Optimization-based process synthesis makes use of optimization tools to identify the best configuration of a process flowsheet. For this, the definition of a super­structure that considers a significant amount of variations in the topology of the technological configurations for a given process is required. In general, the evalu­ation and definition of the best process flowsheet is carried out solving a mixed — integer nonlinear programming (MINLP) problem. In this way, process synthesis is accomplished in an automatic way excluding so far as possible the formulation of heuristic rules.

The great advantage of this approach lies in the generation of a generic frame­work to solve a large variety of process synthesis problems carrying out a very rigorous analysis of the global process structure. In particular, all the equations corresponding to the models of each process unit should be specified. This allows the definition of the accuracy level during the description of unit processes and the operations involved. This implies solving an optimization problem while simultaneously taking into consideration all the models of the units (equation- oriented approach). In contrast, most of the commercial process simulators some­times used for knowledge-based strategy are based on the modular-sequential approach in which the calculation scheme involves the models for each unit to which the user does not have direct access. Thus, the simulation is solved taking into account the strict order of the units (from feedstocks to end products) mak­ing up the process flowsheet. Main drawbacks of the optimization-based strategy are related to the fact that the optimal configuration can only be found within the alternatives considered in the formulated superstructure (Li and Kraslawski, 2004). Furthermore, this approach has the additional disadvantages of having a significant mathematical complexity as well as the difficulties arising during the definition of the superstructure of technological configurations, i. e., the difficulty to ensure that the initial superstructure contains the “best” solution (Barnicki and Siirola, 2004). In this sense only, it is possible to formulate the design problem as a mathematical programming (optimization) problem when all the alternatives to be considered can be enumerated and evaluated quantitatively (Westerberg, 2004). In addition, this approach presents a number of difficulties when deal­ing with the optimization of under-defined design problems and uncertainties that result from the multi-objective requirements of the design problem (Li and Kraslawski. 2004).