ITR/AP: Model-Based Integration of Methods for the Optimization of Process Systems

Carnegie Mellon University, Pittsburgh PA

Investigators

Abstract

This ITR research medium research project will put together an interdisciplinary research group whose aim will be to develop a new modeling and a solution framework that nontrivially integrates a wide class of solution methods, including mathematical programming, constraint satisfaction, and a variety of heuristic algorithms. The objective of this framework is to be able to effectively tackle challenging engineering optimization problems. The group will be composed of faculty and students from Computer Science, Operations Research and Chemical Engineering. The key goal for the group will be to identify principles according to which algorithms can be integrated. These principles will then form the basis for a modeling language that guides the integration. Furthermore, new solution techniques will be developed for hybrid models and for global optimization of nonconvex discrete/continuous problems. Two paths of integration will be investigated. One is to exploit problem structure at the modeling stage by using an idea developed in constraint logic programming: namely, by writing the problem in the form of "global" constraints that individually invoke algorithms tailored to the special structure of each constraint. A second is to make a fundamental distinction, at both the modeling and solution stage, between two types of algorithms: checkers and solvers. The research will be applied primarily to the optimization of process systems, with a focus on synthesis of reactive distillation systems and the integrated testing of new products and batch manufacturing. Because these problems have both discrete and continuous aspects, and the second involves difficult nonlinearities, they provide a natural context for combining optimization and logic-based methods. However, the research will develop a general-purpose approach to modeling and solution, rather than one that is exclusively for process systems applications.