Evolution of parallel strategies for chemical process flowsheeting.

摘要

This thesis presents in two parts the results of a study in process flowsheeting. In the first part, parallelism was applied to solve simple flowsheets, using a facility called SCHEDULE. The second part consists of the attempts to explore the potential of parallelism in a modular flowsheeting simulator, viz., ASPEN PLUS, using distillation optimization test problems.; The effect of parallelism can be quantified by calculating speedup, which is the ratio of the time taken to solve a flowsheet in a sequential fashion to that taken using multiple processors in parallel. Speedups were calculated for problems considered in both parts of the study with positive results.; In the first part, two sets of interconnected equilibrium flash problems were solved with parallelism being applied at the module level, i.e., the individual flash module calculations for either a function evaluation or a Jacobian evaluation were performed on separate processors in parallel. The allocation of parallel tasks was done using SCHEDULE, a facility developed at the Argonne National Laboratories.; In the second part of the thesis, six distillation optimization problems were solved using a novel approach on the ASPEN PLUS simulator. A flowsheet consisting of one or several units was converted to one in which the units were split into more units. The units in the flowsheets were made independent by tearing all interconnecting streams. The independency of the modules paved way for parallel evaluation of the flowsheet. The tear streams were included as constraints in the NLP problem.; Since a large part of the solution can be performed in parallel, the unsplit problems were analyzed for parallel tasks and theoretical speedups were calculated. Two different speedup variables have been calculated. One was obtained by comparing the time taken for (theoretical) parallel split problem (on one processor) to that obtained for the unsplit problem. The second speedup was obtained by comparing the time taken for (theoretical) parallel split problem (on several processors) to that taken for the (theoretical) parallel split problem (on one processor). It was found that as the number of independents units increases in the flowsheet, parallelism can be used gainfully.