A Model Simplification Technique For Computer Flowsheeting

The objective of this research has been to develop a rapid accurate method for computing steady-state heat and material balances for process flowsheets. Although there are many flow sheeting systems used routinely in the process industries, the computational times for optimisation or sensitivity studies can be excessive, and there is an incentive to develop substantially faster systems. A dual-level flow sheeting system is developed which links a conventional ‘rigorous’ flow sheeting package with a set of simplified unit-operation computational procedures. The link between the two levels is made by a set of parameters generated by the rigorous routines which ensure that, for a given set of conditions, both routines compute the identical flowsheets. In operation the system performs one computation of the rigorous models from which the required matching parameters are generated. The iteration then proceeds using simplified models. After a number of cycles of iteration, the rigorous models are re-computed to produce a revised set of matching parameters. The computation continues switching between rigorous and simplified models until convergence is achieved. The simplified models developed omit numerical integrations, stage-by-stage calculations and reference to rigorous physical property computations. They include adiabatic and isothermal flash routines and distillation models. These models are of the order of 100 times faster than rigorous models from an efficient well established commercial flow sheeting system. The optimal frequency for re-computing the linking parameters is determined by a treatment based on functional analysis, in which the progress to convergence is monitored during computation. The overall improvement in speed is determined primarily by the time taken to compute rigorous models and matching parameters, not by the method used to solve the simplified models. It is shown that the dual-level system exactly reproduces the results of the rigorous system, with a speed improvement of better than a factor of 20. For problems requiring more time-consuming rigorous computations (e.g. more complex physical properties), there is potential for much greater speed improvement. The examples given have been restricted to first order convergence of simulation problems. Beneficial application of the technique to superliner convergence and to flexibility and optimization studies is, nevertheless, considered.