A superstructure modeling framework for process synthesis using surrogate models

One of the fundamental problems in chemical engineering regards the design of processing facilities tailored for the production of specific chemical products from a specific set of raw materials. The conceptual development of these facilities requires the integration of elemental processing units into an optimal processing system capable of fulfilling production requirements while acknowledging a number of constraints including raw material availability, operational safety, environmental regulations etc.;In order to address this conceptual "process synthesis problem", several strategies have been developed over the years. One of such approaches is called "Superstructure Optimization". Here, a complex process diagram including all potentially useful processing units and interconnections is initially proposed, and an optimization model featuring a suitable objective function as well as mathematical models for all units and connections is developed. The solution of this model identifies the best process structure and operational conditions. Since all aspects of the design (both structural and operational) are considered simultaneously, good solutions are generally identified.;This thesis is focused on the development of a novel methodology for the formulation of superstructure optimization models to support the solution of process synthesis problems. It includes a new way to represent and generate process superstructures, as well as all the elements required to formulate complete superstructure optimization models using an entirely modular approach and standard processing unit models. The models for all superstructure elements (i.e. processing units and connectivity elements) are created from detailed simulation models, where complex non-linear equation blocks are replaced by general non-linear surrogate mappings capable of fitting realistic process data. This approach leads to realistic superstructure models which can be solved to optimality given their reduced mathematical complexity. A novel approach to the design of these hybrid surrogate models and the implementation details required to obtain the surrogate mappings are also discussed.;Several illustrative examples and case studies are included to demonstrate the capabilities of our methodology; in particular, how the combination of realism and reduced mathematical complexity allows the formulation of tractable optimization modes which are generally larger than tractable models based exclusively on detailed unit models.