Artificial Intelligence Assisted Gas Diffusion Electrode Development for Carbon Dioxide Reduction

Intergovernmental panel on climate change (IPCC) presented the 6^th IPCC report that the temperature of Earth is likely to exceed the 1.5 °C rise. The IPCC forecasts 2025 will be the tipping point at which greenhouse gas emissions must turn to a decreasing trend, otherwise preventing global warming after 2030 below the 2 °C limit will be difficult. We are coming to the point where we must not only switch to renewable energy, but also remove already emitted greenhouse gases.[1]<br/>Electrochemical carbon dioxide (CO₂) reduction has been investigated since 1980s-1990s as a promising greenhouse gas elimination method. The activity of CO₂ reduction in aqueous electrolytes is far lower than economically viable minimum. The low solubility (&lt; 30 mM) and diffusivity of CO₂ in aqueous electrolyte limits reduction current approximately to 30 mA in ambient condition. The fact that the majority of feedstocks in the real world of greenhouse gases are in the gaseous state has also fueled technological innovation. [2]<br/>The gas diffusion electrode (GDE) emerged to address this problem and enable current densities a few orders of magnitude higher than planar electrode. The development of GDEs has been investigated by applying materials with various properties, such as porosity, hydrophobicity, mechanical strength, and conductivity. However, the complex GDE structures and various roles of each layer in GDEs cause difficulties in theory-based development of the optimal GDE.<br/>In order to overcome the aforementioned complexities of developing the optimal GDE, this study focuses on utilizing a structure data-based machine learning to optimize the structures in the GDEs. Imaging of GDEs can be performed using characterization techniques such as scanning electron microscopy and commutated tomography to obtain the structural information.[3] Relationship between CO₂ reduction performance and structural characteristics of GDEs are used as training data for 3D convolutional neural network. AI unsupervised learning, such as Affinity Propagation and MiniBatch k-Means cluttering, are utilized to reveal unknown GDE design parameters.<br/>We provided method for dealing with complexities involved in the GDE development through 3D structure information measurements, segmentation, learning and clustering. The method presented in this study can be applied not only to the development of GDEs for electrochemical CO2 reduction but also to many other applications in the energy fields.<br/><br/>[1] IPCC, 2022: Summary for Policy Makers. In: Climate Change: Mitigation of Climate Change of Working Group III to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change (2022, April)<br/>[2] Nitopi, Stephanie, et al. "Progress and perspectives of electrochemical CO2 reduction on copper in aqueous electrolyte." Chemical reviews 119.12 (2019): 7610-7672.<br/>[3] Primakov, Sergey P., et al. "Automated detection and segmentation of non-small cell lung cancer computed tomography images." Nature communications 13.1 (2022): 1-12.