(246l) Mosaic: Parallel Computing, Multi-Objective Optimization Applications

The use of high performance computing provides the opportunity to perform complex optimization analysis of large chemical engineering systems. The goal of this project is to use parallel computing techniques within the MOSAIC modeling framework to perform multi-objective optimization of chemical process systems with the purpose of simultaneously improving environmental performance, profit, and safety. Multi-objective optimization is a process in which two to three objective functions are optimized to Pareto optimality, meaning no objective can be improved without weakening the other objective(s). MOSAIC (Kraus et al., 2014) is a web-based modeling, simulation, and optimization environment, which allows for the formulation of individualized equation systems and custom algorithms to describe and solve chemical processes and unit operations within a chemical plant. Using chemical engineering plant design specifications and MOSAIC’s ability to employ custom algorithms, multi-objective optimization problems are constructed and solved in parallel. Multi-objective optimization problems require a significant amount of computational power, but finding the Pareto set can be efficiently completed using the parallelism available in a graphics processing unit, which is more effective than a traditional computer for implementing large, data-heavy algorithms. In this work, a parallel algorithm was developed in MOSAIC to find the Pareto optimality set for a series of plant design examples. The computational results determined that MOSAIC can effectively implement a multi-objective optimization algorithm. This utilization and application of MOSAIC will allow for the efficient parallel solution of chemical engineering systems with multiple objectives which were previously too complex to solve.

Kraus, R., Fillinger, S., Tolksdorf, G., Minh, Duc H., Merchan-Restrepo, Victor A. and Wozny, G. (2014), Improving Model and Data Integration Using MOSAIC as Central Data Management Platform, Chemie Ingenieur Technik, doi: 10.1002/cite.201400007