SIMULATION-BASED DESIGN OF PROCESSES WITH MASS EXCHANGERS AND RECYCLES USING BAYESIAN OPTIMIZATION

MARTÍNEZ, ERNESTO C.; FISCHER, CARLOS D.

Santa Fe

Congreso; CAIQ2019 - X CONGRESO ARGENTINO DE INGENIERÍA QUÍMICA; 2019

AAIQ - Asociación Argentina de Ingenieros Químicos

For systematic search in the space of alternative process designs, the alternatives of choice are either resorting to a process simulator or mathematical programming. When resorting to a software package for process simulation, the interaction between components, first-principles balances and constitutive laws are quite simple since most commercial simulation environments have a large library of components, operations, and thermodynamic correlations. The main disadvantages of this alternative are the difficulty to include new operation models and tedious design optimization using mathematical programming models. Industrial practitioners often resort to their own expertise for tweaking decision variables upon simulation results. Taking a human out of the simulation-optimization loop would easy industrial applications of commercial packages. Tightly integrating a simulation package with Bayesian optimization is proposed. By aptly choosing the acquisition function, the problem of convergence for some input combinations is circumvented. Results obtained considering both real and categorical decision variables are very promising.