(185x) Application of Sequential Design of Experiments (SDoE) to a MEA-Based CO2 Capture Pilot Plant

The U.S. DoE’s Carbon Capture Simulation for Industry Impact (CCSI²) is a partnership among national laboratories, industry, and academic institutions, focused on the application and enhancement of a suite of computational tools and models developed in its predecessor project, the Carbon Capture Simulation Initiative (CCSI). The CCSI Toolset was developed to accelerate the development and deployment of novel carbon capture technologies in part by reducing the risk associated with process scale-up through the use of fundamental predictive models with full uncertainty quantification (UQ).

As a part of CCSI, a modeling framework for CO₂ capture solvent systems was developed with the capability to quantify uncertainty in submodel (e.g. physical properties, mass transfer, kinetics) parameters. In this work, this model is used to identify optimal test conditions through a sequential design of experiments (SDoE) implemented in a pilot plant. Due to the expense of pilot-scale testing, it is important to allocate resources to maximize learning during the test period. In a typical experimental design, test cases are chosen using a space-filling approach that does not consider the process output space or leverage the collected process data to update the test plan. The SDoE approach, however, incorporates the process model and UQ to optimally select experiments, to achieve a specific objective or goal for the test campaign. The data collected are incorporated into a Bayesian framework in which model uncertainty can be quantified and with that knowledge used to improve the model by reducing its parametric uncertainty. A new set of experiments are then selected using the updated model, thus utilizing the knowledge gained in the previous test runs.

The objective function of the proposed SDoE in this work seeks to minimize the cost of CO₂ capture and considers both capital and operating costs as well as thermodynamic improvements. Moreover, the CO₂ capture percentage is treated as an optimization variable in this objective function, whereas this is often considered fixed at a typical level (85-90%) in many campaigns. To reduce the computational expense of the SDoE procedure, multiple-input/multiple-output surrogate models are developed for the absorber and the stripper columns. These reduced models accurately predict the output variables of interest over the operating and parametric spaces of interest, effectively reproducing the rigorous column model results within 5% error for most cases. The SDoE will be implemented at Norway’s Technology Centre Mongstad’s 12 MWe scale pilot facility in an upcoming test campaign.