Development and Application of Simulation-based Methods for Engineering Optimization Under Uncertainty

This study presents a methodology that combines Monte Carlo methods for uncertainty propagation along with modern stochastic programming methods for optimization of chemical engineering process models. The aim is to integrate uncertainty information of model parameters directly into optimization workflow. Successful implementation will give statistically strong optimum designs that do not require the use of safety factors that could compromise the cost competitiveness of the process design. Assuming that uncertainty behaves as random disturbances, Monte Carlo sampling techniques of model parameters uncertainty are used to quantify disturbances effects in model output. The resulting information is then used in a surrogate assisted optimization approach designed specifically for stochastic simulations. The surrogate methods used are the developed Stochastic Kriging methods along with a novel infill criterion. This work will also include an Artificial Neural Network alternative approach. The methodology is applied to two distillation flowsheets built in Aspen Plus. Both proposed methods showcase superior results to a traditional SQP in a case study. Stochastic Kriging approach showcases superior results when optimizing a two-column distillation loop of a DIPE + IPA + 2MEt system, showcasing three times lower cost evaluation compared to other methods while respecting propagation of uncertainty. This work demonstrates the applicability of simulation-based optimization for engineering designs under uncertainty.

Previous chapter in volume