Efficient handling of the implicit constraints problem for the ASPEN MINLP synthesizer

Abstract

A new process synthesizer has been recently implemented in the ASPEN chemical process simulator. The process synthesizer determines optimal flowsheet configurations and is based on a mathematical programming (MINLP) optimization algorithm. This MINLP algorithm consists of solving alternating sequences of NLP subproblems and MILP master problems. The NLP subproblem provides the linearization information of the nonlinear constraints which relate the output variables to input variables specified explicitly. For a simulator like ASPEN, most of the relations are implicit, which prevents the NLP optimizer from transferring this crucial information to the master problem. In our earlier work, a strategy was outlined to circumvent this problem by adding additional variables and constraints to the NLP problem. This procedure invariably increases the load on the NLP optimizer, which is normally the efficiency-determining factor in the large-scale MINLP process synthesis. In this paper, we present a new and efficient strategy for handling these additional variables and constraints by means of partitioning the variables. This approach is shown to decrease the computational time significantly.