Accurate process models are essential for optimising CO2 capture processes, but existing models can be slow to calibrate and difficult to adapt to different capture technologies. OptiCap AI aims to develop the first open-architecture, physics-based digital twin for CO2 capture. The digital twin uses artificial intelligence (AI) to calibrate the model with process data, enabling more efficient CO2 capture. Combining rigorous first-principles models with physically constrained AI is expected to reduce energy use and the cost of CO2 capture by 10-15%, while de-risking scale-up across emission sources and process concepts. By embedding dynamic simulation, uncertainty propagation and sensitivity analysis, the project will identify the parameters that have the greatest influence on cost and risk, supporting engineering and investment decisions.
OptiCap AI combines two complementary developments. First, it develops physically constrained AI calibration using grey-box parameters that honour mass and energy balances when fitting the digital twin to plant data, reducing months of manual tuning to hours. Second, it develops an open, high-fidelity digital twin in Julia/SciML, integrating Q-props electrolyte thermodynamics to accurately and efficiently model different solvents and flue gas compositions. Together, these developments address key bottlenecks in CO2 capture modelling, including slow model calibration, limited solvent coverage and opaque black-box digital twins. The approach will be demonstrated across three real-world capture processes (power generation, biogas production and photo-electrochemical direct air capture (DAC)), as well as through a world-first process-level solvent screening demonstration and system analyses evaluating uncertainty and sensitivity for next-generation techno-economic and life-cycle assessments.
Expected impacts include 10-15% energy savings for post-combustion capture and biogas production. This could unlock approximately 1 Mt CO2/year of additional economically viable capture from Denmark’s estimated 10.5 Mt/year point-source potential. The project will also support de-risked scale-up, faster troubleshooting and modelling of novel solvents and process concepts, including photo-electrochemical DAC. An MIT-licensed Julia core will be released for public beta in 2029, allowing Danish emitters and engineering, procurement and construction (EPC) companies to test OptiCap AI.
