Process Systems Synthesis & AI-Driven Design

Research Contents

• Superstructure enumeration: encode alternatives for units, connections, and recycle structures.

• MINLP/MILP & decomposition: exploit convexification, cuts, and hierarchical strategies for tractable solves.

• Bayesian inverse design: learn surrogate objectives/constraints and search flowsheets efficiently.

• Planning & scheduling: couple strategic designs to production/maintenance under uncertainty.

Conventional workflows optimize steady-state designs first and address operations later, which can produce narrow margins and fragile control when markets or feed conditions shift. We treat design and operation together by constructing superstructures that enumerate feasible units, pathways, and interconnections and by solving mixed-integer nonlinear programs to select reconfigurable networks that remain operable over wide envelopes. Inverse-design and Bayesian-optimization frameworks guide flowsheet discovery and parameter tuning with minimal expensive simulations, while decomposition and convexification techniques maintain tractability. Planning and scheduling layers ensure that strategic designs are consistent with day-to-day constraints such as maintenance, logistics, and market participation. This end-to-end methodology yields flowsheets that are optimal on paper, resilient in practice, and defensible economically and environmentally under uncertainty. 

Associated members: Naeun Choi, Sunwook Kim, Hyunji Kwon