A Modular Approach for Surrogate Modeling of Flowsheets
Flowsheet simulation, an important building block in chemical process development and design, generally requires the solution of a nonlinear system of equations. Such a simulation can fail if either no solution exists for the chosen specifications or if the initial values for the solver are chosen unfavorably. In this work, to overcome such convergence failures, AI-based surrogate models are trained for unit operations of a flowsheet and then interconnected. For a pressure swing distillation with recycle, it is shown that such an interconnection of surrogate models yields more accurate results compared to a surrogate model for the whole flowsheet at once, and that adequate starting points for the flowsheet simulation can be obtained from the interconnected surrogates.