional approach, which assigns a binary decision variable to each column tray in a superstructure that describes its existence, an approach with a minimal number of binary variables is studied. Modelling feed streams on all trays by continuous variables considerably reduces computing time but leads to a large number of nonzero feed streams which is not realistic. In contrast, the distribution of the reaction zone can be appropriately described by continuous variables. The influence of the modelling on the solution effort and the quality of the solution is described. The optimisation problem is implemented in GAMS [4] and is solved by the commercial solvers SBB, CONOPT, and OQNLP with non-standard parameters. The scatter search algorithm implemented in OQNLP [5] generates a series of promising initial values from which the problem is solved, thus enhancing the probability of a good solution by the price of an increased solution time. The approaches are applied to the production of MTBE and of methyl acetate in a reactive distillation column.
[1] A. Ciric, D. Gu, Synthesis of Nonequilibrium Reactive Distillation Processes by MINLP Optimization, AIChE Journal 40 (1994), 1479-1487
[2] T. Frey, J. Stichlmair, MINLP Optimization of Reactive Distillation Columns, ESCAPE-10 (2000), 115-120
[3] J. Jackson, I. Grossmann, A disjunctive programming approach for the optimal design of reactive distillation columns, Computers and Chemical Engineering 25 (2001), 1661-1673
[4] A. Brooke, D. Kendrick, A. Meeraus, R. Raman, GAMS – A User’s Guide (1998)
[5] M. Laguna, R. Martí, Scatter Search – Methodology and Implementations in C, Kluwer Academic Publishers (2003)
Presented Tuesday 18, 16:40 to 17:00, in session Process Synthesis & Design - II (T4-1b).
