Design and analysis of solvents in reaction media

Date of Completion

January 2001


Engineering, Chemical




In this dissertation, a systematic framework for the design and analysis of solvents in reaction systems is developed using computer aided molecular design (CAMD) technique. The framework consists of problem formulation, solution strategy development, and uncertainty analysis. Solvents as mass separation agents (MSAs) to increase the productivity of the reaction play an important role in a fermentation or reversible reaction. A number of solvent characteristics including biocompatibility, inertness, and phase-splitting capability are identified as important properties of the candidate solvent. Group contribution approaches are employed to characterize the relation between physical properties and the structure of the compound. The resulting mathematical programming model is a mixed-integer nonlinear programming (MINLP) model, which involves continuous and binary structural variables. An outer approximation (OA) based algorithm is tailored to solve the MINLP model. Using the above methodology, n-Octane is identified as the optimal solvent for ethanol fermentation with maximum extraction efficiency, while isopropyl propionate and methyl isovalerate are the optimal solvents with the minimum mass flow rate. ^ To overcome the limitation of the OA based algorithm that may end up with a locally optimal solution for solving nonconvex MINLP, a hybrid global optimization approach, namely OA_Global, is developed. It incorporates a soft computing (simulated annealing) approach with outer approximation with equality relaxation and augmented penalty (OA/ER/AP). The simulated annealing method can find a global solution with an asymptotic convergence guarantee in probability. Although the proposed approach is not provably globally optimal, computational experience with benchmark examples and solvent design MINLP models indicate that this approach gives near global optimal solutions. ^ Uncertainty resulting from the physical property modeling is characterized through the flexibility analysis technique on chemical processes. The value of a feasibility function is calculated to measure how the uncertainty issues affect the designed molecules. Due to the multiextremality and nondifferentiability of the feasibility function, novel optimization approaches (BB and BB_ACTIVE) that avoid explicit enumerations are developed to evaluate this feasibility function. Case studies on chemical processes and molecular design indicate that these two approaches are computationally efficient. ^