The design and operation of several engineering systems entails the solution of large-scale
optimization problems with several decision variables (both continuous and discrete) and nonlinear
constraints. Despite advances in general-purpose optimization solvers, several engineering problems
may not converge to a solution. In this talk, we explore the role of domain knowledge in systems
optimization. Can we leverage knowledge of the system to design better formulations of optimization
problems as well as improved algorithms? To answer this question, we delve deep into superstructure
optimization- an optimization approach to process synthesis that is prone to non-convergence.
Superstructure optimization problems suffer from numerical singularities at zero-valued flowrates to
process units that are deselected. We build on prior knowledge of chemical process models to develop
an exact and well-defined reformulation of the superstructure optimization problem. We demonstrate
the use of our new formulation on distillation column synthesis. We briefly provide additional examples
from integrated molecular and process design, power systems as well as remote sensing to illustrate
the role of domain knowledge in optimization. In conclusion, we show that domain-knowledge can
influence every aspect of optimization: problem formulation, development of convex-relaxations for
global optimization and the design of efficient optimization algorithms. Algorithms developed here are
open-source and parallelized for high performance computing (HPC).