The design of staged distillation units where chemical reactions occur is performed using detailed and complex models. The values of the design variables are determined so that an economic criterion is optimised subject to the modelling equations of the process and a set of safety, operating and product quality specifications. The distillation column is divided into different sections to account for reactive and non-reactive parts of the column. The design variables include the number of stages

more »

... in the different column sections, the location of the feed streams, the column tray liquid phase holdup for each column section for homogeneous liquid phase reactions, or the catalyst load per tray for catalytic reactions, and the optimal operating conditions. An equilibrium stage model based on the MESH equations is utilised with kinetically controlled chemical reactions. The economic objective function consists of the annualised investment capital costs for the design and the operating costs for the column. The MESH equations for reactive distillation units are solved using orthogonal collocation on finite elements (OCFE) techniques. OCFE models assume that the modeling equations are satisfied exactly only at selected points within the column, namely the collocation points. OCFE models require that the position in the column in a staged distillation unit is treated as a continuous domain. This allows the number of stages in the column to vary as a continuous variable. Such a property will transform the integer valued decision variable that account for the number of stages in each column section in an equivalent continuous one. The composition and temperature profiles in the column are also expressed as continuous functions of position in the column. As a result the optimisation design problem using an OCFE model is solved using conventional nonlinear programming methods. In order to provide greater stability for the OCFE model, inert components that exhibit very low composition in certain column sections are discarded from the modeling equations. Subsequently, OCFE models allow the reduction of the size of the modeling equation set, by usually requiring fewer collocation points to describe a set of column stages while providing accurate prediction of the column optimum, given that a sufficient number of collocation points is used. The model reduction property of the OCFE formulation allows the use of complex steady-state models in the design procedure without significant increase in computational effort. A methodology is proposed that enhances the robustness of the designed process to uncertainties in the model parameters. This is based on the ability of the process to meet the operating and product specifications for perturbations in the model parameter set. The model parameters are perturbed along directions in the parameter space that cause the maximum variability in the decision variable space. The directions of maximum variability are determined from local sensitivity information by decomposing the sensitivity matrix, resulted from the solution of the optimal design problem at a nominal parameter point, into its eigenvector components (singular value decomposition). The optimal solution variation is determined by pathfollowing the parameterised first-order optimality conditions, while keeping the structural design variables (e.g. number of stages in the columns) at the optimal values from the nominal case, using continuation methods. Pathfollowing investigates whether process constraints are satisfied as parameters vary along the specified direction within a fixed range. Failure to maintain feasibility will initiate the creation of an augmented optimisation ii problem consisting of the nominal case and an additional scenario that accounts for the critical perturbation that led to violation of the process constraints. The procedure continues until the parametric perturbations for a specified magnitude along the direction of maximum variability for the system do not violate the process specifications. The proposed design procedure is applied successfully in the process design for the production of high purity ethyl-acetate from ethanol and acetic acid. The process involves a reactive distillation column with a recovery system. The optimal design is determined under the influence of uncertainties in the reaction kinetic parameters and the feed composition. Critical operating constraints are identified for the process and valuable insight regarding the effects of uncertainty in the feasible operation is acquired.