A systems engineering framework for solvent discovery

Date of Completion

January 1999


Engineering, Chemical|Engineering, Industrial




The search for new solvents is often dictated by market demands as a result of new applications and processing requirements. On the other hand the need for solvent substitutes is often dictated by changing environmental and safety requirements. For example, many traditional solvents are on the environmental “hit lists” and are to phased out within the next few years. To respond to these concerns and more importantly to respond to rapidly changing operational requirements and market forces, there is a need for efficient and systematic strategies for “just in time” optimal design of solvent alternatives. ^ In response to the above needs, this doctoral project has resulted in a systems engineering framework for solvent design and substitution using computer-aided product design (CAPD) approaches. Although the framework is general enough for solvent design, the target application considered are cleaning solvents (blanket washes) for the printing industry. These solvents need to have maximum dissolution rate with swelling of the rubber blankets. ^ These models have rather complex mathematical behavior. Specifically, the CAPD models are highly nonlinear and nonconvex. As a result they have multiple and non unique solutions which are in general very difficult to obtain. In other words they exhibit multiextremality (multiple local optima). ^ To address the computational issue alluded above, efficient global optimization algorithms have been developed that exploit the mathematical structure of models. Specifically, two algorithms tailored for pure component design and for mixture design has been developed. For pure component design, a reduced dimension branch and bound (RD-BB) algorithm has been developed. For mixture design, a domain reduction algorithm (LIBRA) using interval analysis has been developed. LIBRA can efficiently solve constrained nonlinear problems. ^ Critical to the framework is the need to estimate physicochemical properties of the solvents. Very frequently, these thermodynamic models which tend to be semi-empirical, have parametric uncertainties associated with them. In this doctoral project, a quantitative approach to characterize uncertainty via use of joint-confidence region has been developed. These uncertainty representations are subsequently incorporated into the solvent design framework. Thus, the solvents designed are guaranteed to satisfy the specified requirements even in the presence of uncertainties. ^