Polymers have been gradually replacing metals in many applications due to their versatility. Nowadays, many consumer products such as computers and automobile components rely on the technology and production from polymer companies. To ensure the sustainability of these companies, it is important to design reliable processes. To analyze and improve the processing of plastic parts, advanced computer simulation tools have been developed. Yet, the difficulty of optimizing polymer processes is that the performance measures (objectives) involved usually show conflicting behaviors. Therefore the best processing conditions for one performance measure are usually not the best for some other performance measures. This thesis proposes an Optimization via Simulation method that considers multiple performance measures simultaneously. The method is able to approximate a set of Pareto solutions without having to evaluate a large number of simulations. In order to accomplish this, design of experiments, metamodeling, Data Envelopment Analysis, and Pareto optimality are combined.
After first testing the method with well-known multiobjective optimization test problems, it is illustrated with several injection molding case studies. The method is also compared with an alternative metamodel-based multiple criteria simulation optimization method. The performance of the method is evaluated, in light of the quality of the obtained Pareto frontier as well as the number of simulation runs required to obtain such a frontier. The quality of the approximated Pareto frontier is assessed by the percentage of solutions dominated by the frontier using the hypervolume indicator.
In addition to the test problems and the simulation-based case studies, we applied the optimization method to two case studies where only experimental data was used. In these cases we molded the parts under study and evaluated the values of the performance measures from the actual molded parts.
At the end, we present some preliminary ideas on how to build process windows. A process window is the range of the controllable variables at which one needs to operate the process to obtain the best compromises between the performance measures.