TY - JOUR
T1 - Scale-up study of electrochemical carbon dioxide reduction process through data-driven modelling
AU - Zhang, Guyu
AU - Liu, Xiaoteng
AU - Lei, Hanhui
AU - Wang, Yucheng
AU - Bildan, Denise
AU - Zhuge, Xiangqun
AU - Xing, Lei
AU - Luo, Kun
PY - 2024/10/1
Y1 - 2024/10/1
N2 - Efficient electrochemical carbon dioxide reduction (eCO2RR) depends on addressing mass transfer kinetics hindering CO2 diffusion to the cathode surface. Gas diffusion electrodes (GDE) have enhanced this process, but the shift from lab-scale research to industrial use is to be explored, and we systematically assessed four variable factors: electrode area, gas flow rate, catalytic layer (CL) thickness and gas diffusion layer (GDL) porosity for scaling-up the electrolyser with a comprehensive two-dimensional physical model was developed to investigate the concentration, distribution, and consumption of CO2. Random Forest (RF) coupled with Latin Hypercube Sampling (LHS) data collection method demonstrate a prediction accuracy of 98.67 % and a RMSE of 0.00058 for the average CO2 concentration. A maximum CO2 consumption rate of 98 % was achieved at a CL thickness of 73 μm and a GDL with a porosity of 0.8, for an electrode area of 100 cm2 and a gas flow rate of 91 mL/min. This high level of CO2 consumption was sustained throughout the scaling-up process, consistently at 96.7 %, as the evidence attests to the reliability and feasibility of the scale-up approach.
AB - Efficient electrochemical carbon dioxide reduction (eCO2RR) depends on addressing mass transfer kinetics hindering CO2 diffusion to the cathode surface. Gas diffusion electrodes (GDE) have enhanced this process, but the shift from lab-scale research to industrial use is to be explored, and we systematically assessed four variable factors: electrode area, gas flow rate, catalytic layer (CL) thickness and gas diffusion layer (GDL) porosity for scaling-up the electrolyser with a comprehensive two-dimensional physical model was developed to investigate the concentration, distribution, and consumption of CO2. Random Forest (RF) coupled with Latin Hypercube Sampling (LHS) data collection method demonstrate a prediction accuracy of 98.67 % and a RMSE of 0.00058 for the average CO2 concentration. A maximum CO2 consumption rate of 98 % was achieved at a CL thickness of 73 μm and a GDL with a porosity of 0.8, for an electrode area of 100 cm2 and a gas flow rate of 91 mL/min. This high level of CO2 consumption was sustained throughout the scaling-up process, consistently at 96.7 %, as the evidence attests to the reliability and feasibility of the scale-up approach.
KW - CO reduction
KW - Electrochemical
KW - Machine learning
KW - Mass transfer
KW - Scale-up
UR - http://www.scopus.com/inward/record.url?scp=85197293287&partnerID=8YFLogxK
U2 - 10.1016/j.fuel.2024.132400
DO - 10.1016/j.fuel.2024.132400
M3 - Article
SN - 0016-2361
VL - 373
SP - 1
EP - 9
JO - Fuel
JF - Fuel
M1 - 132400
ER -