Sun_cmu_0041E_10647.pdf (2.35 MB)
Download file

Physical properties and chemical product design

Download (2.35 MB)
posted on 09.02.2022, 21:59 by Yijia SunYijia Sun
Chemical product design is the problem of identifying chemical compounds that satisfy a set of previously identified functional properties for a specific application. Throughout the industrial era, chemical product design has played an increasingly important societal role by addressing the growing demand for better products. Naturally, product design has become a focal area for chemical scientists and engineers. This thesis develops new methodologies for chemical product design problems where the design targets are physical properties. We address at problems with dynamic design targets and stationary design targets separately. For design applications with stationary property targets, we rely on algebraic optimization models to efficiently locate optimal compounds from the vast chemical design space. In particular, we exploit efficient mixed-integer linear programming (MILP) models and solve for optimal pure chemical components for a specific application: identifying better cooling fluids for electronic equipment. We use group contribution (GC) methods as the major property estimation tool along with additional accurate property models. We derive a metric to measure the cooling performance of a two-phase cooling system consisting of micro-channel heat sinks. Additionally, we carry out a
sensitivity analysis on property predictions to assess the effect of prediction uncertainty on the final design outcome.
To extend the coolant design space to include silicon containing structures, we develop a new GC method capable of property prediction for organosilicon compounds. Along the way, we propose a functional group selection method that deterministically decomposes each molecular structure into the smallest number of non-overlapping functional groups while ensuring each group holds a
maximum amount of information. The group selection method is applied to construct GC models to predict eight pure component properties. Utilizing the new GC models, we are able to identify organosilicon cooling fluids with considerably improved heat transfer properties.
As for design application with time-varying targets, or functionalities that are difficult to model via algebraic expressions, we leverage the state-of-art derivative-free optimization (DFO) methods to solve for a diverse span of candidate chemicals. We assess DFO algorithms and demonstrate the viability of DFO in a polymer configuration design example. Our computational results suggest that
a collection of derivative-free algorithms can successfully search the chemical design space and identify good solutions with high computational efficiency. Whether the candidate compounds can be put into production depends on their likelihood to be synthesized. We investigate contemporary synthetic planning methodologies and provide an overview on retrosynthesis frameworks. We address potential challenges and opportunities facing automated synthetic planning. Lastly, we summarize the major contributions from this work and offer future research directions.




Degree Type



Chemical Engineering

Degree Name

  • Doctor of Philosophy (PhD)


Nick Sahinidis