Synthesis of heat exchanger networks using mathematical programming and heuristics in a two-step optimisation procedure with detailed exchanger design

This study makes use of a novel methodology for the synthesis of heat exchanger networks, which is aimed at overcoming the shortcomings associated with the use of shortcut models to represent individual exchangers in the synthesis network. The new approach entails the use of a number of correction factors to get networks which are based on the use of shortcut models, such as the mixed integer non-linear programming (MINLP) stage-wise superstructure (SWS) of Yee and Grossmann (1990) to more closely represent physically achievable heat exchangers and ensure that the MINLP network topology optimisation step of these models converge on a real design, rather than an approximated one. In this paper, the SWS formulation is used for the generation of an initial network after which its objective function is modified to include the correction factors that force its objective function towards the cost of a network whose individual exchangers are designed using methods such as Bell-Delaware and heuristics. The modified objective function includes parameters that modify the areas obtained by the shortcut based MINLP model so as to more closely represent the areas obtained by the detailed models and also includes a novel method for including the number of shells required for each exchanger duty. The correction factors account for pressure drops, Ft correction factors, number of shells, TEMA considerations, and changes to the overall heat transfer co-efficient of each stream match. The methodology is applied to two examples and the solutions are comparable with other solutions obtained in literature and were shown to produce good solutions. The reason that the method is effective is because many potential networks are evaluated during the iterative procedure and the best network, based on the detailed exchanger designs, is chosen. In this way it is possible to use the detailed exchanger designs to “guide” the MINLP optimisation towards more realistic networks and also to generate many different potential networks.