Stony Brook researchers are using an artificial intelligence (AI) approach to reduce carbon emissions by converting carbon dioxide into methane, which will be made into a non-fossil fuel energy source.
Led by Anatoly Frenkel, a professor of Materials Science and Chemical Engineering, the team aimed to determine the properties of a catalyst — a substance that speeds the rate of a chemical reaction without being consumed — that can most efficiently convert carbon dioxide to methane. In order to measure catalytic properties like the size, structure and chemistry, the team used a machine-learning approach to extract the data.
Frenkel, who is a Brookhaven National Lab (BNL) joint appointee, worked with his team in the lab’s National Synchrotron Light Source II facility, which provides equipment to further research in areas like energy security and environmental sustainability. BNL frequently fosters projects that study the natural exchange of greenhouse gases, the carbon cycle and how shifting atmospheric and environmental factors affect plant growth.
Carbon emissions will decrease while storing and transporting the methane to make a non-fossil fuel energy source, Frenkel said, since the “methane will not pollute the atmosphere because it will be converted into feedstock for fuel.”
By creating resources that do not come from fossil fuels, Frenkel spoke about how methane can be used for chemical manufacturing. This form of renewable methane can be manufactured into everyday products, like several plastics used as building materials. Methane is used as feedstock for synthesizing methanol and is processed into formaldehyde — a chemical compound used in these plastics.
The team found that copper is a promising candidate as a catalyst that can survive during catalysis — when the rate or outcome of the reaction is influenced by the catalyst — because it can lower the temperature of the carbon dioxide-to-methane reaction while deriving the desired product: methane and water vapor.
By performing X-ray absorption spectroscopy — a technique that emits x-ray beams to determine the electronic structure of materials — with BNL’s synchrotron, the team analyzed how copper atoms absorbed the synchrotron X-rays. The X-ray absorption spectrum then holds the information about the catalyst’s chemical composition and structure, which is fundamental during the catalysis of carbon dioxide to methane.
A challenge that arises, according to Frenkel, has to do with the fact that copper clusters are small nanoparticles that have a few atoms to collect data from; it is difficult to determine the structure and configuration of atoms that makes the best catalyst to facilitate the conversion of carbon dioxide to methane.
In order to tackle this challenge, the team used AI to extract the catalytic properties found in the X-ray signals of the catalysts. AI utilizes machine learning to create an artificial neural network that mimics how the human brain transmits and processes information.
“When we have experimental X-ray spectrum, which is mirrored during the reaction condition, then we can utilize our machine-learning model to predict the structure from this experimental spectrum,” third year chemistry Ph.D. candidate Yang Liu said.
After determining the structure of the catalyst, Liu said that “at the same time when we do the experiment, we also mirror the reactant and the product, so we can correlate that reaction performance with the structure property of the catalyst.”
The neurons are taught the known spectrums with their corresponding structures of catalysts that were generated during the experiment.
Fourth year materials science Ph.D. candidate, Nicholas Marcella — who worked on the project — said that the team’s next steps in their research will be to “put statistical significance on [the analytic methods used to understand X-ray spectroscopy] so we can say that with certainty, we know that the structural characteristics that we’re looking at are real and accurate.”