Researchers have developed an artificial intelligence model that dramatically accelerates the discovery of catalysts essential for producing green hydrogen, identifying promising new materials in a matter of days rather than the years typically required for such work. A team led by engineers at Worcester Polytechnic Institute (WPI) used a specialized machine learning approach to find high-performance, low-cost catalysts, a breakthrough that could remove a major economic barrier to the large-scale production of clean energy.
The new method moves away from the expensive, rare metals like platinum and ruthenium that have long been the standard for industrial catalysts. Instead, the AI focused on earth-abundant metal alloys, ultimately identifying bimetallic pairings such as nickel-iron that offer comparable performance at a fraction of the cost. This advance, published in Nature Chemical Engineering, not only addresses the high price of catalysts but also mitigates supply chain vulnerabilities tied to precious metals, paving the way for more resilient and economically viable hydrogen production.
A New Method for Catalyst Design
The traditional process for discovering new catalysts has long been a bottleneck in materials science, relying on a painstaking, trial-and-error approach. Scientists would typically synthesize and test a limited number of material combinations, a resource-intensive method with a low rate of success. The sheer number of potential elemental combinations makes exhaustive experimental screening practically impossible. This Edisonian approach has slowed the adoption of new materials for critical applications, including the electrolysis and ammonia decomposition needed for green hydrogen.
Targeting Abundant Materials
The WPI-led team, in collaboration with the Dalian University of Technology and Northeastern University, developed transparent machine learning models to overcome this challenge. Their system was designed to analyze thousands of potential bimetallic alloys, using fundamental physics principles to predict their catalytic performance. The AI models incorporated key parameters like electronic structure descriptors and adsorption energies to evaluate each candidate’s likely activity, stability, and cost. Within two days, the AI effectively screened 95% of the initial candidates, narrowing a vast field down to a dozen promising alloys built from common metals like cobalt, iron, and nickel. This rapid, intelligent down-selection allowed the researchers to focus their experimental efforts exclusively on the most viable options.
Plasma-Assisted Decomposition
The newly discovered catalysts were designed for a specific and innovative process: plasma-assisted ammonia decomposition. Unlike traditional steam methane reforming, which is energy-intensive and produces significant carbon dioxide, this method offers a cleaner pathway to hydrogen. The process involves feeding ammonia (NH₃) into a reactor and using a low-temperature plasma field to help the catalyst break the ammonia into hydrogen (H₂) and nitrogen (N₂). The team found an optimal temperature range of 350–450 °C for the catalyst bed, which balances high performance with material longevity. The use of plasma reduces the overall energy required by up to 20% compared to conventional thermal methods, and when powered by renewable electricity and used with green ammonia, the process can cut CO₂ emissions by over 70%.
Significant Cost and Supply Chain Advantages
The economic implications of this research are substantial. An analysis conducted by researchers at Northeastern University found that switching from ruthenium-based catalysts to the newly identified nickel-iron alloys could slash material costs from approximately $150 per gram to less than $5 per gram. This dramatic cost reduction fundamentally changes the financial model for setting up new hydrogen production facilities. High catalyst cost has been a persistent barrier, accounting for a large portion of the capital expense for electrolyzers and reactors. By replacing scarce precious metals with widely available alternatives, the technology also insulates the green hydrogen supply chain from the price volatility and geopolitical risks associated with sourcing platinum-group metals.
Broader Research Trends in a Competitive Field
The success at WPI is part of a larger global trend where AI is being applied to solve critical materials science problems in the energy sector. Multiple research groups are pursuing similar goals, underscoring the importance of this work. For example, researchers at the University of Toronto recently used an AI to analyze over 36,000 different metal oxide combinations for water-splitting applications. Their model identified a novel alloy of ruthenium, chromium, and titanium that proved significantly more stable and durable than existing ruthenium catalysts, which often degrade quickly. Similarly, a team at the University of Houston is using machine learning in conjunction with microwave plasma technology to discover new catalysts for hydrogen generation and carbon capture, a project funded by the National Science Foundation. These parallel efforts highlight a paradigm shift in materials discovery, moving from manual experimentation to AI-driven design.
Verification and Future Prospects
The catalysts identified by the WPI team’s AI were not merely theoretical. Researchers at the Dalian University of Technology synthesized the recommended alloys and tested them in laboratory-scale plasma reactors. These experiments confirmed the AI’s predictions, demonstrating that the new materials could reliably produce hydrogen under continuous operation while showing the required stability. The explainability of the AI models was a key feature, allowing the scientists to understand the underlying physical properties that made the chosen alloys effective. The next phase of the research involves scaling up the technology for real-world validation. The team is exploring demonstration projects with partners in the maritime and mining industries to prove the system’s viability in demanding, off-grid environments. If these pilots are successful, this AI-guided approach to catalyst design could play a pivotal role in displacing steam methane reforming as the dominant method of hydrogen production, accelerating the transition to a global hydrogen economy.
