Machine learning provides green ammonia efficiency

Beyond its role as an important component of fertilizers that help feed the 8 billion people around the world, ammonia can also function as a hydrogen transporter. When produced cleanly, ammonia is easier to transport and store than the hydrogen it contains, making it a potentially versatile enabler of the future green hydrogen economy. The problem is that almost all of the world's 200 million tonnes produced each year are produced in large energy-intensive factories that consume around 2% of the total world energy and emit a similar share of global CO2.

But now, researchers at the University of New South Wales in Sydney have developed an enhanced method for creating green ammonia through electrolysis. Using machine learning, researchers have found innovative catalysts that speed up chemical reactions and achieve “seven-fold improvements to ammonia production.” “At the same time, it's nearly 100% efficient.”

The new catalyst is designed to improve the lab-built prototype method that uses renewable energy to create ammonia from air and water. The prototype system, developed in 2021 in collaboration with researchers at the University of Sydney, aimed to produce small and clean ammonia at a low cost for distributed use. However, commercializing the process requires improved speeds of ammonia production, which requires an innovative catalyst search.

To formulate a catalyst, researchers have chosen 13 prospective metals, from which combinations may improve the results. “But that would have meant testing over 8,000 different combinations,” Jalili said. “And after two or three ingredients, chemistry is too intertwined due to old-fashioned trial and error.”

Machine learning boosts ammonia production

Instead, they turned to machine learning. Researchers wrote the program in Python based on Gaussian processing learning that can detect patterns in small amounts of data. They provided unique lab data taking into account the program properties of metals that act as catalysts and ammonia production rates, cost, long-term stability, and falada efficiency.

When the AI model ran the data and made proposals, researchers conducted electrolysis tests, provided feedback, and generated new batches of proposals. In the four rounds of testing, it was sufficient to create 28 candidate experiments and find ways to significantly affect ammonia production performance by adding or removing elements.

“It took less than a week to run these 28 tests, but it straightened to an efficient five-metal alloy of iron, bismuth, nickel, tin and zinc, with all other combination benchmarks performing,” says Jalili. “We've significantly reduced the discovery time that could have taken months while improving the process of seven times green ammonia.

Ammonia often uses electrolytic factors to separate water molecules and combines hydrogen with atmospheric nitrogen. New catalysts, discovered in part through AI systems, improve the efficiency of electrolytic agents.University of New South Wales

The alloy was then formed on the electrodes for use in prototype systems constructed in previous labs. The Air-water-Ammonia module includes a nanosecond pulsed plasma reactor, an electrochemical cell (currently an electrochemical cell with new catalysts installed), and a process optimization tool. A plasma reactor called “tube lightning” by Jaliri uses small bursts of electricity to energize the air, breaking down nitrogen molecules and exerts sufficient reactivity to form ammonia. Electrochemical cells, on the other hand, speed up the chemical process of converting nitrogen compounds to ammonia.

“It's a hybrid process running on renewable power, which bridges the chemistry and electrochemistry of plasmas and we developed,” says Jalili. “Air and water are ingested at one end, and green ammonia comes out from the other side.”

He describes the lab module as an ammonia factory from self-contained air stuffed into a standard 6-meter shipping container. A pilot setup with power from a nearby solar array or wind array only air and water could produce 50-100 kilograms of green ammonia per day. “And if you want more capacity, you can add additional modules, like Lego blocks,” says Jalili.

However, outside of lab conditions, pilot modules tested on the farm are connected to the grid for stability rather than carrying renewable energy. According to Jalili, it supplies 0.5 kg of nitrogen-based fertilizer, which is derived from ammonia per day. It is enough to grow 500 cucumber plants in the season. Additionally, a much larger system is in the planning stage and supported by the New South Wales government and commercial partners. The farm aims to produce 90 tons of nitrogen-based fertilizer from ammonia each year, and will feature a solar energy plant of several megawatts. Jaliri has also spoken with the Bill Gates Foundation about the possibility of using the technology to produce fertilizer in the Sahara.

Meanwhile, researchers continue to develop current systems for commercialization. The goal is to reduce the lab setup to suitcase size. The size of the suitcase is that it can produce several kilograms of ammonia per day for less than $1 million (approximately $655,000 USD).

Wanting to look further, Jalili points out that ammonia is one of the world's most produced chemicals and that while the industry knows how to safely liquefy, store and ship it, docking points, pipelines, tanks and terminals are all in operation now.

“Instead of waiting for $1 billion in hydrogen plants, our containerized modules allow renewable energy-rich regions to create green ammonia in real time, allowing them to build support infrastructure,” says Jalili. “And fuel can then be used in three ways: as fertilizer, it goes back to hydrogen for the fuel cell vehicle, burns it with the turbine and engine to burn the backup power.”