Comparative techno-economic analysis and optimization of CO liquefaction processes using Bayesian optimization
- Journal
- Computers & Chemical Engineering
- Page
- 109770
- Year
- 2026
This study presents a techno-economic and exergy-based analysis of CO liquefaction processes by combining rigorous Aspen Plus simulation with Bayesian optimization. Three established liquefaction process configurations — open-loop, closed-loop, and open/closed-loop configurations — were evaluated under common boundary conditions for product pressures of 7.5 and 20 bar and production capacities of 1000, 2000, and 3000 TPD. This work provides an enumeration-based topology-screening framework in which feasible compression-stage combinations and refrigerant choices are explicitly defined, and the continuous operating variables of each topology candidate are optimized using BoTorch-based Bayesian optimization.
The optimization results show that the levelized cost of liquefaction (LCO) is governed by the combined effects of product pressure, refrigerant selection, compressor power, and CAPEX–OPEX trade-offs. Across the evaluated base-case conditions, the closed-loop configuration using CH refrigerant achieved the lowest LCO. At 3000 TPD, the optimized closed-loop CH system achieved integrated LCOs of $12.45/tCO for the 7.5 bar product and $11.33/tCO for the 20 bar product when the common feed-pretreatment section was included. An electricity-price sensitivity analysis from 70% to 130% of the base electricity price confirmed that this ranking was robust for 41 of 42 product-pressure, capacity, and electricity-price combinations considered.
The results demonstrate that thermodynamic efficiency alone does not determine the cost optimum. Additional compression stages can reduce compressor work but may increase the levelized CAPEX sufficiently to offset the OPEX saving. The proposed workflow therefore provides a reproducible decision-making basis for comparing CO liquefaction topologies under consistent simulation, economic, and exergy-analysis assumptions.
Sung Hyun Ju and Jisung Byun contributed equally to this work.